chore(gui): 版本号 1.9.6 → 1.9.7

## ① 图像模型曾与对话模型不可区分

根因是结构性的，不是漏了个过滤：

    Provider.Models()   → []string   ← 只有 ID，Kind 被丢掉
    SourceStatus.Models → []string   ← 扁平列表

于是 /api/status 给前端的 34 个模型（29 对话 + 5 图像）长得一模一样。选了
gpt-image-2 就 POST 到 /api/chat —— 那个端点服务不了图像模型，必然失败，
而且报错完全不提「这是图像模型」。

修法：新增 ModelKinds []ModelKindInfo（[{id,kind}]），纯增量、omitempty，
老 UI 不受影响；前端按 optgroup 分组，且这些 option 是 **disabled** 的。

分组还不够 —— 分组只是视觉分隔，operator 照样能选，选了照样失败。
disabled 的 option 会自解释（「请到生图页」），可选的只会失败。

★ 写这段时踩了一次 String.replace 的坑：`imgOpts.join("").replace(...)`
只替换**第一个**匹配，于是 6 个图像模型里只有第 1 个被禁用。浏览器实测
逐条读 e.disabled 才抓到（siliconflow 禁用了、vela 的 5 个没有）。
判据只看 value 里有没有 "image" 的话，第一次跑就会永远报警——那是判据
自己错了，比没判据更糟。

## ② 生图二级页面

侧栏新增 tab，与「对话」并列。网格展示 + 提示词输入 + 数量/尺寸，
调用 /v1/images/generations（不是 /api/chat）。

按钮显示已用秒数：实测该上游 15.7s~112.5s，波动 7.6 倍，不显示进度
看起来就像卡死了。

浏览器实测（几何测量，非仅 DOM 断言）：
    tab-image 容器   986×523  visible
    img-grid         840×374  visible
    提示词输入框      634×64   visible
    生图页 6 个模型，全部 kind=image，全部可选
    对话页 6 个图像模型，全部 disabled；97 个对话模型全部可选

## ③ base_url 引导不足

文案直接给出截断位置和具体后果，而不是泛泛地说「填 URL」：

    填主机地址，通常不带 /v1 —— 网关会自动拼接路径。但各家约定不同：
    若填 https://x.com/v1，对话路径会正确，而图像生成会变成 /v1/v1/… 而 404。
    不确定就用上方「自动探测模型」。

/v1/v1 不是假设：ImageURL 是 BaseURL + ImageEndpoint()，而图像默认路径
是**绝对**的 /v1/images/generations，chat 的默认是相对的 /chat/completions。
两者对「版本段归谁管」的假设相反。今天在 vela 上为此排查了两次。

## 判据

新增 6 个契约测试：
- 对话页必须读 model_kinds（不读就会退回扁平列表）
- 禁用图像模型不能用 joined-HTML 的 String.replace（只改第一个）
- 生图页存在且注册进 TABS（漏了则 pane 永不可见，症状是「页面空白」）
- sendImage 必须打 /v1/images/generations
- urlHint 必须同时提到 /v1 和 404
- TestImageEndpointComposesUnderRealConditions 用真实形态验证：
  相对路径拼出 /v1/images/generations 得 200，默认路径拼出 /v1/v1/… 得 404

.gitignore

@@ -26,3 +26,7 @@ cmd/gui/dist/
 
 # local design/working notes (not part of the shipped repo)
 /plan.md
+
+# scratch build/probe dirs used while debugging (GOTMPDIR, throwaway binaries)
+.build-work/
+.probe/

README.md

@@ -24,6 +24,7 @@
 ### 强大的多租户调度能力
 
 - **多密钥多租户**：支持无限密钥，每个密钥独立角色、模型范围、Token 配额、重置周期
+- **按模型配额**：每把 key 的每个模型单独配 token + 请求数配额与重置周期（小时/周/月/自定义 N 小时）；一个模型用满只影响该模型，同 key 其它模型照常；耗尽返 429 + `Retry-After` 并点名模型，admin key 永不受限
 - **AUTO 智能调度**：基于优先级档位的分级调度，同优先级源自动轮询负载均衡，故障自动毫秒级故障转移
 - **自愈冷却**：冷却上限 5 分钟，过半后放行 1 个探测请求，上游/额度恢复即刻回归轮询，无需等满冷却窗口
 - **Token 配额管理**：精确到模型级别的 Token 配额控制，支持小时/周/月/自定义小时周期自动重置
@@ -149,6 +150,47 @@ sources:
     timeout: 120s              # 请求超时，默认 120s
 ```
 
+
+### 自动探测模型列表
+
+不想手抄模型 id 时，只填 URL 和 key，让网关去扫：
+
+```bash
+# 只探测，不写任何东西
+curl -X POST http://127.0.0.1:8080/api/sources/discover \
+  -H "Authorization: Bearer sk-gw-…" -H "Content-Type: application/json" \
+  -d '{"base_url":"https://api.vendor.com/v1","api_key":"sk-…"}'
+# → {"ok":true,"base_url":"…","endpoint":"","models":["gpt-x",…],
+#    "image_models":["gpt-image-2"],"models_path":"/models","latency_ms":1387,…}
+
+# 探测后直接落库
+curl -X POST http://127.0.0.1:8080/api/sources/discover/apply \
+  -H "Authorization: Bearer sk-gw-…" -H "Content-Type: application/json" \
+  -d '{"base_url":"https://api.vendor.com/v1","api_key":"sk-…","name":"vendor"}'
+```
+
+WebUI 源编辑框里也有「自动探测模型」按钮：填好 URL 和 key 点一下，列表自动
+填好，仍需点保存才生效。
+
+几点值得知道：
+
+- **URL 形态不用猜。** 厂商对「`/v1` 放 base 还是放 path」没有共识，探测会
+  依次尝试若干候选并返回答对的那个，同时给出归一化后的 `base_url` +
+  `endpoint`（`endpoint` 平时留空，因为 openai adapter 自带 `/chat/completions`）。
+- **`image_endpoint` 是另一套规则。** 网关的图像默认路径是**绝对**的
+  `/v1/images/generations`，而 chat 默认是相对的。所以 `base_url` 带 `/v1` 时
+  必须显式给 `image_endpoint: /images/generations`，否则图像请求 404。
+  探测会替你填好。
+- **图像模型只看上游自己声明的 modality**，绝不从模型名猜。`gpt-image-2`
+  这类 id 也出现在同时提供对话服务的目录上，猜错会让 AUTO 链选中一个只支持
+  图像生成的模型，其后每一层都被它拖垮。
+- **不会逐模型探活。** 34 个模型的目录跑一轮要花钱且耗时数分钟，`/models`
+  是足够便宜的信号；模型是否真能应答留给正常请求去发现。
+- **对已存在的源，必须显式 `replace_models: true`** 才会覆盖模型列表。重新
+  探测顺手抹掉为 AUTO 链手工调过的 priority，是最难排查的一类数据丢失。
+- **`api_key` 留空表示沿用已存值**，不会把凭据清空。
+
+
 `config.yaml`、`runtime.json`、`master.key`、`adapters/` 均不提交版本库（见
 `.gitignore`），请随备份带走——`master.key` 丢失后运行时密文无法解密。
 
@@ -178,6 +220,81 @@ sources:
 - 客户端用任意一个已授权的密钥明文作为 Bearer（`Authorization: Bearer <key>`）。
 - 删除密钥即从运行时存储移除，立即失效。
 
+#### 密钥用量配额
+
+**配额按模型单独设置**：每个密钥的 `models[]` 里，每一条模型范围各自带一份
+token 配额与请求数配额。一个模型用满只影响该模型，同一密钥的其它模型照常工作。
+
+```yaml
+keys:
+  - key: sk-gw-<hex>
+    role: user
+    name: agent-alice
+    models:
+      - model: deepseek-v4-flash
+        token_quota: 1000000   # 本周期内该 key 用这个模型的 token 预算
+        req_quota: 20000       # 本周期内的请求次数
+        period: nhour          # "" | hour | week | month | nhour
+        hours: 6               # 仅 nhour
+      - model: AUTO
+        token_quota: 5000000   # AUTO 也是一条独立配额
+        period: day            # ← 这种写法会被拒绝（词表只有 hour/week/month/nhour）
+      - model: kimi-k3        # 未列配额 = 无限
+```
+
+- **没有「整钥总配额」**：这是刻意的设计。整钥总额会让「把 A 模型的额度挪给
+  B 模型」变成一次全局重分配；按模型独立计费则每个模型各自可控，运维可以
+  看出哪个模型吃掉了预算。
+- `period` 词表：空 = 永不过期（累计总量），`hour` / `week` / `month` = 固定窗口，
+  `nhour` + `hours` = 自定义小时数。**拼错的周期在写入时就被拒**，不会静默变成
+  永不过期。
+- 配额严格按密钥隔离，且**同一密钥内按模型隔离**：A 密钥用满 `m1` 不会消耗
+  B 密钥的额度，同一密钥的 `m2` 也不受影响。
+- 配额统计含聊天、流式、生图，跨重启从审计日志回放（保留 40 天，覆盖最长的
+  month 窗口）。
+- 配额耗尽返回 **429 + `Retry-After`**（`rate_limit_exceeded`），消息里点名是哪个
+  模型用满了，客户端可等窗口重置后自动恢复；模型越权才是 403。
+  **admin 密钥永不受配额限制**（其 scope 上的配额也不强制），避免把管理员
+  锁在门外。
+- 窗口用量按整点小时分桶统计，实际释放比配置窗口最多晚 1 小时（配额宁可晚释放
+  也不超发）。
+- `PUT /api/keys/{key}` 提交 `models` 即同时提交它们的配额（配额就是 scope 的一部分，
+  不存在会与模型列表脱节的第二份预算）。显式 `0` = 解除该模型的限制。
+
+##### 配额拒绝 vs 容量拒绝：两种「拒绝」含义不同
+
+配额在**入口**检查，不占用任何上游槽位，因此是廉价拒绝；容量不足则要走
+`busyWait` 有界等待后才返回 503。实测（容量 8、0.6s/请求、100 并发）：
+
+| 情况 | 状态码 | 延迟 | 是否打到上游 |
+|---|---|---|---|
+| 配额用尽（`req_quota` 50 → 100 并发） | `429 rate_limit_exceeded` + `Retry-After` | **19–21ms** | 否 |
+| 容量打满（同一拓扑，未限配额） | `503 upstream_error` | ~2.6s（`busyWait`） | 超出部分否 |
+
+客户端因此可以区分「等窗口重置」与「等上游腾出容量」：前者只需等待，后者
+通常该降并发或换源。
+
+##### 配额的开销
+
+配额桶按 (密钥, 模型, 整点小时) 分桶保留 40 天，实测（AMD 7840HS）：
+
+- 窗口查询按窗口长度而非保留总量扫描：1h 窗口 49ns、24h 窗口 55ns、
+  30d 窗口 3.9μs（此前全扫保留总量，960 桶时 5.9μs）。
+- 记录一条请求：283ns、3 分配（与引入配额前相同，分配来自 ring buffer）。
+- 内存：生产形态（7 密钥 × 8 模型 × 2 源 × 满 40 天 retention）约 **3.7MB**。
+  按源 pin 的 `source::model` 桶**惰性创建**——只有当某条配额真的 pin 了
+  某个源时才维护，否则每条记录多写一份桶，在 20 密钥 × 8 模型 × 3 源下会
+  多占 18MB。
+  代价：配置 pinned 配额之前的历史用量无法事后按源拆分，所以 **pinned 配额的
+  首个窗口可能少算**。
+
+```bash
+# 配额耗尽时客户端看到（点名了具体模型）
+HTTP/1.1 429 Too Many Requests
+Retry-After: 2100
+{"error":{"type":"rate_limit_exceeded","message":"token quota exceeded for \"deepseek-v4-flash\" (5000000/5000000)"}}
+```
+
 ### 模型路由
 
 `/v1/chat/completions` 的 `model` 解析顺序：
@@ -344,7 +461,10 @@ Environment=MALLOC_ARENA_MAX=2
   支持按时间范围导出 CSV；点击模型可生成 pin 到该模型的连接配置。
 - **对话页**：流式/非流式调试。
 - **密钥页**：创建/编辑网关 key，为每个 key 配模型范围（模型 + 源 + token 配额 + 周期），
-  管理员管理全部 key，用户只看到自己的 key。
+  管理员管理全部 key，用户只看到自己的 key。每个模型砖块显示自己的配额徽标
+  （如 `1.0K 77×·1h`，未设配额显示 `∞`），点开可编辑该模型的 token 配额、
+  请求数配额与重置周期 —— 配额按模型独立生效，一个用满不影响同一 key 的其它模型。
+  「我的密钥」页对用户展示本 key 的模型范围与各自配额。
 - **优先级页**：拖拽积木配置 AUTO 链档位。
 - **源页**：在线增删改上游源（API key 等敏感字段加密落盘）。
 - **适配器页**：上传 / 删除 Lua 适配器脚本。

README_EN.md

@@ -38,6 +38,11 @@ Extracted and independently evolved from the multi-source LLM adapter layer of
   `reasoning_content`, `tool_calls`, `usage`).
 - **Image generation**: `POST /v1/images/generations`, routed to models with
   `kind: image`.
+- **Per-model quota**: each key gives every model its own token and request caps
+  plus a reset period (hour/week/month/custom N hours). One model running out
+  affects only that model — the key's other models keep working. Exhaustion
+  answers 429 + `Retry-After` naming the model, so a client resumes when the
+  window rolls over; admin keys are never capped.
 - **Multimodal**: `content` arrays (`image_url` etc.) pass through losslessly;
   Anthropic/Gemini/Ollama are translated automatically.
 - **LuaJIT VM**: golua-binding LuaJIT; each adapter has its own VM + worker
@@ -171,11 +176,103 @@ under the `keys` field of the runtime file (encrypted at rest):
   delete the seed key.
 - The WebUI **Keys page** creates/deletes keys. Each key has a role (`admin`
   manages everything, `user` sees only its own key) and an optional **model
-  scope** (model + source + token quota + reset period).
+  scope** (model + source + token quota + reset period). Each model brick shows
+  its own budget badge (e.g. `1.0K 77×·1h`, `∞` when uncapped); clicking it
+  edits that model's token quota, request quota and reset period. Quotas apply
+  per model, so one model running out never blocks the key's others. The
+  "My key" view shows a user its model scopes and their budgets.
 - Clients authenticate with any authorized key's plaintext as
   `Authorization: Bearer <key>`.
 - Deleting a key removes it from the store immediately.
 
+#### Per-key usage quota
+
+**Quotas are per model.** Each key's `models[]` list gives every model its own
+token budget and request budget. One model running out affects only that model
+— the key's other models keep working.
+
+```yaml
+keys:
+  - key: sk-gw-<hex>
+    role: user
+    name: agent-alice
+    models:
+      - model: deepseek-v4-flash
+        token_quota: 1000000   # this key's token budget for this model
+        req_quota: 20000       # requests within the window
+        period: nhour          # "" | hour | week | month | nhour
+        hours: 6               # n-hour only
+      - model: AUTO
+        token_quota: 5000000   # AUTO is a quota entry like any other
+        period: hour
+      - model: kimi-k3        # no quota listed = unlimited
+```
+
+- **There is deliberately no key-wide total.** A key-wide cap would make
+  "move A's budget to B" a global reallocation; per-model budgets keep each
+  model independently controllable, so it stays visible which model is
+  actually consuming the spend.
+- `period`: empty = never resets (lifetime total); `hour` / `week` / `month` =
+  fixed windows; `nhour` + `hours` = a custom hour count. **A misspelled
+  period is rejected at write time** rather than silently becoming a
+  never-resetting quota.
+- Quotas are isolated per key *and* per model within a key: one key exhausting
+  `m1` never draws on another key's budget, and never blocks the same key's
+  `m2`.
+- Usage counts chat, streaming and image requests, and survives a restart by
+  replaying the audit log (40 days retained, covering the longest `month`
+  window).
+- An exhausted quota returns **429 + `Retry-After`**
+  (`rate_limit_exceeded`) and the message names the model that ran out, so a
+  client resumes when the window rolls over; a model the key may not use stays
+  403. **Admin keys are never capped** (quotas on their scopes are not
+  enforced either), so a cap can never lock the operator out.
+- Buckets are whole unix hours, so a window frees up at most an hour late
+  (deliberately freeing late rather than overspending).
+- `PUT /api/keys/{key}` submits quotas by submitting `models` — the caps are
+  part of the scope, so there is no second budget that can drift out of sync
+  with the model list. An explicit `0` lifts that model's cap.
+
+##### Quota rejection vs capacity rejection
+
+A quota is checked at the entry point and occupies no upstream slot, so it is a
+cheap rejection. Running out of capacity instead waits out the bounded
+`busyWait` before answering 503. Measured (capacity 8, 0.6 s/request, 100
+concurrent clients):
+
+| Case | Status | Latency | Reaches upstream |
+|---|---|---|---|
+| Quota spent (`req_quota` 50, 100 concurrent) | `429 rate_limit_exceeded` + `Retry-After` | **19–21 ms** | no |
+| Capacity exhausted (same topology, no quota) | `503 upstream_error` | ~2.6 s (`busyWait`) | overflow: no |
+
+So a client can tell "wait for the window to roll over" apart from "wait for
+upstream capacity" — the first only needs patience, the second usually means
+back off concurrency or switch sources.
+
+##### What the quota costs
+
+Buckets are kept per (key, model, whole unix hour) for 40 days. Measured on an
+AMD 7840HS:
+
+- A window query scans the window, not the whole retention: 49 ns for 1 h,
+  55 ns for 24 h, 3.9 µs for 30 d.
+- Recording one request: 283 ns, 3 allocations (unchanged from before this
+  feature; the allocations come from the record ring buffer).
+- Memory: the production shape (7 keys x 8 models x 2 sources at full 40-day
+  retention) costs about **3.7 MB**. The `source::model` bucket used by a
+  source-pinned quota is created **lazily** — it is maintained only once some
+  quota actually pins that source, because writing it on every record would
+  cost an extra 18 MB at 20 keys x 8 models x 3 sources.
+  The trade-off: usage recorded before a pinned quota was configured cannot be
+  split by source afterwards, so **a pinned quota may under-count its first
+  window**.
+
+```
+HTTP/1.1 429 Too Many Requests
+Retry-After: 2100
+{"error":{"type":"rate_limit_exceeded","message":"token quota exceeded for \"deepseek-v4-flash\" (5000000/5000000)"}}
+```
+
 ### Model routing
 
 `/v1/chat/completions` `model` resolution order:

cmd/gui/main.js

@@ -17,7 +17,7 @@ const fs = require("fs");
 const http = require("http");
 const os = require("os");
 const crypto = require("crypto");
-const { spawn } = require("child_process");
+const { spawn, execFileSync } = require("child_process");
 
 const LOG_FILE = path.join(app.getPath("userData"), "gui.log");
 function log(msg) {
@@ -97,24 +97,99 @@ function readAdminKey() {
   // gateway_keys entries get marked seed:true by the core (-> "replace the
   // initial key" warning in the web UI), while a plain keys entry authenticates
   // without any seed warning.
+  const raw = readConfigRaw();
+  if (!raw) return "";
+  // keys:
+  //     - key: sk-gw-...
+  const k = /^\s*keys:\s*\n(?:\s*-\s*key:\s*"?([^\s"#]+)"?[^\n]*\n)+/m.exec(
+    raw,
+  );
+  let v = k && k[1] ? k[1] : "";
+  if (!v) {
+    // legacy: gateway_keys list (pre-migration configs)
+    const g = /^\s*gateway_keys:\s*\n(?:\s*-\s*"?([^\s"#]+)"?\s*\n)+/m.exec(raw);
+    if (g && g[1]) v = g[1];
+    if (!v) {
+      const g2 = /gateway_keys:[^[\n]*\[\s*"?([^\s"\]]+)"?/m.exec(raw);
+      if (g2 && g2[1]) v = g2[1];
+    }
+  }
+  // The core seals credentials at rest (enc:v1:...), so from the second start
+  // onward the file holds ciphertext. Injecting that as a cookie is a
+  // guaranteed 401, which surfaces to the user as "enter your key" for a key
+  // they never set. Ask the core to unseal rather than reimplementing its
+  // crypto here. -show-secrets only reads, prints and exits: it starts nothing
+  // and writes nothing.
+  if (v && v.indexOf(SEALED_PREFIX) === 0) return unsealViaCore();
+  return v;
+}
+
+const SEALED_PREFIX = "enc:v1:";
+
+function readConfigRaw() {
   try {
     if (!fs.existsSync(CONFIG_FILE)) return "";
-    const raw = fs.readFileSync(CONFIG_FILE, "utf-8");
-    // keys:
-    //     - key: sk-gw-...
-    const k = /^\s*keys:\s*\n(?:\s*-\s*key:\s*"?([^\s"#]+)"?[^\n]*\n)+/m.exec(
-      raw,
+    return fs.readFileSync(CONFIG_FILE, "utf-8");
+  } catch (e) {
+    return "";
+  }
+}
+
+// Returns the admin key in the clear, or "" if the core cannot be asked.
+//
+// cwd must be the directory that holds master.key. The core derives that path
+// from the config's runtime_file, so a call made from any other directory has
+// the core generate a *second* master key in the CWD and then fail to decrypt
+// ("master key changed?"). Electron's CWD is not the profile dir, so this is
+// not optional: without it the desktop build cannot read its own key back.
+//
+// readAdminKey runs on every outbound request, and this spawns a process, so
+// the result is cached against the config's mtime. Editing the config (or the
+// port rewrite, or the first write) changes the mtime and invalidates it, which
+// keeps the "read live so ordering never matters" property the auth rule
+// depends on.
+function unsealViaCore() {
+  const st = safeStat(CONFIG_FILE);
+  const stamp = st ? st.mtimeMs : 0;
+  if (unsealCache && unsealCache.stamp === stamp) return unsealCache.key;
+  let key = "";
+  try {
+    const out = execFileSync(
+      CORE_EXE,
+      ["-config", CONFIG_FILE, "-show-secrets"],
+      {
+        encoding: "utf-8",
+        timeout: 10000,
+        windowsHide: true,
+        cwd: PROFILE_DIR,
+      },
     );
-    if (k && k[1]) return k[1];
-    // legacy: gateway_keys list (pre-migration configs)
-    const g = /^\s*gateway_keys:\s*\n(?:\s*-\s*"?([^\s"#]+)"?\s*\n)+/m.exec(
-      raw,
-    );
-    if (g && g[1]) return g[1];
-    const g2 = /gateway_keys:[^[\n]*\[\s*"?([^\s"\]]+)"?/m.exec(raw);
-    if (g2 && g2[1]) return g2[1];
-  } catch (e) {}
-  return "";
+    // -show-secrets prints one line per credential:
+    //   key    <name>  role=<role>  <value>
+    const admin = /^key\s+\S+\s+role=admin\s+(\S+)\s*$/m.exec(out);
+    if (admin && admin[1]) key = admin[1];
+    // Tolerate a field-order or spacing change rather than locking the user out.
+    if (!key) {
+      const loose = /role=admin\s+(\S+)/.exec(out);
+      if (loose && loose[1]) key = loose[1];
+    }
+  } catch (e) {
+    console.error("unseal via core failed:", e.message);
+  }
+  // Only cache a success. A transient failure must not pin an empty key until
+  // the config next changes, or a momentary spawn error locks the user out.
+  if (key) unsealCache = { stamp, key };
+  return key;
+}
+
+let unsealCache = null;
+
+function safeStat(p) {
+  try {
+    return fs.statSync(p);
+  } catch (e) {
+    return null;
+  }
 }
 
 const embeddedBaseUrl = () =>
@@ -596,6 +671,78 @@ ipcMain.handle(
   (e) => !!BrowserWindow.fromWebContents(e.sender)?.isMaximized(),
 );
 
+// ---- plugin management over IPC -------------------------------------------
+//
+// The renderer cannot call the embedded core directly: it has no key and no
+// network identity, and the core binds a loopback port that only the main
+// process knows about. So every plugin action is proxied through the main
+// process, which already knows how to obtain the admin key (unsealViaCore).
+//
+// The proxy is deliberately a raw (method, path, body) pass-through rather than
+// a fixed set of commands. A fixed set would have to be extended for every new
+// plugin endpoint, and the one thing worse than "no button for this" is "a
+// button that silently does nothing" — with a pass-through the renderer can talk
+// to any /api/plugins route the core grows, and the path is validated here so
+// this channel cannot be used to reach arbitrary endpoints.
+function pluginProxy(req) {
+  const { method, path, body } = req || {};
+  const M = ["GET", "POST", "PUT", "DELETE"];
+  if (!M.includes(method)) throw new Error("bad method: " + method);
+  // The path must stay inside the plugin namespace. A prefix check alone would
+  // still allow /api/plugins/../keys, so reject any traversal outright.
+  if (typeof path !== "string" || !path.startsWith("/api/plugins")) {
+    throw new Error("path must start with /api/plugins");
+  }
+  if (path.includes("..") || path.includes("\\")) {
+    throw new Error("path traversal rejected");
+  }
+  const key = unsealViaCore();
+  if (!key) throw new Error("no admin key available yet");
+  return new Promise((resolve, reject) => {
+    const u = new URL(embeddedBaseUrl() + path);
+    const data = body == null ? null : JSON.stringify(body);
+    const headers = { Authorization: "Bearer " + key };
+    if (data) {
+      headers["Content-Type"] = "application/json";
+      headers["Content-Length"] = Buffer.byteLength(data);
+    }
+    const r = http.request(
+      {
+        hostname: u.hostname,
+        port: u.port,
+        path: u.pathname + u.search,
+        method,
+        headers,
+      },
+      (res) => {
+        let raw = "";
+        res.setEncoding("utf8");
+        res.on("data", (c) => (raw += c));
+        res.on("end", () => {
+          let parsed = null;
+          try {
+            parsed = raw ? JSON.parse(raw) : null;
+          } catch (e) {
+            parsed = { raw };
+          }
+          if (res.statusCode >= 400) {
+            const msg =
+              (parsed && parsed.error && parsed.error.message) ||
+              "HTTP " + res.statusCode;
+            reject(new Error(msg));
+            return;
+          }
+          resolve(parsed);
+        });
+      },
+    );
+    r.on("error", reject);
+    if (data) r.write(data);
+    r.end();
+  });
+}
+
+ipcMain.handle("plugins:proxy", (_e, req) => pluginProxy(req));
 ipcMain.handle("core:state", () => ({
   running: coreStarted() && coreReady,
   ready: coreReady,

cmd/gui/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "modelrouter-gui",
-  "version": "1.4.2",
+  "version": "1.9.3",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "modelrouter-gui",
-      "version": "1.4.2",
+      "version": "1.9.3",
       "license": "Proprietary",
       "devDependencies": {
         "electron": "^33.0.0",
@@ -582,7 +582,7 @@
       }
     },
     "node_modules/@peculiar/webcrypto": {
-      "version": "1.7.1",
+      "version": "1.7.5",
       "resolved": "https://registry.npmjs.org/@peculiar/webcrypto/-/webcrypto-1.7.1.tgz",
       "integrity": "sha512-ODOov0sGMJMf3jPonOkgGqPknTsu+DdQ7kD++gz8aI+aFMOMHFbWAA2taqXXVTdP+OTOQR/znGvSpmkeI0WTYQ==",
       "dev": true,
@@ -3335,7 +3335,7 @@
       }
     },
     "node_modules/resedit": {
-      "version": "1.7.2",
+      "version": "1.7.5",
       "resolved": "https://registry.npmjs.org/resedit/-/resedit-1.7.2.tgz",
       "integrity": "sha512-vHjcY2MlAITJhC0eRD/Vv8Vlgmu9Sd3LX9zZvtGzU5ZImdTN3+d6e/4mnTyV8vEbyf1sgNIrWxhWlrys52OkEA==",
       "dev": true,

cmd/gui/package.json

@@ -1,6 +1,6 @@
 {
   "name": "modelrouter-gui",
-  "version": "1.4.2",
+  "version": "1.9.7",
   "description": "ModelRouter Desktop — embedded ModelRouter core with tray",
   "author": "ModelRouter",
   "main": "main.js",

cmd/gui/preload.js

@@ -16,6 +16,13 @@ contextBridge.exposeInMainWorld("modelrouter", {
     key: () => ipcRenderer.invoke("core:key"),
     onState: (cb) => ipcRenderer.on("core:state", (_e, d) => cb(d)),
   },
+  plugins: {
+    // Raw pass-through to the embedded core's /api/plugins surface. The main
+    // process validates the path and attaches the admin key; the renderer never
+    // sees either.
+    request: (method, path, body) =>
+      ipcRenderer.invoke("plugins:proxy", { method, path, body }),
+  },
   settings: {
     get: () => ipcRenderer.invoke("settings:get"),
     set: (patch) => ipcRenderer.invoke("settings:set", patch),

cmd/gui/renderer/app.js

@@ -155,6 +155,7 @@ async function openSettings() {
   $("#set-tray").checked = !!state.settings.minimizeToTray;
   $("#settings-overlay").style.display = "flex";
   renderRail();
+  loadPlugins();
 }
 function closeSettings() {
   $("#settings-overlay").style.display = "none";
@@ -182,6 +183,296 @@ async function saveSettings() {
   }
 }
 
+// esc / escAttr escape text for innerHTML. The WebUI has its own copies; the
+// shell needs its own because renderer/app.js is a separate document that
+// never loads index.html's script.
+function esc(s) {
+  return String(s == null ? "" : s).replace(
+    /[&<>"']/g,
+    (c) => ({ "&": "&amp;", "<": "&lt;", ">": "&gt;", '"': "&quot;", "'": "&#39;" })[c],
+  );
+}
+function escAttr(s) {
+  return esc(s).replace(/`/g, "&#96;");
+}
+
+// ===== plugin management =====
+//
+// The desktop shell manages plugins through the embedded core's /api/plugins
+// surface, proxied over IPC (see plugins:proxy in the main process). The
+// renderer never holds the admin key.
+//
+// Scope note: the desktop build has no plugin_dir configured by default, so this
+// panel normally reports "plugins disabled" with the one-line fix. That is a
+// deliberate state, not an error — the packaged profile is a per-user directory
+// and seeding a plugin tree into someone's home without asking would be rude.
+
+async function loadPlugins() {
+  const list = $("#pl-list");
+  const hint = $("#set-plugins-hint");
+  if (!list || !hint) return;
+  let j;
+  try {
+    j = await window.modelrouter.plugins.request("GET", "/api/plugins");
+  } catch (e) {
+    hint.textContent = "内核未就绪：" + e.message;
+    list.innerHTML = "";
+    return;
+  }
+  if (!j.plugin_dir) {
+    hint.innerHTML =
+      '未配置 <code>plugin_dir</code>，插件功能未启用。在 config.yaml 加一行后重启内核即可。';
+    list.innerHTML = "";
+    return;
+  }
+  const rows = j.on_disk || [];
+  window.__plRows = rows; // read by editPlugin/deletePlugin for the builtin flag
+  const active = rows.filter((p) => p.loaded && !p.disabled).length;
+  const broken = rows.filter((p) => !p.loaded).length;
+  hint.textContent =
+    `${rows.length} 个插件 · ${active} 个启用中` +
+    (broken ? ` · ${broken} 个加载失败` : "");
+  list.innerHTML = rows.length
+    ? rows
+        .map((p) => {
+          const cls = !p.loaded ? "pl-broken" : p.disabled ? "pl-off" : "pl-on";
+          const label = !p.loaded
+            ? "加载失败"
+            : p.disabled
+              ? "已禁用"
+              : "启用中";
+          const btn = p.loaded
+            ? `<button class="ghost" data-act="toggle" data-name="${escAttr(
+                p.name,
+              )}" data-en="${p.disabled ? "1" : "0"}">${
+                p.disabled ? "启用" : "禁用"
+              }</button>`
+            : "";
+          const builtin = p.builtin
+            ? '<span class="pl-builtin">内置</span>'
+            : "";
+          // Edit / view source / delete are the operations an operator
+          // actually needs after enable/disable: a plugin that cannot be
+          // read back cannot be fixed, and one that cannot be removed cannot
+          // be rolled back.
+          const acts = [
+            btn,
+            `<button class="ghost" data-act="code" data-name="${escAttr(p.name)}">${
+              p.builtin ? "查看源码" : "编辑"
+            }</button>`,
+            builtin
+              ? ""
+              : `<button class="ghost danger" data-act="del" data-name="${escAttr(
+                  p.name,
+                )}">删除</button>`,
+          ]
+            .filter(Boolean)
+            .join("");
+          const pages = Array.isArray(p.pages) ? p.pages : [];
+          return `<div class="pl-item ${cls}">
+            <div class="pl-head"><b>${esc(p.name)}</b>${builtin}<span class="pl-state">${label}</span></div>
+            ${p.description ? `<div class="pl-desc">${esc(p.description)}</div>` : ""}
+            ${pages.length ? `<div class="pl-desc">页面：${esc(pages.join(", "))}</div>` : ""}
+            ${p.error ? `<div class="pl-err">${esc(String(p.error).slice(0, 160))}</div>` : ""}
+            <div class="pl-acts">${acts}</div>
+          </div>`;
+        })
+        .join("")
+    : '<div class="pl-empty">插件目录为空</div>';
+  list.querySelectorAll('button[data-act="toggle"]').forEach((b) => {
+    b.onclick = () => togglePlugin(b.dataset.name, b.dataset.en === "1");
+  });
+  list.querySelectorAll('button[data-act="code"]').forEach((b) => {
+    b.onclick = () => editPlugin(b.dataset.name);
+  });
+  list.querySelectorAll('button[data-act="del"]').forEach((b) => {
+    b.onclick = () => deletePlugin(b.dataset.name);
+  });
+}
+
+// editPlugin shows the plugin source in a modal. Built-in plugins are
+// read-only: their file ships inside the binary, so a save would either be
+// ignored or silently do nothing — showing an editable text area that
+// discards on save is worse than saying so up front.
+function editPlugin(name) {
+  // The plugin list is the only place `builtin` is reported; keep the last
+  // fetch so the modal does not need a second round trip (and cannot guess).
+  const row = (window.__plRows || []).find((p) => p.name === name);
+  const isBuiltin = !!(row && row.builtin);
+  const wrap = document.createElement("div");
+  wrap.className = "overlay";
+  wrap.innerHTML = `<div class="overlay-card pl-wide">
+    <div class="overlay-head"><h2>${esc(name)}${isBuiltin ? "（内置，只读）" : ""}</h2></div>
+    ${isBuiltin
+      ? `<p class="hint">内置插件随内核发布，源码在二进制内，不可编辑。</p>`
+      : `<textarea id="pl-code" spellcheck="false"></textarea>
+         <p class="hint" id="pl-code-err"></p>`}
+    <div class="row actions">
+      ${isBuiltin ? "" : `<button id="pl-code-save">保存并热重载</button>`}
+      <button class="ghost" id="pl-code-close">关闭</button>
+    </div>
+  </div>`;
+  document.body.appendChild(wrap);
+  wrap.addEventListener("click", (e) => {
+    if (e.target === wrap) wrap.remove();
+  });
+  $("#pl-code-close").onclick = () => wrap.remove();
+  const save = $("#pl-code-save");
+  if (save) {
+    save.onclick = async () => {
+      save.disabled = true;
+      try {
+        // Saving writes through POST /api/plugins — the same install path the
+        // installer uses, which validates the name, compiles the source and
+        // hot-reloads. PUT /{name} only toggles enable/disable, so pointing the
+        // save there would silently disable nothing and store nothing.
+        await window.modelrouter.plugins.request("POST", "/api/plugins", {
+          name,
+          code: $("#pl-code").value,
+        });
+        toast(`${name} 已保存`);
+        wrap.remove();
+        await loadPlugins();
+      } catch (e) {
+        $("#pl-code-err").textContent = e.message;
+        save.disabled = false;
+      }
+    };
+  }
+  if (isBuiltin) return;
+  (async () => {
+    const ta = $("#pl-code");
+    try {
+      const j = await window.modelrouter.plugins.request(
+        "GET",
+        `/api/plugins/${encodeURIComponent(name)}`,
+      );
+      ta.value = j.code || "";
+      const err = $("#pl-code-err");
+      if (err && j.parse_error) err.textContent = String(j.parse_error);
+    } catch (e) {
+      $("#pl-code-err").textContent = e.message;
+    }
+  })();
+}
+
+function deletePlugin(name) {
+  const wrap = document.createElement("div");
+  wrap.className = "overlay";
+  wrap.innerHTML = `<div class="overlay-card">
+    <div class="overlay-head"><h2>删除插件</h2></div>
+    <p>确定删除 <b>${esc(name)}</b>？插件目录中的文件会被移除，核心将不再加载它。</p>
+    <div class="row actions">
+      <button class="danger" id="pl-del-ok">删除</button>
+      <button class="ghost" id="pl-del-no">取消</button>
+    </div>
+  </div>`;
+  document.body.appendChild(wrap);
+  wrap.addEventListener("click", (e) => {
+    if (e.target === wrap) wrap.remove();
+  });
+  $("#pl-del-no").onclick = () => wrap.remove();
+  $("#pl-del-ok").onclick = async () => {
+    try {
+      await window.modelrouter.plugins.request(
+        "DELETE",
+        `/api/plugins/${encodeURIComponent(name)}`,
+      );
+      toast(`已删除 ${name}`);
+      wrap.remove();
+      await loadPlugins();
+    } catch (e) {
+      toast(e.message, true);
+    }
+  };
+}
+
+// installPlugin creates a new plugin from source. The name becomes the file
+// name, so it is validated client-side to save a round trip on the obvious
+// mistakes (empty, path separators, non-portable characters).
+function installPlugin() {
+  const wrap = document.createElement("div");
+  wrap.className = "overlay";
+  wrap.innerHTML = `<div class="overlay-card pl-wide">
+    <div class="overlay-head"><h2>安装插件</h2></div>
+    <div class="row"><label>名称</label><input id="pl-new-name" placeholder="my_plugin" /></div>
+    <p class="hint" id="pl-new-err"></p>
+    <textarea id="pl-new-code" spellcheck="false" placeholder="-- 单文件 Lua 插件&#10;plugin = {}&#10;..."></textarea>
+    <div class="row actions">
+      <button id="pl-new-ok">安装</button>
+      <button class="ghost" id="pl-new-no">取消</button>
+    </div>
+  </div>`;
+  document.body.appendChild(wrap);
+  wrap.addEventListener("click", (e) => {
+    if (e.target === wrap) wrap.remove();
+  });
+  $("#pl-new-no").onclick = () => wrap.remove();
+  $("#pl-new-ok").onclick = async () => {
+    const name = $("#pl-new-name").value.trim();
+    const code = $("#pl-new-code").value;
+    const err = $("#pl-new-err");
+    if (!/^[A-Za-z0-9_-]+$/.test(name)) {
+      err.textContent = "名称只能包含字母、数字、下划线和连字符";
+      return;
+    }
+    if (!code.trim()) {
+      err.textContent = "源码不能为空";
+      return;
+    }
+    $("#pl-new-ok").disabled = true;
+    try {
+      await window.modelrouter.plugins.request("POST", "/api/plugins", {
+        // The backend appends .lua itself (installPlugin writes name+".lua"),
+        // so passing it here would create name.lua.lua.
+        name,
+        code,
+      });
+      toast(`已安装 ${name}`);
+      wrap.remove();
+      await loadPlugins();
+    } catch (e) {
+      err.textContent = e.message;
+      $("#pl-new-ok").disabled = false;
+    }
+  };
+}
+
+async function togglePlugin(name, disabled) {
+  try {
+    await window.modelrouter.plugins.request("PUT", `/api/plugins/${encodeURIComponent(name)}`, {
+      enabled: disabled,
+    });
+    toast(disabled ? `已禁用 ${name}` : `已启用 ${name}`);
+    await loadPlugins();
+  } catch (e) {
+    toast(e.message, true);
+  }
+}
+
+async function enableAllPlugins() {
+  let j;
+  try {
+    j = await window.modelrouter.plugins.request("GET", "/api/plugins");
+  } catch (e) {
+    return toast(e.message, true);
+  }
+  const off = (j.on_disk || []).filter((p) => p.loaded && p.disabled);
+  for (const p of off) {
+    try {
+      await window.modelrouter.plugins.request(
+        "PUT",
+        `/api/plugins/${encodeURIComponent(p.name)}`,
+        { enabled: true },
+      );
+    } catch (e) {
+      toast(`${p.name}: ${e.message}`, true);
+    }
+  }
+  toast(off.length ? `已启用 ${off.length} 个插件` : "没有处于禁用状态的插件");
+  await loadPlugins();
+}
+
 // ===== theme =====
 function applyTheme() {
   document.documentElement.dataset.theme = state.theme;
@@ -197,6 +488,12 @@ function init() {
   $("#tb-close").onclick = () => window.modelrouter.win.close();
   $("#tb-settings").onclick = openSettings;
   $("#rail-settings").onclick = openSettings;
+  const plReload = document.getElementById("pl-reload");
+  if (plReload) plReload.onclick = loadPlugins;
+  const plAll = document.getElementById("pl-toggle-all");
+  if (plAll) plAll.onclick = enableAllPlugins;
+  const plInstall = document.getElementById("pl-install");
+  if (plInstall) plInstall.onclick = installPlugin;
   $("#rail-autostart").onclick = toggleAutoStart;
   $("#rail-silent").onclick = toggleSilent;
   $("#rail-theme").onclick = () => {

cmd/gui/renderer/index.html

@@ -207,6 +207,16 @@
               ><input type="checkbox" id="set-tray" /> 关闭时最小化到托盘</label
             ><span class="hint">点关闭按钮隐藏到系统托盘</span>
           </div>
+          <div class="row">
+            <label>插件</label>
+            <span class="hint" id="set-plugins-hint">加载中…</span>
+          </div>
+          <div id="pl-list" class="pl-list"></div>
+          <div class="row actions">
+            <button class="ghost" id="pl-toggle-all">全部启用</button>
+            <button class="ghost" id="pl-install">安装插件</button>
+            <button class="ghost" id="pl-reload">刷新</button>
+          </div>
           <div class="row actions">
             <button class="ghost" id="set-dir">打开数据目录</button>
             <button class="ghost" id="set-log">查看日志</button>

cmd/gui/renderer/style.css

@@ -596,3 +596,129 @@ html[data-theme="dark"] .overlay {
 #toast.err {
  border-color: var(--danger);
 }
+
+/* ===== plugin management panel =========================================
+ * The existing .ghost/.primary rules are scoped to `.form .actions`, so a
+ * button outside that selector gets browser defaults. The plugin rows live in
+ * their own list, hence their own rules — reusing a scoped class here would have
+ * produced unstyled buttons that still worked, which is the kind of thing that
+ * looks fine until someone themes the shell.
+ */
+.pl-list {
+ display: flex;
+ flex-direction: column;
+ gap: 8px;
+ margin: 8px 0 4px;
+}
+.pl-item {
+ border: 1px solid var(--line);
+ border-radius: 9px;
+ padding: 10px 12px;
+}
+.pl-item.pl-broken {
+ border-color: var(--danger, #d1435b);
+}
+.pl-head {
+ display: flex;
+ align-items: center;
+ gap: 8px;
+ font-size: 13px;
+}
+.pl-builtin {
+ font-size: 10px;
+ padding: 1px 6px;
+ border-radius: 999px;
+ background: var(--primary-50);
+ color: var(--primary-h);
+}
+.pl-state {
+ margin-left: auto;
+ font-size: 11px;
+ color: var(--muted);
+}
+.pl-item.pl-broken .pl-state {
+ color: var(--danger, #d1435b);
+}
+.pl-desc {
+ font-size: 12px;
+ color: var(--muted);
+ margin-top: 3px;
+}
+.pl-err {
+ font-size: 11px;
+ color: var(--danger, #d1435b);
+ margin-top: 4px;
+ word-break: break-word;
+}
+.pl-acts {
+ margin-top: 8px;
+ display: flex;
+ gap: 8px;
+}
+.pl-acts button {
+ padding: 5px 12px;
+ font-size: 12px;
+ border-radius: 7px;
+ border: 1px solid var(--line);
+ background: var(--bg-s2, #fff);
+ color: var(--fg, inherit);
+ cursor: pointer;
+ transition: all 0.15s;
+}
+.pl-acts button:hover {
+ border-color: var(--primary);
+ color: var(--primary-h);
+}
+.pl-empty {
+ font-size: 12px;
+ color: var(--muted);
+ padding: 10px 0;
+}
+
+/* ---- plugin editor / installer --------------------------------------------
+   The plugin source editor and the install dialog reuse the settings overlay
+   (overlay / overlay-card) rather than a second popup system. They need three
+   things the settings card does not: a wider column for code, a monospace
+   textarea, and a destructive button for delete. */
+.overlay-card.pl-wide {
+ width: 760px;
+}
+.pl-wide textarea {
+ width: 100%;
+ min-height: 380px;
+ resize: vertical;
+ padding: 10px 12px;
+ border-radius: 10px;
+ border: 1px solid var(--line-strong);
+ background: var(--bg-sunken, rgba(127, 127, 127, 0.06));
+ color: inherit;
+ font-family: ui-monospace, SFMono-Regular, Menlo, Consolas, monospace;
+ font-size: 12.5px;
+ line-height: 1.55;
+ tab-size: 2;
+}
+.pl-wide textarea:focus {
+ outline: none;
+ border-color: var(--primary);
+}
+.pl-wide .row input {
+ flex: 1;
+ padding: 7px 10px;
+ border-radius: 9px;
+ border: 1px solid var(--line-strong);
+ background: var(--bg-sunken, rgba(127, 127, 127, 0.06));
+ color: inherit;
+}
+.pl-editor .hint,
+.pl-wide .hint {
+ min-height: 16px;
+ color: var(--muted, #888);
+}
+button.danger {
+ color: #e5484d;
+ border-color: rgba(229, 72, 77, 0.4);
+}
+button.danger:hover {
+ border-color: #e5484d;
+ color: #e5484d;
+}

cmd/llmsproxy/main.go

@@ -24,6 +24,7 @@ import (
 func main() {
 	cfgPath := flag.String("config", "config.yaml", "path to gateway config file")
 	checkOnly := flag.Bool("check", false, "validate the config file and exit (0 = valid, 1 = invalid); nothing is started and no file is written")
+	showSecrets := flag.Bool("show-secrets", false, "print the config's credentials in the clear and exit; nothing is started and no file is written")
 	flag.Parse()
 
 	// -check is the deploy-time preflight: parse and validate the config
@@ -41,6 +42,20 @@ func main() {
 		return
 	}
 
+	// -show-secrets is the operator escape hatch for a config whose
+	// credentials are sealed at rest: it prints them to stdout and exits
+	// without starting anything or writing anything. Nothing else in the
+	// program prints a credential, and this path never logs to a file.
+	if *showSecrets {
+		if _, err := os.Stat(*cfgPath); err != nil {
+			log.Fatalf("[llmsproxy] show-secrets: %v", err)
+		}
+		if err := config.PrintSecrets(*cfgPath); err != nil {
+			log.Fatalf("[llmsproxy] show-secrets: %v", err)
+		}
+		return
+	}
+
 	created, err := config.EnsureDefault(*cfgPath)
 	if err != nil {
 		log.Fatalf("[llmsproxy] config: %v", err)
@@ -58,19 +73,26 @@ func main() {
 		}
 	}
 
-	gw, err := gateway.New(c, c.GatewayKeys())
+	gw, err := gateway.New(c)
 	if err != nil {
 		log.Fatalf("[llmsproxy] gateway: %v", err)
 	}
 
 	// Ops hygiene: surface the two most common footguns instead of silently
 	// running with them.
-	if keys := c.GatewayKeys(); len(keys) == 0 {
-		log.Printf("[llmsproxy] WARNING: gateway_keys is EMPTY — without a key every request is rejected")
+	//
+	// Read the AUTHORITATIVE source. Auth uses core.ListKeys() (cfg.Keys), not
+	// the legacy cfg.GatewayKeys list: seedKeys copies gateway_keys into keys[]
+	// on first start and the YAML list is ignored for auth afterwards. Checking
+	// GatewayKeys() here made the warning lie — once the legacy list was empty
+	// (e.g. after rotating the starter key away) it reported "every request is
+	// rejected" while seven working keys, one of them admin, were in service.
+	if keys := c.ListKeys(); len(keys) == 0 {
+		log.Printf("[llmsproxy] WARNING: no gateway key configured — every request will be rejected. Add one in the WebUI or set gateway_keys")
 	} else {
 		for _, k := range keys {
-			if k == "sk-gw-local-0001" || k == "sk-local-0001" {
-				log.Printf("[llmsproxy] WARNING: gateway key %q looks like the starter/example key — rotate it before exposing the gateway", k)
+			if k.Key == "sk-gw-local-0001" || k.Key == "sk-local-0001" {
+				log.Printf("[llmsproxy] WARNING: gateway key %q looks like the starter/example key — rotate it before exposing the gateway", k.Key)
 			}
 		}
 	}

deploy.sh

@@ -244,6 +244,117 @@ verify_config() {
     return 1
 }
 
+# ---------- 密钥可解封校验 ----------
+# config.yaml 里的凭据以 enc:v1: 密文落盘，解密钥匙是 master.key（或
+# LLMS_PROXY_MASTER_KEY）。一旦钥匙丢失/错位，配置在**启动时**才会报错，
+# 那时服务已经处于「二进制已替换 + 配置已替换」的半死状态，只能靠回滚。
+# 所以在替换任何东西之前先验：能不能用现有的钥匙解开现有的密文。
+#
+# 两种情形要分开：
+#   1. 配置里**有**密文 ⇒ 钥匙必须能解开（否则拒绝部署）
+#   2. 配置里**没有**密文 ⇒ 全新安装/首次加密，无需钥匙（但会顺手记一条，
+#      因为它意味着这次部署会把明文写成密文，而钥匙必须同时就位）
+verify_master_key() {
+    log "密钥可解封校验"
+    [[ -f "$TARGET_CONFIG" ]] || return 0   # 配置都不在，交给 verify_config 报错
+
+    local sealed
+    sealed=$(grep -c 'enc:v1:' "$TARGET_CONFIG" 2>/dev/null || echo 0)
+    if [[ "$sealed" -eq 0 ]]; then
+        warn "配置内无密文（尚未加密或全为明文）——首次加密需要 master.key 可写"
+        return 0
+    fi
+
+    # 钥匙来源与 config 包一致：env 优先，否则 runtime_file 同目录的 master.key
+    local runtime_file key_file
+    runtime_file=$(awk '/^runtime_file:/{print $2; exit}' "$TARGET_CONFIG")
+    [[ -n "$runtime_file" ]] || runtime_file="$(dirname "$TARGET_CONFIG")/runtime.json"
+    key_file="$(dirname "$runtime_file")/master.key"
+
+    if [[ -n "${LLMS_PROXY_MASTER_KEY:-}" ]]; then
+        log "  钥匙来源: LLMS_PROXY_MASTER_KEY（环境变量）"
+        # 走真实解密路径验证，而不是只检查格式
+        if out=$("$TARGET_BIN" -show-secrets -config "$TARGET_CONFIG" 2>&1); then
+            log "✓ 凭据可解封（$(echo "$out" | grep -cE '^(source|key) ') 项）"
+            return 0
+        fi
+        err "凭据无法用 LLMS_PROXY_MASTER_KEY 解封，未替换任何文件："
+        printf '%s\n' "$out" >&2
+        return 1
+    fi
+
+    if [[ ! -f "$key_file" ]]; then
+        err "配置含 $sealed 处密文，但钥匙文件不存在: $key_file"
+        err "  没有它，部署后服务将无法启动。恢复 master.key 或设置 LLMS_PROXY_MASTER_KEY 后重试。"
+        return 1
+    fi
+    log "  钥匙文件: $key_file ($(stat -c '%A' "$key_file" 2>/dev/null || echo '?')，$(wc -c < "$key_file") 字节)"
+
+    if out=$("$TARGET_BIN" -show-secrets -config "$TARGET_CONFIG" 2>&1); then
+        log "✓ 凭据可解封（$(echo "$out" | grep -cE '^(source|key) ') 项）"
+        return 0
+    fi
+    err "凭据无法解密（钥匙与配置不匹配），未替换任何文件："
+    printf '%s\n' "$out" | head -5 >&2
+    err "  密钥来源: $key_file"
+    err "  常见原因：master.key 被重新生成、恢复了不匹配的备份，或换了机器。"
+    return 1
+}
+
+# ---------- 适配器备份保留策略 ----------
+# 每次部署都新建一个 adapters.bak.<时间戳> 目录，而回滚只用到最近一次
+# （$BACKUP_BIN / $BACKUP_CONFIG 都是单文件覆盖）。多出来的目录从没人清理，
+# 实测线上已积累 71 个，/etc/llmsproxy 因此涨到 122M。
+#
+# 保留最近 KEEP_ADAPTER_BACKUPS 份足够回滚，同时给目录数设上限——否则
+# 一次误配置（比如 adapter_dir 指错）就可能在几秒内造出成百上千个目录。
+# 只删名字严格匹配 adapters.bak.<14位时间戳> 的目录，避免误伤人工放的目录。
+KEEP_ADAPTER_BACKUPS=5
+
+prune_adapter_backups() {
+    local base="/etc/llmsproxy"
+    [[ -d "$base" ]] || return 0
+
+    # 先按数量上限硬裁：即使时间戳排序失效也不会无上限增长。
+    local all
+    mapfile -t all < <(find "$base" -maxdepth 1 -type d -name 'adapters.bak.*' -printf '%f\n' | sort)
+    local cap=$((KEEP_ADAPTER_BACKUPS * 4))
+    if (( ${#all[@]} > cap )); then
+        warn "适配器备份目录有 ${#all[@]} 个（异常），裁到 $cap"
+        local i=0
+        for d in "${all[@]}"; do
+            i=$((i + 1))
+            # 从最旧的开始删（sort 后升序）。名字不规范的跳过不删。
+            if (( i <= ${#all[@]} - cap )); then
+                if [[ "$d" =~ ^adapters\.bak\.[0-9]{14}$ ]]; then
+                    rm -rf "${base:?}/$d"
+                else
+                    warn "跳过名字不规范的备份目录（不删）: $d"
+                fi
+            fi
+        done
+        mapfile -t all < <(find "$base" -maxdepth 1 -type d -name 'adapters.bak.*' -printf '%f\n' | sort)
+    fi
+
+    # 再按时间保留最近 KEEP_ADAPTER_BACKUPS 份（sort 后最新在末尾）。
+    local total=${#all[@]} i=0
+    for d in "${all[@]}"; do
+        i=$((i + 1))
+        # i <= total-KEEP 的是较旧的，要删。
+        if (( i <= total - KEEP_ADAPTER_BACKUPS )); then
+            if [[ "$d" =~ ^adapters\.bak\.[0-9]{14}$ ]]; then
+                rm -rf "${base:?}/$d"
+            else
+                warn "跳过名字不规范的备份目录（不删）: $d"
+            fi
+        fi
+    done
+
+    local left
+    left=$(find "$base" -maxdepth 1 -type d -name 'adapters.bak.*' | wc -l)
+    log "  适配器备份保留最近 $KEEP_ADAPTER_BACKUPS 份（当前剩 $left 个）"
+}
+
 # ---------- 同步适配器 ----------
 sync_adapters() {
     log "同步适配器"
@@ -261,6 +372,7 @@ sync_adapters() {
         cp -rf "$TARGET_ADAPTERS/"*.lua "$BACKUP_DIR/" 2>/dev/null || true
         log "  旧适配器已备份到 $BACKUP_DIR"
     fi
+    prune_adapter_backups
 
     cp -f "$SRC_ADAPTERS/"*.lua "$TARGET_ADAPTERS/"
     chmod 0644 "$TARGET_ADAPTERS/"*.lua
@@ -393,6 +505,10 @@ main() {
     parse_args "$@"
     log "═══ llmsproxy 原子部署开始 ═══"
     precheck
+    # 密钥校验放在 build/替换之前：它的失败后果最重（配置里的密文永久解不开），
+    # 必须在动任何文件之前就挡住。此阶段用的是**当前线上二进制**，它已经
+    # 支持 -show-secrets（首次引入该子命令的版本除外，此时跳过并提示）。
+    verify_master_key
     build
     sync_adapters
     backup_old_state

docs/git-workflow-en.md

@@ -15,7 +15,7 @@ deployable.
 |---|---|---|---|
 | Main | `main` | permanent | Only long-lived branch. Always deployable. Accumulates the next version. |
 | Feature | `feature/<desc>` | short (dev → merge → delete) | New features / ordinary fixes. Born from `main`, merged back into `main`. |
-| Release | `release/vX.Y.Z` | one version cycle | Cut from `main`, tagged for release. Version-specific hotfixes land here. |
+| Release | `release/vX.Y.x` | one version cycle | Cut from `main`, tagged for release. Version-specific hotfixes land here. |
 
 ## Change flow (important)
 
@@ -26,15 +26,21 @@ deployable.
                            │                        │
                            │ cut                    │ cut
                            ▼                        ▼
-                    release/v1.4.2            release/v1.4.3
+                    release/v1.6.x            release/v1.7.x
                            │                        │
-                    tag: v1.4.2               tag: v1.4.3
+                      tag: v1.6.0               tag: v1.7.6
                            │                        │
               hotfix ◄─────┘                hotfix ◄─────┘
                    │                              │
                    └── cherry-pick back ──────────┘
 ```
 
+The patch position in the branch name is a literal `x`, while the tag carries
+the concrete version: one `release/v1.7.x` can hold tags v1.7.0 … v1.7.6.
+Spanning several patches on a single minor branch is deliberate — patches are
+revisions of the same feature batch, hotfixes land on one branch, and back-port
+to main never has to resolve dependencies between several release branches.
+
 ### Key rules
 
 1. **main is always deployable**: never leave half-done work on `main`.
@@ -43,9 +49,9 @@ deployable.
    then `git merge --no-ff feature/xxx` (or squash) when done.
 3. **Release = cut a release branch from main + tag**:
    ```bash
-   git checkout -b release/v1.4.2 main
-   git tag -a v1.4.2 -m "ModelRouter v1.4.2"
-   git push origin release/v1.4.2 v1.4.2
+   git checkout -b release/v1.7.x main
+   git tag -a v1.7.0 -m "ModelRouter v1.7.0"
+   git push origin release/v1.7.x v1.7.0
    ```
    Build installers and upload the GitCode Release from this tag so the
    published state is exactly reproducible.
@@ -54,7 +60,7 @@ deployable.
    an already-released branch (unless you deliberately ship a minor revision).
 5. **Hotfixes MUST flow back to main**:
    ```bash
-   git checkout release/v1.4.2     # fix in the release branch
+   git checkout release/v1.7.x     # fix in the release branch
    git commit -m "fix: ..."
    git checkout main
    git cherry-pick <hotfix-commit> # and into main
@@ -67,12 +73,23 @@ deployable.
 When the next version ships, the previous release branch retires:
 
 - **Default: delete the remote release branch**
-  (`git push origin :release/v1.4.2`). All hotfixes were already
+  (`git push origin :release/v1.7.x`). All hotfixes were already
   cherry-picked into main, so main contains everything; no merge needed.
 - **Long-term maintenance** (e.g. an enterprise client pinned to an old
   version): keep the branch, accept only security fixes, keep the
   commit-then-cherry-pick loop.
 
+> **Where practice diverged from this section (checked 2026-10-01)**:
+> `release/v1.4.x` and `release/v1.5.x` still exist locally and on the remote,
+> so "retire the previous branch when the next version ships" was never
+> carried out. Keeping them is harmless (hotfixes were back-ported), but it
+> contradicts the rule above and makes a reader wonder whether they should be
+> there at all. **Feature branches, by contrast, are cleaned up**:
+> `feature/key-quota-control`, `feature/toolcall-id-sanitize`,
+> `feature/anthropic-usage-cache` and `feature/agentrouter-id-sanitize` were
+> deleted on 2026-10-01 after confirming with a per-commit `git patch-id`
+> comparison that their work had already landed in main.
+
 ## Explicit non-goals
 
 - **Never rebase main**: main's history stays append-only; anyone pulling gets
@@ -134,8 +151,8 @@ pain points:
 2. No feature branches meant two independent efforts could not proceed in
    parallel without colliding.
 
-With release branches: the published state = `release/vX.Y.Z` branch +
-`vX.Y.Z` tag, exactly reproducible; hotfixes have a clear landing spot; main
-stays "latest + all fixes + deployable".
+With release branches: the published state = `release/vX.Y.x` branch +
+the concrete `vX.Y.Z` tag, exactly reproducible; hotfixes have a clear landing
+spot; main stays "latest + all fixes + deployable".
 
 > 中文版见 [docs/git-workflow.md](git-workflow.md)。

docs/git-workflow.md

@@ -13,7 +13,7 @@ ModelRouter 采用 **GitHub Flow + 发布分支** 模型：`main` 是唯一长
 |---|---|---|---|
 | 主分支 | `main` | 永久 | 唯一长命分支。永远可部署。积攒下一个版本的功能。 |
 | 特性分支 | `feature/<描述>` | 短命（开发→合并即删） | 新特性 / 一般 bug 修复。从 `main` 开出，完成后合回 `main`。 |
-| 发布分支 | `release/vX.Y.Z` | 一个版本周期 | 从 `main` 分出，打 tag 发布。该版本生命周期内的 hotfix 都提交在此分支。 |
+| 发布分支 | `release/vX.Y.x` | 一个版本周期 | 从 `main` 分出，打 tag 发布。该版本生命周期内的 hotfix 都提交在此分支。 |
 
 ## 变更流向（重要）
 
@@ -24,15 +24,20 @@ ModelRouter 采用 **GitHub Flow + 发布分支** 模型：`main` 是唯一长
                            │                        │
                            │ 切出                   │ 切出
                            ▼                        ▼
-                    release/v1.4.2            release/v1.4.3
+                     release/v1.6.x            release/v1.7.x
                            │                        │
-                    tag: v1.4.2               tag: v1.4.3
+                      tag: v1.6.0               tag: v1.7.6
                            │                        │
               hotfix ◄─────┘                hotfix ◄─────┘
                    │                              │
                    └── cherry-pick 回 main ───────┘
 ```
 
+分支名里 patch 位是**字面的 x**，而 tag 打具体版本号：`release/v1.7.x` 这一条
+发布分支上的 tag 可以有 v1.7.0 … v1.7.6 多个。一条 minor 分支跨多个 patch 是
+刻意的：patch 是同一批功能的不同修订，hotfix 落在同一条分支上，回流 main 时
+也不必处理多条 release 分支之间的依赖。
+
 ### 关键规则
 
 1. **main 永远可部署**：不在 main 上留半成品。任何未完成的工作必须在特性分支上。
@@ -40,16 +45,16 @@ ModelRouter 采用 **GitHub Flow + 发布分支** 模型：`main` 是唯一长
    开发完 `git merge --no-ff feature/xxx` 或 squash 合回。
 3. **发布 = 从 main 切 release 分支 + 打 tag**：
    ```bash
-   git checkout -b release/v1.4.2 main
-   git tag -a v1.4.2 -m "ModelRouter v1.4.2"
-   git push origin release/v1.4.2 v1.4.2
+   git checkout -b release/v1.7.x main
+   git tag -a v1.7.0 -m "ModelRouter v1.7.0"
+   git push origin release/v1.7.x v1.7.0
    ```
    构建安装包、上传 GitCode Release 都基于这个 tag，保证可精确回溯发布态。
 4. **版本生命周期内只收该版本的 hotfix**：新特性一律并入 `main` 等下一个版本，
    绝不塞进已发布的 release 分支（除非主动选择在该版本内发次要版）。
 5. **hotfix 必须回流 main**：
    ```bash
-   git checkout release/v1.4.2     # 在发布分支提交修复
+   git checkout release/v1.7.x     # 在发布分支提交修复
    git commit -m "fix: ..."
    git checkout main
    git cherry-pick <hotfix-commit> # 回主分支
@@ -61,11 +66,20 @@ ModelRouter 采用 **GitHub Flow + 发布分支** 模型：`main` 是唯一长
 
 下一个版本发布时，上一个 release 分支退役：
 
-- **默认：直接删除远端 release 分支**（`git push origin :release/v1.4.2`）。
+- **默认：直接删除远端 release 分支**（`git push origin :release/v1.7.x`）。
   因为 hotfix 都已逐个 cherry-pick 回 main，main 已包含全部修复，无需再合并。
 - **如需要长期维护旧版**（例如企业大客户卡在旧版本）：保留分支，仅 stopship 接受
   该版本的安全修复，继续走「提交 + cherry-pick 回 main」循环。
 
+> **实践与本节的历史出入（2026-10-01 核对）**：`release/v1.4.x` 与 `release/v1.5.x`
+> 至今仍在本地与远端，说明"下一个版本发布就删上一个分支"实际没有执行。
+> 保留无害（hotfix 已回流），但它与上面写的规则不一致，读文档的人会以为
+> 这些分支不该存在。**特性分支则确实在清理**：`feature/key-quota-control`、
+> `feature/toolcall-id-sanitize`、`feature/anthropic-usage-cache`、
+> `feature/agentrouter-id-sanitize` 四个分支在 2026-10-01 删除——它们的工作
+> 早已全部进入 main（逐提交用 `git patch-id` 比对确认），留着只是给下个版本
+> 制造 cherry-pick/merge 陷阱。
+
 ## 明确不做的事
 
 - **不 rebase main**：`main` 的历史保持追加式，任何人拉取后 `git pull` 都得到直接可用的历史。

docs/plugins.md

@@ -0,0 +1,507 @@
+# 插件系统（Plugin System）
+
+> **English**: this document is the reference for writing ModelRouter plugins.
+> The Chinese version is the primary one; section titles map 1:1.
+
+ModelRouter 的插件是**单个 `.lua` 文件**，放在 `config.yaml` 的 `plugin_dir` 目录里。
+插件能做两件事：
+
+1. **挂钩子**：在请求流水线的若干 stage 上注册回调，看到每个请求的完整信息，
+   并可以把结果累加进自己的状态。
+2. **贡献界面**：在启动时返回 HTML / CSS / JS，由内核注入 WebUI——可以是一整个
+   新页面，也可以是往现有页面里追加一个组件。
+
+两者互相独立：只想统计请求数的插件不必碰界面；只想加个仪表盘的插件不必碰钩子。
+
+---
+
+## 1. 快速上手
+
+一个最小的插件：
+
+```lua
+-- plugins/hello.lua
+local plugin = {
+  name = "hello",
+  version = "1.0.0",
+  description = "示例插件",
+  author = "you",
+}
+
+-- 声明钩子
+plugin.hooks = {
+  request_end = "on_request_end",
+}
+
+-- 钩子实现
+function plugin.on_request_end(payload)
+  -- payload 是解码后的 table，不是 JSON 字符串
+  log("info", string.format("%s via %s: %d prompt tokens",
+    payload.model, payload.source, payload.prompt_tokens or 0))
+  return nil  -- 最后一个 stage 没有下游，return 无意义
+end
+
+-- 贡献界面
+plugin.ui = {
+  page = {
+    page_id = "hello",     -- kebab-case
+    title = "Hello",
+    icon = "👋",
+    order = 90,            -- 侧栏排序
+    mount = [[<div id="hello">hello</div>]],
+  },
+}
+
+return plugin   -- 必须返回一个 table
+```
+
+放进 `plugin_dir` 后重启即生效。`GET /api/plugins` 确认它被加载了。
+
+---
+
+## 2. 加载与生命周期
+
+```
+core.New
+  └─ lua.NewVM(adapter_dir).Start()        适配器状态
+  └─ lua.NewPlugins(vm, plugin_dir)
+       ├─ SeedBundled()                    仅当目录不存在时写入内置插件（目前是 billing）
+       └─ LoadDir()                        按文件名字典序逐个加载
+```
+
+**加载失败不影响网关启动。** 一个语法错误的插件会被记录在 `GET /api/plugins`
+的 `error` 字段里，永远不会被调用。这与适配器一致，但理由更强：插件是可选的
+第三方扩展，因为一个 `.lua` 打错字就让网关起不来是错误的取舍。
+
+**目录一旦存在就是权威的。** 与适配器同规则：首启会 seed 内置插件，之后目录里
+的文件说了算，删除或编辑内置插件都是真实生效的操作。
+
+### 2.1 热更新
+
+| 方式 | 效果 |
+|---|---|
+| `POST /api/plugins {name, code}` | 写文件 + 立即加载新版本（旧的 Lua 状态被关闭重建，**累计量清零**） |
+| `DELETE /api/plugins/{name}` | 删文件 + 卸载 |
+| 改文件后 `POST` 同名 | 同上 |
+
+改文件但**不** POST，需要重启才生效。
+
+---
+
+## 3. 流水线 stage
+
+一个请求依次经过三个 stage。插件可以为任意 stage 注册钩子；未注册的 stage
+被忽略，所以插件不会因为网关将来新增 stage 而报错。
+
+```
+        客户端请求
+             │
+   ┌─────────▼──────────┐
+   │  request_start     │  已解析、已鉴权，尚未选源
+   │  · type            │  "chat" | "stream" | "image"
+   │  · model           │  客户端请求的原始 model（"AUTO" 也在这里）
+   │  · key / role      │  掩码后的网关 key id（"***a1b2c3"）与角色
+   │  · source          │  空（还没选源）
+   │  · stream          │
+   │  · messages_count  │
+   │  · tools_count     │
+   │  · ts              │  unix 秒
+   └─────────┬──────────┘
+             │  （调度：tier 遍历 → 槽位轮转 → 冷却/配额过滤）
+   ┌─────────▼──────────┐
+   │  routed            │  已选定 (source, model)，尚未发往上游
+   │  · source / model  │  实际选中的
+   │  · tier            │  AUTO 链的档位；直连 = -1；AUTO = -2
+   │  · stream / key    │
+   └─────────┬──────────┘
+             │  （HTTP 往返 / SSE 流）
+   ┌─────────▼──────────┐
+   │  request_end       │  每个请求恰好一次，成功失败都触发
+   │  · ok / status     │
+   │  · latency_ms      │
+   │  · first_byte_ms   │  流式的首字节时间
+   │  · prompt_tokens   │  上游真实 usage，缺失时为字节估算
+   │  · completion_tokens
+   │  · cache_hit_tokens / cache_miss_tokens
+   │  · image_count     │  生图数量（图片不计 token）
+   │  · error           │  失败原因，成功时为 ""
+   │  · time            │  unix **毫秒**
+   └─────────┬──────────┘
+             │
+        写审计 + 聚合统计
+```
+
+### 3.1 触发点在哪
+
+| stage | 代码位置 | 说明 |
+|---|---|---|
+| `request_start` | `gateway/chat.go` `handleChat` / `fireImageStart` | 每个请求一次（chat 与生图各一条），在配额闸门**之前** |
+| `chain_step` | `gateway/chat.go` `chainTraceSink` | **仅 AUTO 路径**，每步一次 |
+| `routed` | `singleChat` / `streamChat` / `singleChatAuto` / `streamChatAuto` | 成功选定源之后，各一次 |
+| `request_end` | `gateway/chat.go` `writeRec` | **所有出口的唯一汇合点**，每个请求一次 |
+
+### 3.1b 为什么单独有 `chain_step`
+
+`routed` 只在**遍历结束后**触发一次，只带最终胜出的槽位。所以
+"tier 1 冷却所以降级到 tier 3"和"tier 1 正常接单"在它眼里**完全一样**——
+而这恰恰是优先级链存在的全部理由。
+
+`chain_step` 补上这条信息，四种 `kind`：
+
+| kind | 含义 | 何时产生 |
+|---|---|---|
+| `tier_skip` | 整档被跳过 | 该档所有槽位冷却中/配额用尽 |
+| `slot_fail` | 某个槽位硬失败 | 上游报错 / 适配器输出不可用 |
+| `tier_busy` | 整档全忙且有界等待超时 | 2s 内没等到空位 |
+| `selected` | 这个槽位接了单 | 每次成功遍历**恰好一次**，且是最后一步 |
+
+顺序保证：所有 `chain_step` 都在 `routed` 之前，`selected` 是最后一步。
+所以只订阅 `request_end` 的插件也能拿到轨迹摘要（见下）。
+
+### 3.1c `request_end` 里的轨迹摘要
+
+除了逐个 `chain_step`，`request_end` 还带三个便于做报表的字段：
+
+| 字段 | 含义 |
+|---|---|
+| `chain_walk` | 整个遍历的步骤数组（上限 12 步，超出截断） |
+| `degraded` | 布尔。`true` = 有过跳过/失败，即**发生了降级** |
+| `tier_served` | 实际服务的那一档；直连或全失败时为 `-1` |
+
+> **计费口径**：按**实际服务的模型**计费。降级到 tier 3 仍按 tier 3 的价算，
+> `chain_step` / `degraded` / `tier_served` 只作**观测**，不参与计价。
+> 理由见 §7.5。
+
+`request_end` 放在 `writeRec` 是因为四条入口路径（直连/AUTO × 流式/非流式）都
+经过它，既不会漏（流式的 token 数只有流结束才知道），也不会重复。
+
+**hot path 注意事项**：没有插件注册某 stage 时，`Fire` 立刻返回（一次 `RLock`
+加一次 map 查找）。装了插件之后，每个请求会在该 stage 上多一次 Lua 调用——
+这是同步的，在关键路径上。计费插件那种"每请求一次"是正常的；把重活放进钩子是
+反模式。
+
+### 3.2 钩子的返回值
+
+- 返回 `nil` 或不返回 = **没有意见**，payload 原样传给下一个插件
+- 返回 table = 其中的键会**合并进 payload**，并作为 `Fire` 的返回值
+
+**同一 stage 的多个插件并行执行**，但返回值按**插件加载顺序**合并，所以结果是
+确定的（不依赖 goroutine 调度）。代价是一个插件看不到另一个插件刚加的字段：
+每个钩子拿到的是**同一份 payload 快照**。
+
+这与早期版本不同 —— 早期是顺序执行，后一个插件能看到前一个的返回值。它从未被
+实际依赖（随核心发布的 billing 在每个 stage 都 `return nil`，注释里写着
+"nobody downstream would read a return value"），但这是一处**契约变化**：如果你的
+插件依赖「读到前一个插件写的字段」，并行的两个插件之间必须改用外部通信
+（例如各自写 `plugin.state`，由 `/api/plugins/<name>/state` 读取）。
+
+单个插件时不启 goroutine，直接调用。
+
+前三个 stage 的返回值目前没有内部消费者（最后一个 stage 之后就是写审计），
+所以计费插件改用 `plugin.state` + `/state` 端点来暴露数据。
+
+---
+
+## 4. 界面扩展
+
+### 4.1 整页
+
+```lua
+plugin.ui = {
+  page = {
+    page_id = "billing",   -- 必填,kebab-case。侧栏 data-tab 与 #tab-billing
+    title = "Billing",     -- 必填,侧栏文字
+    icon = "💰",           -- 可选
+    order = 40,            -- 侧栏排序,默认 100
+    mount = [[...HTML...]],
+  },
+}
+```
+
+### 4.2 往现有页面追加元素
+
+```lua
+plugin.ui = {
+  elements = {
+    {
+      target = "status",   -- status | chat | keys | sort | sources | adapters
+      anchor = "top",      -- "top" | "bottom" | "before:<sel>" | "after:<sel>"
+      order = 5,
+      mount = [[...HTML...]],
+    },
+  },
+}
+```
+
+### 4.3 `mount` 里可以带 `<script>` 和 `<style>`
+
+内核的注入顺序是：**先插 HTML，再执行 `<script>`**，所以脚本里访问
+`document.getElementById` 一定能拿到已渲染的节点。
+
+### 4.4 插件可用的浏览器端 API
+
+| API | 作用 |
+|---|---|
+| `window.pluginAPI.fetchState(name)` | 等价于 `GET /api/plugins/<name>/state` |
+| `window.pluginAPI.onTabShown(fn)` | 注册"页面切到可见时"的回调（轮询类组件用） |
+| `window.pluginAPI.postState(name, obj)` | `PUT` 自己的 state（写操作，需 admin） |
+
+### 4.5 一个完整例子
+
+见 `internal/lua/plugins/billing.lua`——它同时用了两种 UI 形式、完整的价格配置、
+以及全部三个 stage。
+
+---
+
+## 5. 插件状态与 HTTP API
+
+### 5.1 端点
+
+```
+GET    /api/plugins                  已加载插件清单 + 钩子 + UI + 错误
+POST   /api/plugins                  上传/替换（admin）{name, code}
+DELETE /api/plugins/{name}           删除（admin）
+GET    /api/plugins/{name}/state     读取插件自己发布的状态
+PUT    /api/plugins/{name}/state     替换状态（admin）
+```
+
+`GET .../state` 对**任意角色开放**：它是报表数据（开销、计数），用户自己的
+计费组件要能渲染。而 `PUT` 需要 admin。
+
+### 5.2 约定：`prices` 与 `state` 分离
+
+`PUT .../state` 的载荷里如果有 `prices` 键，它会被写进插件的 **`plugin.prices`**
+字段并从 `state` 里剔除。
+
+**为什么**：`state` 是累计量（历史），`prices` 是配置。一次改价如果整体替换
+`state`，累计量就没了——账目会在改价那一刻清零。分离之后：
+
+- 只带 `prices` → 改配置，**不动** `state`（累计量保留）
+- 带其它键 → 替换 `state`（这是显式的重置）
+
+> 插件可以不遵守这个约定，把整个载荷放进 `state` 自行处理。约定只为内置的
+> 计费插件存在。
+
+### 5.3 `GET /api/plugins` 的响应
+
+```json
+{
+  "plugins": [{
+    "name": "billing", "version": "1.0.0",
+    "description": "...", "author": "...",
+    "hooks": ["request_end"],
+    "ui": { "page": "billing", "elements": 1 },
+    "loaded": true
+  }],
+  "hook_errors": { "request_end": { "count": 3, "last_error": "billing: ..." } },
+  "plugin_dir": "/etc/llmsproxy/plugins"
+}
+```
+
+`hook_errors` 是**排错入口**：一个坏掉的插件表现为"功能缺失"而不是报错，
+这个计数让它可见。
+
+---
+
+## 6. 运行时约束（重要）
+
+### 6.1 一个插件 = 一个 Lua 状态
+
+适配器可以有多个独立的 worker 状态（它们无状态，这是对的）。**插件不行**：
+钩子通常往 `plugin.state` 里累加，多个状态就意味着总额被劈成几份，而写价格只
+写进其中一个状态，钩子恰好跑到另一个时**所有请求按 0 计费**。
+
+因此插件持有**唯一一个**状态，所有入口（`Fire` / `State` / `SetState`）用互斥锁
+串行化。代价：**一个卡住的钩子会卡住所有插件的钩子**。所以——
+
+> **钩子必须短、同步、不阻塞。** 不要在里面做 HTTP 请求、睡眠或重计算。
+
+### 6.2 异常被隔离
+
+钩子里 `error()` 不会影响转发：异常被捕获、记进 `hook_errors`、跳到下一个插件。
+适配器变换出错会让源进入冷却，但插件出错**不产生任何惩罚**——插件是可选功能。
+
+### 6.3 内置可用函数
+
+与适配器相同：`json.encode` / `json.decode`、`log(level, msg)`、
+`hmac_sha256_hex` / `sha256_hex` / `base64_encode` / `tohex`。
+
+⚠️ `os.date` 在极简运行时里可能不可用（计费插件因此有降级路径）。
+
+### 6.4 命名
+
+- 全局 `__llmsproxy_plugin` 存放插件返回的 table，**不要占用**
+- 插件间状态完全隔离，一个插件改不了另一个的全局
+
+---
+
+## 7. 内置计费插件
+
+`plugins/billing.lua`，默认随首启 seed 进来，开箱可用。
+
+### 7.1 计费维度
+
+| 维度 | 用途 | 优先级 |
+|---|---|---|
+| `sources` | 源的固定价（如按请求计费的转发服务） | 中 |
+| `models` | 模型的 per-token 价 | 高 |
+| `keys` | 单个网关 key 的覆盖价 | **最高** |
+
+**token 价**优先级：`keys` > `models` > `default`。
+**`per_request` 固定价是叠加的**（不覆盖），所以一个生图模型可以既算 token 又收固定费。
+**`cache_discount`** 见 §7.7。
+
+### 7.2 价格单位
+
+**USD / 单个 token**。这是各家 provider 的公布口径，所以典型数值长这样：
+`1.25e-6`。插件内部乘以 token 数，全程不做单位换算。
+
+```json
+{
+  "prices": {
+    "currency": "USD",
+    "default":  { "prompt": 0, "completion": 0, "per_request": 0 },
+    "sources":  { "trae": { "per_request": 0.01 } },
+    "models":   {
+      "gpt-5.4": { "prompt": 1.25e-6, "completion": 1e-5 },
+      "kolors":  { "per_request": 0.04 }
+    },
+    "keys":     { "***a1b2c3": { "prompt": 1.1e-6, "completion": 9e-6 } }
+  }
+}
+```
+
+配置价目：
+
+```bash
+curl -X PUT http://127.0.0.1:8080/api/plugins/billing/state \
+  -H "Authorization: Bearer $ADMIN_KEY" -H "Content-Type: application/json" \
+  -d '{"prices":{"models":{"gpt-5.4":{"prompt":1.25e-6,"completion":1e-5}}}}'
+```
+
+读回账目：
+
+```bash
+curl -H "Authorization: Bearer $ADMIN_KEY" \
+  http://127.0.0.1:8080/api/plugins/billing/state
+```
+
+### 7.3 累计维度
+
+`total` / `by_source` / `by_model` / `by_key` / `by_day`（`YYYY-MM-DD` UTC），
+每项含 `cost`、`requests`、`prompt_tokens`、`completion_tokens`、`failures`。
+
+另有两个**降级观测**维度（来自 `chain_step`）：
+
+| 字段 | 含义 |
+|---|---|
+| `degraded_reqs` | 发生过降级的请求数 |
+| `by_tier_served` | 各档实际接单数（`{"1": 812, "2": 37}`） |
+| `skip_reasons` | 跳过原因计数，等待时长已归一（`no free slot within <wait>`） |
+
+这三项是"网关是不是在悄悄降级"的核心指标：一个持续降级的网关，账单结构和健康
+网关看起来一模一样——**除非**单独统计降级次数。
+
+### 7.4 计费策略：失败的请求怎么算
+
+**保留 token 费用，丢弃固定费用。** 理由：上游在生成后才 500，token 确实被消耗
+了；但那个从未真正发生的固定费不该收。
+
+这是本插件里**最可争议的一条**。想改成"失败也收固定费"：
+
+```json
+{ "prices": { ... }, "count_failures": true }
+```
+
+### 7.5 ⚠️ 计费插件只报表，不执法
+
+**网关自己的配额会计（`internal/gateway/stats.go`，入口处强制）才是限额权威。**
+本插件不参与任何路由或配额决策。
+
+理由：两套独立的会计路径如果对不上，比一套功能略少的更糟。计费是**观察**，
+配额是**控制**，二者分开。
+
+### 7.6 未定价流量（重要）
+
+**任何维度都没配价的请求，成本记 0。** 这是最危险的失败模式：账单照样能加总，
+只是**悄悄少报**，而且没有任何报错。
+
+所以插件单独统计它们：
+
+| 字段 | 含义 |
+|---|---|
+| `unpriced_reqs` | 没有任何价目覆盖的请求数 |
+| `unpriced_models` | 按模型点名（`{"MYSTERY-MODEL": 12}`）——直接告诉你价目表缺了哪一行 |
+
+仪表盘上有 "Unpriced" 卡片。**这个数应该是 0**；不是 0 就去补价目。
+
+注意"未定价"不等于"免费"：这些请求的 `requests` / token 数**照常计入**
+`total` 与各维度，只有金额是 0。
+
+### 7.7 提示缓存计价
+
+**缓存命中的 prompt token 不按全价算。** 绝大多数 provider 对缓存读给很深的折扣
+（常见是 1/10），而 agent 流量会反复重放长前缀——正是缓存要让它便宜的那类流量。
+
+```
+fresh  = prompt_tokens - cache_hit_tokens   → 全价
+cached = cache_hit_tokens                   → 全价 × cache_discount
+```
+
+`cache_discount` 默认 **0.1**（10 倍折扣），因为 DeepSeek / Qwen / Kimi 等都是这个
+量级。它是**每个 provider 的事实、不是自然常数**，所以可以按条目覆盖：
+
+```json
+"models": { "gpt-5.4": { "prompt": 1.25e-6, "completion": 1e-5, "cache_discount": 0.25 } }
+```
+
+设成 `1` 恢复成旧的"prompt 一律全价"行为，设成 `0` 表示该 provider 不打折。
+
+优先级与 token 价一致（`keys` > `models` > `default`）。
+
+> **这一条改过行为。** 修复前缓存命中按全价算：100 万 prompt token 里 90 万是
+> 缓存命中，会算出 10 USD 而不是 ~1.9——**高估约 10 倍**，而且恰好发生在缓存
+> 最有价值的高频流量上。
+
+### 7.8 数据从哪来
+
+`request_end` 的 `prompt_tokens` / `completion_tokens` 优先取上游真实的
+`usage`；上游没报时网关用字节估算（`len/3+1`）。流式请求在流结束后用上游真实
+数字**覆盖**估算值。所以计费数字的精度取决于上游是否报 usage。
+
+---
+
+## 8. 排错
+
+| 现象 | 查什么 |
+|---|---|
+| 插件没出现在 `/api/plugins` | 目录对不对（响应里的 `plugin_dir`）；文件是不是 `.lua` |
+| `loaded: false` 且有 `error` | 语法错误或没 `return table`，错误信息在 `error` 字段 |
+| 功能"没反应"但无报错 | 查 `hook_errors`——坏钩子只记录不抛出 |
+| 钩子没被调用 | 该 stage 确实触发了吗（`routed` 只在**成功**选源后触发，调度全失败时不触发） |
+| 界面空白 | 内核的 `GET /api/ui-inject` 里有没有你的 `page_id`；脚本有没有报错（浏览器 console） |
+| `pluginAPI` 未定义 | 脚本在注入前执行了；确认用的是 `mount` 而不是别的方式插入 |
+
+### 调试钩子
+
+插件的 `log(level, msg)` 输出到网关的日志：
+
+```
+journalctl -u llmsproxy -f | grep -i plugin
+```
+
+---
+
+## 9. 与适配器的区别
+
+| | 适配器 | 插件 |
+|---|---|---|
+| 位置 | `adapter_dir` | `plugin_dir` |
+| 作用 | 上游协议转换 | 请求流水线 + 界面 |
+| 钩子 | `transform_request` / `transform_response` / `transform_stream_chunk` / `build_headers` / `transform_error` | `request_start` / `routed` / `request_end` |
+| 状态 | 每个源独立，多 worker | 单状态，见 §6.1 |
+| 出错后果 | 该 (源,模型) 进入冷却退避 | 仅该功能缺失 |
+| 鉴权 | 需要 `gateway_keys` 之外的独立凭据 | 网关 key |
+| 必需性 | 核心 | 可选，缺了网关照常跑 |
+
+完整适配器协议见 [lua-adapters.md](lua-adapters.md)。

e2e/e2e_test.go

@@ -326,3 +326,144 @@ func statusOf(resp *http.Response) int {
 	}
 	return resp.StatusCode
 }
+
+// The startup warning about key configuration must describe the key set that
+// AUTH actually uses (cfg.Keys, via core.ListKeys), not the legacy
+// gateway_keys YAML list.
+//
+// It used to check GatewayKeys() only. seedKeys copies gateway_keys into keys[]
+// on first start and the YAML list stops being consulted, so once that list was
+// emptied — e.g. after rotating the starter key away — the process logged
+// "every request will be rejected" while seven working keys, one of them admin,
+// were in service. Observed in production: the warning appeared and that same
+// key returned HTTP 200 on /v1/models.
+//
+// The fixture reproduces exactly that shape: gateway_keys EMPTY, with the
+// working key present only in keys[].
+func TestStartupWarningReflectsRealKeysNotLegacyList(t *testing.T) {
+	up := newMockUpstream(t)
+	dir := t.TempDir()
+
+	// Seed a real install once so keys[] gets populated from gateway_keys, then
+	// rewrite the file to look like a rotated install: legacy list cleared, the
+	// seeded key still present in keys[].
+	path := writeConfig(t, dir, "127.0.0.1:0", map[string]*mockUpstream{"good": up})
+	bin := buildBinary(t)
+
+	// First run creates the sealed config with keys[] populated.
+	seedPort := freePort(t)
+	cfg1 := writeConfig(t, dir, "127.0.0.1:"+seedPort, map[string]*mockUpstream{"good": up})
+	g1 := startGateway(t, bin, "127.0.0.1:"+seedPort, cfg1)
+	g1.stop()
+
+	data, err := os.ReadFile(cfg1)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(data), "keys:") {
+		t.Fatalf("first run did not populate keys[]:\n%s", data)
+	}
+
+	// Now clear the legacy gateway_keys list, keeping keys[] intact. The
+	// rewritten file may use different indentation, so match loosely.
+	cleared, n := clearGatewayKeys(string(data))
+	if n == 0 {
+		t.Fatalf("could not find a gateway_keys entry to clear in:\n%s", data)
+	}
+	if err := os.WriteFile(cfg1, []byte(cleared), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	// Start again and inspect the log. The config pins its listen address, so
+	// rewrite that field to the port we will actually poll.
+	port2 := freePort(t)
+	rehosted, ok := replaceListen(string(cleared), "127.0.0.1:"+port2)
+	if !ok {
+		t.Fatalf("could not rewrite listen in:\n%s", cleared)
+	}
+	cfg2 := filepath.Join(dir, "config2.yaml")
+	if err := os.WriteFile(cfg2, []byte(rehosted), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	g2 := startGateway(t, bin, "127.0.0.1:"+port2, cfg2)
+	defer g2.stop()
+
+	// Sanity: the key really does authenticate, so the warning would be a lie.
+	resp, _ := g2.do("GET", "/v1/models", "", true)
+	if resp == nil || resp.StatusCode != http.StatusOK {
+		t.Fatalf("key from keys[] did not authenticate (status %v); test fixture is wrong",
+			resp)
+	}
+
+	logs := g2.logString()
+	if strings.Contains(logs, "gateway_keys is EMPTY") {
+		t.Errorf("startup logged 'gateway_keys is EMPTY ... every request is rejected' while "+
+			"a working admin key was in service — the warning must read the authoritative "+
+			"key set (core.ListKeys), not the legacy YAML list.\nlog:\n%s", logs)
+	}
+	_ = path
+	_ = g1
+}
+
+func freePort(t *testing.T) string {
+	t.Helper()
+	l, err := net.Listen("tcp", "127.0.0.1:0")
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer l.Close()
+	_, port, _ := net.SplitHostPort(l.Addr().String())
+	return port
+}
+
+// stop terminates the gateway and waits, so its log buffer is complete.
+func (g *gatewayUnderTest) stop() {
+	if g == nil || g.cmd == nil || g.cmd.Process == nil {
+		return
+	}
+	_ = g.cmd.Process.Kill()
+	_, _ = g.cmd.Process.Wait()
+}
+
+// logString returns everything the gateway has written so far.
+func (g *gatewayUnderTest) logString() string { return g.log.String() }
+
+// clearGatewayKeys empties the gateway_keys list in a config, tolerating any
+// indentation the YAML writer chose. Returns the new text and how many entries
+// it removed.
+func clearGatewayKeys(src string) (string, int) {
+	lines := strings.Split(src, "\n")
+	var out []string
+	removed := 0
+	inList := false
+	for _, l := range lines {
+		trimmed := strings.TrimSpace(l)
+		if strings.HasPrefix(l, "gateway_keys:") {
+			out = append(out, "gateway_keys: []")
+			inList = true
+			removed++
+			continue
+		}
+		if inList {
+			if strings.HasPrefix(trimmed, "- ") {
+				removed++
+				continue
+			}
+			inList = false
+		}
+		out = append(out, l)
+	}
+	return strings.Join(out, "\n"), removed
+}
+
+// replaceListen rewrites the top-level listen: value.
+func replaceListen(src, addr string) (string, bool) {
+	lines := strings.Split(src, "\n")
+	for i, l := range lines {
+		if strings.HasPrefix(l, "listen:") {
+			lines[i] = "listen: " + addr
+			return strings.Join(lines, "\n"), true
+		}
+	}
+	return src, false
+}

internal/billing/compile.go

@@ -0,0 +1,323 @@
+package billing
+
+import (
+	"fmt"
+	"strings"
+
+	"llmsproxy/internal/config"
+)
+
+// Compile turns one billing profile into the prices table the billing plugin
+// expects, resolving URL rules against the gateway's actual sources.
+//
+// WHY URL RULES NEED RESOLVING AT ALL: the operator declares pricing by URL
+// because that is what a provider's price list is keyed on, and because several
+// sources can point at the same URL. The plugin, however, looks up by
+// `payload.source` (a source NAME) and by model — it has no idea what URL a
+// request went to. So the URL match happens here, at compile time, where the
+// config's name->base_url mapping is known, and the result is expressed in the
+// dimensions the plugin already supports.
+//
+// Profiles are the "let the user choose" axis: the same URL can appear in
+// several profiles and switching recomputes this table, so a gateway can be
+// repriced without editing the plugin.
+func Compile(profile *config.BillingProfile, sources []config.Source) (map[string]interface{}, error) {
+	return CompileOpts(profile, sources, false)
+}
+
+// CompileOpts is Compile with an explicit policy for rules that match no
+// source.
+//
+// The default is STRICT and that is correct for startup: a profile with a
+// typo'd URL prices nothing, every request on it is recorded as unpriced, and
+// the bill silently comes out wrong. Failing the load is the right response.
+//
+// The rules editor needs the opposite. An operator editing a half-finished
+// profile — adding a rule before the source that will use it exists, or
+// renaming a URL while typing — must be able to save and then see the warning,
+// not be blocked by an error they can only resolve by guessing. Lenient mode
+// keeps the offending rule in the config and drops it from the compiled table,
+// so it is preserved as text and clearly reported, without pretending it
+// prices anything.
+func CompileOpts(profile *config.BillingProfile, sources []config.Source, lenient bool) (map[string]interface{}, error) {
+	if profile == nil {
+		return nil, fmt.Errorf("no billing profile")
+	}
+	prices := map[string]interface{}{
+		"currency": "USD",
+		"default":  map[string]interface{}{"prompt": 0.0, "completion": 0.0, "per_request": 0.0},
+		"sources":  map[string]interface{}{},
+		"models":   map[string]interface{}{},
+		"keys":     map[string]interface{}{},
+	}
+	if profile.Default != "" {
+		prices["default_mode"] = profile.Default
+	}
+
+	// Currency is a profile-level statement; a rule may override it.
+	for i := range profile.Rules {
+		if c := profile.Rules[i].Currency; c != "" {
+			prices["currency"] = c
+			break
+		}
+	}
+
+	matched := map[string]bool{}
+	for i := range profile.Rules {
+		rule := &profile.Rules[i]
+		targets := matchSources(rule.URL, sources)
+		if len(targets) == 0 && rule.URL != "*" {
+			// A rule for a URL no source uses is almost always a typo (or a
+			// source that was removed). Failing loudly beats a profile that
+			// silently prices nothing — the whole reason this is a config file
+			// instead of a hand-written JSON blob.
+			if lenient {
+				continue // kept in the config text, reported as a warning
+			}
+			return nil, fmt.Errorf("rule url %q matches no configured source base_url", rule.URL)
+		}
+		for _, srcName := range targets {
+			if matched[srcName] {
+				// First rule wins. Two rules matching one source is ambiguous,
+				// and silently letting the later one win makes the file's
+				// meaning depend on ordering the operator cannot see.
+				continue
+			}
+			matched[srcName] = true
+			if err := applyRule(prices, rule, srcName, sources); err != nil {
+				return nil, err
+			}
+		}
+	}
+
+	// Sources no rule matched fall through to the profile default, which the
+	// plugin applies via prices.default. Recording them explicitly means the
+	// UI can say "this source is unpriced" instead of leaving the operator to
+	// infer it from a zero.
+	var unmatched []string
+	for _, src := range sources {
+		if !matched[src.Name] {
+			unmatched = append(unmatched, src.Name)
+		}
+	}
+	if len(unmatched) > 0 {
+		prices["unmatched_sources"] = unmatched
+	}
+	return prices, nil
+}
+
+// matchSources returns the source names whose base_url matches pattern.
+// "*" matches every source (used as a catch-all default rule).
+func matchSources(pattern string, sources []config.Source) []string {
+	var out []string
+	if pattern == "*" {
+		for _, s := range sources {
+			out = append(out, s.Name)
+		}
+		return out
+	}
+	want := normalizeURL(pattern)
+	for _, s := range sources {
+		if normalizeURL(s.BaseURL) == want {
+			out = append(out, s.Name)
+		}
+	}
+	return out
+}
+
+// normalizeURL compares URLs the way an operator expects: trailing slashes and
+// case in the host are differences the provider's price list does not care
+// about, and requiring an exact byte match would make the config brittle.
+func normalizeURL(u string) string {
+	u = strings.TrimSpace(u)
+	u = strings.TrimRight(u, "/")
+	return strings.ToLower(u)
+}
+
+// applyRule writes one rule's pricing for one source into the prices table.
+func applyRule(prices map[string]interface{}, rule *config.BillingRule, srcName string, sources []config.Source) error {
+	switch rule.Mode {
+	case "free":
+		// Explicitly priced at zero. This is NOT the same as unpriced: a source
+		// the operator says is free must not appear in the unpriced warnings,
+		// or those warnings become noise and stop being read.
+		prices["sources"].(map[string]interface{})[srcName] = map[string]interface{}{
+			"prompt": 0.0, "completion": 0.0, "per_request": 0.0,
+		}
+		return nil
+
+	case "subscription":
+		// A fixed monthly commitment: the per-request MARGINAL cost is zero, and
+		// the flat fee is reported separately. Spreading a monthly fee across
+		// requests would invent a per-request number the provider never charges,
+		// and it would change every time traffic did.
+		prices["sources"].(map[string]interface{})[srcName] = map[string]interface{}{
+			"prompt": 0.0, "completion": 0.0, "per_request": 0.0,
+		}
+		fixed, _ := prices["subscriptions"].(map[string]interface{})
+		if fixed == nil {
+			fixed = map[string]interface{}{}
+			prices["subscriptions"] = fixed
+		}
+		cur := "USD"
+		if rule.Currency != "" {
+			cur = rule.Currency
+		}
+		fixed[srcName] = map[string]interface{}{
+			"monthly": rule.Subscription, "currency": cur,
+		}
+		return nil
+
+	case "unpriced":
+		// Deliberately left out of `sources` so the plugin's unpriced_models /
+		// unpriced_reqs counters catch it. That is the point: a subscription
+		// plan whose credits cannot be converted to tokens must be VISIBLE as
+		// unbilled, not quietly estimated.
+		return nil
+
+	case "token":
+		if len(rule.Models) == 0 {
+			return fmt.Errorf("rule for url %q: mode token requires at least one model", rule.URL)
+		}
+		models := prices["models"].(map[string]interface{})
+		for modelID, t := range rule.Models {
+			entry := map[string]interface{}{}
+			prompt, err := usdPerM(t.Prompt)
+			if err != nil {
+				return fmt.Errorf("model %q prompt: %w", modelID, err)
+			}
+			completion, err := usdPerM(t.Completion)
+			if err != nil {
+				return fmt.Errorf("model %q completion: %w", modelID, err)
+			}
+			entry["prompt"] = prompt
+			entry["completion"] = completion
+			if t.CacheDiscount != nil {
+				entry["cache_discount"] = *t.CacheDiscount
+			}
+			// reasoning is optional and stays ABSENT (not nil, not 0) when the
+			// rule does not set it, which is what makes the billing plugin skip
+			// the repricing branch entirely. Writing an explicit 0 here would
+			// mean "thinking is free" and silently under-bill every request
+			// from a provider that bills it like ordinary output.
+			if strings.TrimSpace(t.Reasoning) != "" {
+				rp, err := usdPerM(t.Reasoning)
+				if err != nil {
+					return fmt.Errorf("model %q reasoning: %w", modelID, err)
+				}
+				entry["reasoning_price"] = rp
+			}
+			if lc := t.LongContext; lc != nil {
+				if lc.Above <= 0 {
+					return fmt.Errorf("model %q long_context: above must be > 0", modelID)
+				}
+				above, err := longContextRate(lc, prompt)
+				if err != nil {
+					return fmt.Errorf("model %q long_context: %w", modelID, err)
+				}
+				entry["long_context"] = map[string]interface{}{
+					"above": lc.Above,
+					// Marginal, not blended. The plugin charges the tokens past the
+					// threshold at this rate and the rest at the base rate, so the
+					// threshold composes with a cache discount (which applies to the
+					// cached portion) instead of forcing the plugin to guess how
+					// the two interact.
+					"prompt": above,
+				}
+			}
+			if rule.Peak != nil {
+				entry["peak"] = compilePeak(rule.Peak)
+			}
+			models[modelID] = entry
+		}
+		// A specific-URL token rule needs a source entry so a model served
+		// from THIS url but absent from `rule.Models` is still recognised as
+		// belonging to a priced provider.
+		//
+		// A WILDCARD rule ("*") must NOT get one. "*" is a model catalogue — it
+		// says "these model ids have these prices, wherever they come from" —
+		// and marking every source priced at {0,0} would make priceFor() set
+		// priced=true for every model the catalogue does NOT list. Paid
+		// sources (commandcode, alittokenplan, …) would then record $0 instead
+		// of showing up in unpriced_reqs, which is precisely the silent
+		// zero-billing failure this whole plugin is built to make visible.
+		// The model lookup p.models[payload.model] already marks a listed
+		// model priced on its own, so the wildcard loses nothing.
+		if rule.URL != "*" {
+			if _, ok := prices["sources"].(map[string]interface{})[srcName]; !ok {
+				prices["sources"].(map[string]interface{})[srcName] = map[string]interface{}{
+					"prompt": 0.0, "completion": 0.0, "per_request": 0.0,
+				}
+			}
+		}
+		return nil
+	}
+	return fmt.Errorf("unknown mode %q", rule.Mode)
+}
+
+// usdPerM converts a USD-per-million string into the per-token rate the plugin
+// expects. Delegates to config.ParseFloatUSDPerM so the DSL validator and the
+// compiler agree byte-for-byte on what a valid price is — a value that passes
+// Validate() but fails here (or vice versa) would be the worst kind of drift.
+func usdPerM(s string) (float64, error) {
+	return config.ParseFloatUSDPerM(s)
+}
+
+// compilePeak renders the peak window in the shape the plugin reads:
+// { multiplier, windows = { { days = {...}, hours = { {lo,hi}, ... } } } }.
+//
+// The plugin reads `days` / `hours` pairs. The first version of the deployed
+// price table used { start, end, weekdays } — a shape nothing reads — so peak
+// traffic was billed at off-peak rates with no error anywhere. Compiling from
+// typed config fields removes the chance of writing the wrong key names by hand.
+func compilePeak(p *config.BillingPeak) map[string]interface{} {
+	win := map[string]interface{}{}
+	if len(p.Weekdays) > 0 {
+		days := make([]interface{}, 0, len(p.Weekdays))
+		for _, d := range p.Weekdays {
+			days = append(days, d)
+		}
+		win["days"] = days
+	}
+	if len(p.Hours) > 0 {
+		hours := make([]interface{}, 0, len(p.Hours))
+		for _, h := range p.Hours {
+			hours = append(hours, []interface{}{h[0], h[1]})
+		}
+		win["hours"] = hours
+	}
+	out := map[string]interface{}{"multiplier": p.Multiplier}
+	if len(win) > 0 {
+		out["windows"] = []interface{}{win}
+	}
+	return out
+}
+
+// longContextRate resolves the marginal per-token rate for the tokens ABOVE a
+// threshold, given the model's base per-million prompt price.
+//
+// Two ways to express it are supported because providers publish both: an
+// absolute rate (OpenAI quotes \$3.00/M past 200k) and a multiple (DeepSeek
+// quotes exactly 2x past 128k). Neither may be missing: a step with no rate is a
+// silent no-op, and an operator who wrote one meant something.
+func longContextRate(lc *config.BillingLongContext, basePerToken float64) (float64, error) {
+	hasPrompt := strings.TrimSpace(lc.Prompt) != ""
+	hasMult := lc.Multiplier > 0
+	switch {
+	case hasPrompt && hasMult:
+		p, err := usdPerM(lc.Prompt)
+		if err != nil {
+			return 0, err
+		}
+		// Both given: the absolute rate wins, because it is the one the provider
+		// actually quotes. Silently multiplying instead would produce a number
+		// the operator never wrote.
+		return p, nil
+	case hasPrompt:
+		return usdPerM(lc.Prompt)
+	case hasMult:
+		return basePerToken * lc.Multiplier, nil
+	default:
+		return 0, fmt.Errorf("needs either prompt or multiplier")
+	}
+}

internal/billing/wildcard_test.go

@@ -0,0 +1,214 @@
+package billing
+
+import (
+	"testing"
+
+	"llmsproxy/internal/config"
+)
+
+// The wildcard token rule must not blanket-mark every source as priced.
+//
+// Why this is a hazard and not a style question: priceFor() sets priced=true as
+// soon as the request's SOURCE has an entry in the price table, and that flag
+// is what keeps a request OUT of unpriced_reqs. A "*" rule is a model
+// catalogue — it prices specific model ids wherever they come from — but
+// compiling it into a {0,0,0} entry on all 21 sources made every model the
+// catalogue does not list look "priced at zero". Paid providers (commandcode,
+// alittokenplan, …) then stopped appearing in the unpriced warnings, which is
+// exactly the silent zero-billing failure the plugin exists to surface.
+func TestWildcardTokenRuleDoesNotMarkEverySourcePriced(t *testing.T) {
+	sources := []config.Source{
+		{Name: "free1", BaseURL: "https://free1.example.com/v1"},
+		{Name: "paid1", BaseURL: "https://paid1.example.com/v1"},
+		{Name: "paid2", BaseURL: "https://paid2.example.com/v1"},
+	}
+	profile := &config.BillingProfile{
+		ID: "p",
+		Rules: []config.BillingRule{
+			{URL: "https://free1.example.com/v1", Mode: "free"},
+			{
+				URL: "*", Mode: "token",
+				Models: map[string]config.BillingToken{
+					"only-catalogued-model": {Prompt: "1.00", Completion: "2.00"},
+				},
+			},
+		},
+	}
+	prices, err := Compile(profile, sources)
+	if err != nil {
+		t.Fatalf("compile: %v", err)
+	}
+	srcs := prices["sources"].(map[string]interface{})
+
+	// The explicitly free source keeps its entry (free is a deliberate 0).
+	if _, ok := srcs["free1"]; !ok {
+		t.Error("the explicit free rule lost its source entry")
+	}
+	// Paid sources must NOT be marked, or their unlisted models bill $0.
+	for _, name := range []string{"paid1", "paid2"} {
+		if _, ok := srcs[name]; ok {
+			t.Errorf("★ wildcard token rule marked %q as priced at 0 — its unlisted "+
+				"models will bill $0 instead of appearing as unpriced", name)
+		}
+	}
+	// The catalogue itself must still be present, or the rule prices nothing.
+	models := prices["models"].(map[string]interface{})
+	if _, ok := models["only-catalogued-model"]; !ok {
+		t.Error("the wildcard rule did not publish its model price")
+	}
+}
+
+// The inverse guard: a token rule for a SPECIFIC url still marks that source,
+// because it genuinely declares pricing for the models served from it.
+func TestSpecificURLTokenRuleStillMarksItsSource(t *testing.T) {
+	sources := []config.Source{
+		{Name: "mine", BaseURL: "https://mine.example.com/v1"},
+		{Name: "other", BaseURL: "https://other.example.com/v1"},
+	}
+	profile := &config.BillingProfile{
+		ID: "p",
+		Rules: []config.BillingRule{{
+			URL: "https://mine.example.com/v1", Mode: "token",
+			Models: map[string]config.BillingToken{
+				"m": {Prompt: "1.00", Completion: "2.00"},
+			},
+		}},
+	}
+	prices, err := Compile(profile, sources)
+	if err != nil {
+		t.Fatalf("compile: %v", err)
+	}
+	srcs := prices["sources"].(map[string]interface{})
+	if _, ok := srcs["mine"]; !ok {
+		t.Error("a specific-url token rule must still mark its own source priced")
+	}
+	if _, ok := srcs["other"]; ok {
+		t.Error("an unrelated source was marked priced by a specific-url rule")
+	}
+}
+
+// TestReasoningPriceIsOmittedUnlessSet pins the absence convention for the
+// thinking-token rate.
+//
+// A rule that says nothing about reasoning must produce a price entry with NO
+// reasoning_price key, so the billing plugin skips the repricing branch and
+// bills thinking at the ordinary output rate — which is what every provider
+// actually charges. Writing an explicit 0 instead would read as "thinking is
+// free" and silently under-bill every request from that model.
+func TestReasoningPriceIsOmittedUnlessSet(t *testing.T) {
+	sources := []config.Source{{Name: "up", BaseURL: "https://up.example/v1"}}
+
+	// Case 1: no reasoning declared → the key must not appear.
+	p1 := &config.BillingProfile{ID: "p", Rules: []config.BillingRule{{
+		URL: "https://up.example/v1", Mode: "token",
+		Models: map[string]config.BillingToken{
+			"m": {Prompt: "1.00", Completion: "2.00"},
+		},
+	}}}
+	prices, err := Compile(p1, sources)
+	if err != nil {
+		t.Fatal(err)
+	}
+	entry := prices["models"].(map[string]interface{})["m"].(map[string]interface{})
+	if _, present := entry["reasoning_price"]; present {
+		t.Errorf("reasoning_price present (%v) without a rule; the plugin would treat 0 as free",
+			entry["reasoning_price"])
+	}
+
+	// Case 2: declared → present, and converted to per-token like every other rate.
+	p2 := &config.BillingProfile{ID: "p", Rules: []config.BillingRule{{
+		URL: "https://up.example/v1", Mode: "token",
+		Models: map[string]config.BillingToken{
+			"m": {Prompt: "1.00", Completion: "2.00", Reasoning: "6.00"},
+		},
+	}}}
+	prices2, err := Compile(p2, sources)
+	if err != nil {
+		t.Fatal(err)
+	}
+	e2 := prices2["models"].(map[string]interface{})["m"].(map[string]interface{})
+	rp, ok := e2["reasoning_price"].(float64)
+	if !ok {
+		t.Fatalf("reasoning_price missing or wrong type: %#v", e2["reasoning_price"])
+	}
+	if rp != 6e-6 {
+		t.Errorf("reasoning_price = %v, want 6e-6 (USD per TOKEN)", rp)
+	}
+	// A malformed rate must be rejected, not silently ignored: an operator who
+	// typed "6 USD" expects either that value or an error, never a rule that
+	// looks applied and bills nothing.
+	p3 := &config.BillingProfile{ID: "p", Rules: []config.BillingRule{{
+		URL: "https://up.example/v1", Mode: "token",
+		Models: map[string]config.BillingToken{
+			"m": {Prompt: "1.00", Completion: "2.00", Reasoning: "six dollars"},
+		},
+	}}}
+	if _, err := Compile(p3, sources); err == nil {
+		t.Error("a non-numeric reasoning rate must fail the compile")
+	}
+}
+
+// TestLongContextStepCompiles pins the long-context step's two accepted forms
+// and its rejection of a step that would silently do nothing.
+//
+// Providers publish both shapes — DeepSeek says "2x past 128k", OpenAI quotes
+// an absolute "$3.00/M past 200k" — so both must compile. Neither may be
+// missing: a step with no rate is a rule that LOOKS like it handles long
+// context and does not, which under-bills without any visible sign.
+func TestLongContextStepCompiles(t *testing.T) {
+	sources := []config.Source{{Name: "up", BaseURL: "https://up.example/v1"}}
+
+	compile := func(lc config.BillingLongContext) (map[string]interface{}, error) {
+		p := &config.BillingProfile{ID: "p", Rules: []config.BillingRule{{
+			URL: "https://up.example/v1", Mode: "token",
+			Models: map[string]config.BillingToken{
+				"m": {Prompt: "1.00", Completion: "2.00", LongContext: &lc},
+			},
+		}}}
+		prices, err := Compile(p, sources)
+		if err != nil {
+			return nil, err
+		}
+		e := prices["models"].(map[string]interface{})["m"].(map[string]interface{})
+		return e["long_context"].(map[string]interface{}), nil
+	}
+
+	// Absolute rate (OpenAI's published shape).
+	lc1, err := compile(config.BillingLongContext{Above: 200000, Prompt: "3.00"})
+	if err != nil {
+		t.Fatalf("absolute long_context rate rejected: %v", err)
+	}
+	if lc1["above"] != 200000 || lc1["prompt"] != 3e-6 {
+		t.Errorf("absolute step = %#v, want above=200000 prompt=3e-6", lc1)
+	}
+
+	// Multiplier (DeepSeek's shape): 2x of the 1.00/M base.
+	lc2, err := compile(config.BillingLongContext{Above: 128000, Multiplier: 2})
+	if err != nil {
+		t.Fatalf("multiplier long_context rejected: %v", err)
+	}
+	if lc2["prompt"] != 2e-6 {
+		t.Errorf("multiplier step = %#v, want prompt=2e-6", lc2)
+	}
+
+	// Both given: the absolute rate wins, because that is the number the
+	// provider quotes. Multiplying instead would produce a figure the operator
+	// never wrote.
+	lc3, err := compile(config.BillingLongContext{Above: 128000, Prompt: "5.00", Multiplier: 2})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if lc3["prompt"] != 5e-6 {
+		t.Errorf("both given = %#v, want the quoted absolute rate 5e-6", lc3)
+	}
+
+	// Neither: a no-op step must fail, not compile into a rule that silently
+	// bills long context at the base rate.
+	if _, err := compile(config.BillingLongContext{Above: 128000}); err == nil {
+		t.Error("a step with neither prompt nor multiplier must be rejected")
+	}
+	// A zero threshold would make every request "long".
+	if _, err := compile(config.BillingLongContext{Above: 0, Multiplier: 2}); err == nil {
+		t.Error("above=0 must be rejected")
+	}
+}

internal/config/config.go

@@ -7,6 +7,8 @@ import (
 	"fmt"
 	"os"
 	"path/filepath"
+	"strconv"
+	"strings"
 	"time"
 
 	"gopkg.in/yaml.v3"
@@ -19,6 +21,7 @@ type Config struct {
 	GatewayKeys   []string     `yaml:"gateway_keys"`
 	DefaultModel  string       `yaml:"default_model"` // e.g. "AUTO" or a model id
 	AdapterDir    string       `yaml:"adapter_dir"`
+	PluginDir     string       `yaml:"plugin_dir,omitempty"` // Lua plugin directory (optional; hooks + WebUI extensions)
 	RuntimeFile   string       `yaml:"runtime_file"`
 	TLSCertFile   string       `yaml:"tls_cert_file,omitempty"`   // PEM cert; when set together with tls_key_file, serve HTTPS
 	TLSKeyFile    string       `yaml:"tls_key_file,omitempty"`    // PEM private key
@@ -27,6 +30,29 @@ type Config struct {
 	Auto          []ModelScope `yaml:"auto,omitempty"`       // AUTO 调度链规则（WebUI 优先级页编辑，chat）
 	AutoImage     []ModelScope `yaml:"auto_image,omitempty"` // AUTO 生图调度链规则（WebUI 优先级页·生图）
 	Keys          []GWKey      `yaml:"keys,omitempty"`       // 网关密钥（WebUI 密钥页管理）
+	// BillingDSL declares per-URL pricing profiles for the billing plugin.
+	//
+	// Declarative, because hand-writing the prices JSON was error-prone in ways
+	// that looked successful: the first PUT went out without the required
+	// `prices` wrapper and silently REPLACED the accumulated totals with the
+	// price table; the second used a peak-window shape the plugin does not read,
+	// so peak traffic was billed at off-peak rates — no error, the bill just
+	// quietly halved. A schema-checked config file catches both at load.
+	BillingDSL *BillingDSL `yaml:"billing,omitempty"`
+	// PluginVisibility overrides the audience a plugin declared for itself,
+	// keyed by plugin name: "admin" hides a plugin (and its state) from user
+	// keys, "all" widens one that shipped admin-only, "user,admin" names roles
+	// explicitly. An absent name leaves the plugin's own declaration in force.
+	//
+	// This exists because a plugin author cannot know the operator's policy, and
+	// the operator should not have to edit a bundled .lua to change one — the
+	// next upgrade overwrites it.
+	PluginVisibility map[string]string `yaml:"plugin_visibility,omitempty" json:"plugin_visibility,omitempty"`
+	// box seals credentials (sources' api_key/headers, keys' key) at rest.
+	// In-memory values are always plaintext; only the bytes on disk are sealed.
+	// Wired by AttachSecretBox — Load leaves it nil so `-check` and tests stay
+	// filesystem-free.
+	box *SecretBox
 }
 
 // Defaults applied to any source (YAML or runtime) that leaves a field unset.
@@ -127,6 +153,7 @@ func writeDefaultConfig(path string) error {
 		GatewayKeys:  []string{key},
 		DefaultModel: "AUTO",
 		AdapterDir:   filepath.Join(abs, "adapters"),
+		PluginDir:    filepath.Join(abs, "plugins"),
 		RuntimeFile:  filepath.Join(abs, "runtime.json"),
 	}
 	out, err := yaml.Marshal(&cfg)
@@ -192,10 +219,15 @@ func RemoveSourceFromYAML(path, name string) error {
 
 // UpsertSourceInYAML adds or updates a source entry in the YAML config file.
 // Uses yaml.Node to preserve the rest of the file's comments and formatting.
-func UpsertSourceInYAML(path, name string, src Source) error {
+// When box is non-nil the source's credentials are sealed before writing, so a
+// newly added source never lands in the file as plaintext.
+func UpsertSourceInYAML(path, name string, src Source, box *SecretBox) error {
 	if path == "" {
 		return fmt.Errorf("config path is empty")
 	}
+	if err := sealSource(&src, box); err != nil {
+		return err
+	}
 	data, err := os.ReadFile(path)
 	if err != nil {
 		return err
@@ -253,11 +285,17 @@ func UpsertSourceInYAML(path, name string, src Source) error {
 
 // Save writes the current config back to the YAML file (preserving comments
 // via yaml.Node round-trip when possible, or full marshaling as fallback).
+// Credentials are sealed on the way out and the in-memory copy is restored to
+// plaintext afterwards, so callers keep working with usable values.
 func (c *Config) Save() error {
 	if c.Path == "" {
 		return fmt.Errorf("config path is empty")
 	}
+	if err := c.sealInPlace(c.box); err != nil {
+		return err
+	}
 	out, err := yaml.Marshal(c)
+	c.unsealAfterWrite(c.box)
 	if err != nil {
 		return fmt.Errorf("marshal config: %w", err)
 	}
@@ -272,6 +310,9 @@ func (c *Config) ApplyDefaults() error {
 	if c.AdapterDir == "" {
 		c.AdapterDir = "adapters"
 	}
+	// PluginDir is deliberately NOT defaulted: with no plugins a missing
+	// directory would be created and seeded on every start for nothing. An
+	// operator opts in by setting it (or by dropping files into it).
 	if c.RuntimeFile == "" {
 		c.RuntimeFile = "runtime.json"
 	}
@@ -352,24 +393,271 @@ type SourceTemplate struct {
 // (full management) or "user" (sees only its own key); Models is the allowed
 // model scope with per-model token quota (0 = unlimited).
 type GWKey struct {
-	Key       string       `yaml:"key" json:"key"`
-	Role      string       `yaml:"role" json:"role"`
-	Name      string       `yaml:"name,omitempty" json:"name,omitempty"`
-	Models    []ModelScope `yaml:"models,omitempty" json:"models,omitempty"`
-	Note      string       `yaml:"note,omitempty" json:"note,omitempty"`
-	CreatedAt int64        `yaml:"created_at,omitempty" json:"created_at,omitempty"`
-	Seed      bool         `yaml:"seed,omitempty" json:"seed,omitempty"` // true if migrated from config gateway_keys
+	Key   string `yaml:"key" json:"key"`
+	Role  string `yaml:"role" json:"role"`
+	Name  string `yaml:"name,omitempty" json:"name,omitempty"`
+	// Auto is this key's own AUTO scheduling chain. It overrides the global
+	// cfg.Auto for requests authenticated with this key; nil (or an empty
+	// list) means "inherit the global chain", so keys created before per-key
+	// chains existed keep working untouched.
+	Auto           []ModelScope `yaml:"auto,omitempty" json:"auto,omitempty"`
+	Models         []ModelScope `yaml:"models,omitempty" json:"models,omitempty"`
+	Note           string       `yaml:"note,omitempty" json:"note,omitempty"`
+	CreatedAt      int64        `yaml:"created_at,omitempty" json:"created_at,omitempty"`
+	Seed           bool         `yaml:"seed,omitempty" json:"seed,omitempty"` // true if migrated from config gateway_keys
 }
 
-// ModelScope is one allowed model for a key, or one AUTO scheduling slot,
-// with an optional token quota and reset period. TokenQuota 0 = unlimited;
-// Period "" = never resets; "hour"/"week"/"month" are fixed windows; "nhour"
-// uses Hours as the window length in hours.
+// HasOwnAuto reports whether the key declares its own AUTO chain rather than
+// inheriting the global one. An explicitly empty list still counts as "no own
+// chain" so a user can clear an override back to the global default.
+func (k GWKey) HasOwnAuto() bool { return len(k.Auto) > 0 }
+
+// BillingDSL holds declarative per-URL pricing profiles for the billing plugin.
+//
+// Profiles are the "let the user pick" axis: the same upstream URL can be
+// declared under multiple profiles, and the operator chooses which is active.
+// Switching recomputes and re-injects the billing plugin's prices table.
+type BillingDSL struct {
+	// Active is the id of the profile in effect. Empty => the first profile.
+	Active   string           `yaml:"active,omitempty" json:"active,omitempty"`
+	Profiles []BillingProfile `yaml:"profiles,omitempty" json:"profiles,omitempty"`
+}
+
+// BillingProfile is one named set of per-URL pricing rules.
+type BillingProfile struct {
+	ID    string `yaml:"id" json:"id"`
+	Label string `yaml:"label,omitempty" json:"label,omitempty"`
+	// Default applies when no rule matches: "free" or "unpriced".
+	Default string        `yaml:"default,omitempty" json:"default,omitempty"`
+	Rules   []BillingRule `yaml:"rules,omitempty" json:"rules,omitempty"`
+}
+
+// BillingRule declares the pricing for everything under one URL.
+type BillingRule struct {
+	// URL matches a source's base_url. "*" = any unmatched URL.
+	URL string `yaml:"url" json:"url"`
+	// Mode is one of: free, token, subscription, unpriced.
+	Mode     string `yaml:"mode" json:"mode"`
+	Currency string `yaml:"currency,omitempty" json:"currency,omitempty"`
+	// Subscription is the flat monthly cost (for mode=subscription).
+	Subscription float64 `yaml:"subscription,omitempty" json:"subscription,omitempty"`
+	// Models declares per-token prices when mode=token.
+	Models map[string]BillingToken `yaml:"models,omitempty" json:"models,omitempty"`
+	// Free is true when mode=free (kept for explicitness).
+	// Peak applies to ALL token-priced requests under this URL.
+	Peak *BillingPeak `yaml:"peak,omitempty" json:"peak,omitempty"`
+}
+
+// BillingToken is the per-model price.
+type BillingToken struct {
+	Prompt        string   `yaml:"prompt" json:"prompt"`         // USD per million tokens, e.g. "0.15"
+	Completion    string   `yaml:"completion" json:"completion"` // USD per million tokens
+	CacheDiscount *float64 `yaml:"cache_discount,omitempty" json:"cache_discount,omitempty"`
+	// Reasoning is the USD-per-million rate for thinking tokens, for the rare
+	// provider that charges them differently from ordinary output. Omit it and
+	// reasoning is billed at Completion — which is what OpenAI, DeepSeek and
+	// Anthropic all do, and what every source on this deployment reports.
+	//
+	// It is a rate for a SUBSET of completion tokens, never an extra charge: the
+	// billing plugin applies only the difference between this and Completion, so
+	// setting it to a multiple of Completion reprices the thinking share rather
+	// than billing it twice.
+	Reasoning string `yaml:"reasoning,omitempty" json:"reasoning,omitempty"`
+	// LongContext is the prompt-price step above a token threshold. Nil = one
+	// flat rate, which is what a rule means when it says nothing.
+	LongContext *BillingLongContext `yaml:"long_context,omitempty" json:"long_context,omitempty"`
+}
+
+// BillingPeak declares peak-hour pricing.
+type BillingPeak struct {
+	Multiplier float64 `yaml:"multiplier" json:"multiplier"`
+	// Weekdays: 1=Mon..5=Fri (Lua os.date !%w, 0=Sun)
+	Weekdays []int `yaml:"weekdays,omitempty" json:"weekdays,omitempty"`
+	// Hours: each pair is [lo,hi] inclusive
+	Hours [][2]int `yaml:"hours,omitempty" json:"hours,omitempty"`
+}
+
+// BillingLongContext is a prompt-price step for long inputs.
+//
+// Every major provider charges the input tokens ABOVE a threshold at a higher
+// rate — DeepSeek doubles past 128k, OpenAI and Anthropic step up as well. A
+// flat prompt price therefore under-bills exactly the traffic an agent workload
+// generates, and it under-bills it silently: the number is plausible, the bill
+// adds up, and nothing says the rate was wrong.
+//
+// The step applies to the PROMPT only. Output and cache reads are billed at
+// their own rates regardless of input length, which is how providers publish
+// it — the higher rate is for the extra context you made the model read, not for
+// what it wrote back.
+type BillingLongContext struct {
+	// Above is the token count the step starts at. A request with exactly this
+	// many prompt tokens is still at the base rate; the step covers the tokens
+	// BEYOND it. Inclusive-below matches how the providers document it.
+	Above int `yaml:"above" json:"above"`
+	// Prompt is the USD-per-million rate for the tokens above Above.
+	Prompt string `yaml:"prompt" json:"prompt"`
+	// Multiplier is an alternative to Prompt: a factor applied to the base rate
+	// (DeepSeek's published step is exactly 2x). When both are set Prompt wins,
+	// and setting neither is a validation error rather than a silent no-op.
+	Multiplier float64 `yaml:"multiplier,omitempty" json:"multiplier,omitempty"`
+}
+
+// Validate rejects a BillingDSL that could not work as written.
+func (b *BillingDSL) Validate() error {
+	if b == nil {
+		return nil
+	}
+	ids := map[string]bool{}
+	for i := range b.Profiles {
+		p := &b.Profiles[i]
+		if p.ID == "" {
+			return fmt.Errorf("billing profile [%d] has no id", i)
+		}
+		if ids[p.ID] {
+			return fmt.Errorf("billing profile id %q appears more than once", p.ID)
+		}
+		ids[p.ID] = true
+		if p.Default != "" && p.Default != "free" && p.Default != "unpriced" {
+			return fmt.Errorf("profile %q: default must be free or unpriced, got %q", p.ID, p.Default)
+		}
+		for j := range p.Rules {
+			r := &p.Rules[j]
+			if r.URL == "" {
+				return fmt.Errorf("profile %q rule [%d]: url is required", p.ID, j)
+			}
+			switch r.Mode {
+			case "free", "token", "subscription", "unpriced":
+			default:
+				return fmt.Errorf("profile %q rule [%d]: mode must be free/token/subscription/unpriced, got %q", p.ID, j, r.Mode)
+			}
+			if r.Mode == "token" {
+				for m, t := range r.Models {
+					if t.Prompt == "" || t.Completion == "" {
+						return fmt.Errorf("profile %q model %q: prompt and completion are required for token mode", p.ID, m)
+					}
+					if _, err := ParseFloatUSDPerM(t.Prompt); err != nil {
+						return fmt.Errorf("profile %q model %q prompt: %w", p.ID, m, err)
+					}
+					if _, err := ParseFloatUSDPerM(t.Completion); err != nil {
+						return fmt.Errorf("profile %q model %q completion: %w", p.ID, m, err)
+					}
+				}
+			}
+		}
+	}
+	// Active must resolve to a profile id.
+	if b.Active != "" && !ids[b.Active] {
+		return fmt.Errorf("billing.active = %q but no profile has that id", b.Active)
+	}
+	return nil
+}
+
+// Resolve returns the profile with the given id; empty id (or an id that does
+// not exist) falls back to the FIRST profile. A config with profiles but no
+// active marker should mean "use the one I wrote first" rather than "price
+// nothing", which is what a nil here would silently do.
+func (b *BillingDSL) Resolve(id string) *BillingProfile {
+	if b == nil || len(b.Profiles) == 0 {
+		return nil
+	}
+	if id != "" {
+		for i := range b.Profiles {
+			if b.Profiles[i].ID == id {
+				return &b.Profiles[i]
+			}
+		}
+	}
+	return &b.Profiles[0]
+}
+
+// parseFloatUSDPerM parses a USD-per-million price string. Returns the
+// per-single-token rate the plugin expects (price / 1e6).
+//
+// Kept as a string in the DSL rather than a float: "0.15" survives round-trips
+// through YAML editors, and a malformed value must fail VALIDATION (with the
+// file and field named) rather than parse as 0 and under-bill silently — which
+// is exactly the failure mode that motivated the DSL.
+func ParseFloatUSDPerM(s string) (float64, error) {
+	f, err := strconv.ParseFloat(strings.TrimSpace(s), 64)
+	if err != nil {
+		return 0, fmt.Errorf("not a number: %q", s)
+	}
+	if f < 0 {
+		return 0, fmt.Errorf("negative price: %q", s)
+	}
+	return f / 1e6, nil
+}
+
+// KeyQuota is retained only to carry a scope entry's caps through the admin
+// API. Quotas are per model, never per key: there is deliberately no key-wide
+// total, so exhausting one model's budget never blocks the others.
+type KeyQuota struct {
+	TokenQuota int64  `json:"token_quota"`
+	ReqQuota   int64  `json:"req_quota"`
+	Period     string `json:"period"`
+	Hours      int64  `json:"hours"`
+}
+
+// NormalizeRole defaults an empty role to "user", so a key can never end up in
+// a state where no role means "neither admin nor user".
+func NormalizeRole(role string) string {
+	if role == "admin" {
+		return "admin"
+	}
+	return "user"
+}
+
+// Validate rejects a quota configuration that could not work as written. A
+// period is only meaningful when at least one cap is set, and a cap of zero
+// means "unlimited" rather than "deny everything", so those are the only two
+// things worth rejecting.
+func (q KeyQuota) Validate() error {
+	if q.TokenQuota < 0 {
+		return fmt.Errorf("token_quota must be >= 0 (0 = unlimited)")
+	}
+	if q.ReqQuota < 0 {
+		return fmt.Errorf("req_quota must be >= 0 (0 = unlimited)")
+	}
+	if q.Hours < 0 {
+		return fmt.Errorf("hours must be >= 0")
+	}
+	if q.TokenQuota > 0 || q.ReqQuota > 0 {
+		if err := ValidatePeriod(q.Period, q.Hours); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
+// ValidatePeriod accepts the quota period vocabulary: "" (never resets),
+// "hour", "week", "month", or "nhour" with hours >= 1. An unknown period is
+// rejected rather than silently treated as "never resets", which would turn a
+// typo into an all-time quota — the opposite of what the operator typed.
+func ValidatePeriod(period string, hours int64) error {
+	switch period {
+	case "", "hour", "week", "month":
+		return nil
+	case "nhour":
+		if hours < 1 {
+			return fmt.Errorf("period %q needs hours >= 1", period)
+		}
+		return nil
+	}
+	return fmt.Errorf("period must be one of \"\", hour, week, month, nhour (got %q)", period)
+}
+
+// ModelScope is one allowed model for a key, or one AUTO scheduling slot.
+// Its TokenQuota and ReqQuota cap THAT entry only, independently of every
+// other entry on the same key: a model that runs out of budget stops being
+// served while the key's other models keep working. TokenQuota 0 / ReqQuota 0
+// = unlimited. Period "" = never resets; "hour"/"week"/"month" are fixed
+// windows; "nhour" uses Hours.
 type ModelScope struct {
 	Model      string `yaml:"model" json:"model"`
 	Source     string `yaml:"source,omitempty" json:"source,omitempty"` // optional: pin to one upstream source; "" = any source
 	Tier       int    `yaml:"tier,omitempty" json:"tier,omitempty"`
 	TokenQuota int64  `yaml:"token_quota" json:"token_quota"`
+	ReqQuota   int64  `yaml:"req_quota,omitempty" json:"req_quota,omitempty"`
 	Period     string `yaml:"period,omitempty" json:"period,omitempty"`
 	Hours      int64  `yaml:"hours,omitempty" json:"hours,omitempty"`
 }
@@ -388,6 +676,7 @@ func (m *ModelScope) UnmarshalJSON(b []byte) error {
 		Source     string `json:"source"`
 		Tier       int    `json:"tier"`
 		TokenQuota int64  `json:"token_quota"`
+		ReqQuota   int64  `json:"req_quota"`
 		Period     string `json:"period"`
 		Hours      int64  `json:"hours"`
 	}
@@ -398,6 +687,7 @@ func (m *ModelScope) UnmarshalJSON(b []byte) error {
 	m.Source = o.Source
 	m.Tier = o.Tier
 	m.TokenQuota = o.TokenQuota
+	m.ReqQuota = o.ReqQuota
 	m.Period = o.Period
 	m.Hours = o.Hours
 	return nil

internal/config/secret_config.go

@@ -0,0 +1,317 @@
+package config
+
+// Secret handling for config.yaml.
+//
+// config.yaml is 0644 world-readable by design (ops need to inspect it), so any
+// credential in it must not sit there in plaintext. Sources' api_key / headers
+// and gateway keys are therefore sealed at rest with the same SecretBox used by
+// the runtime store, and unsealed in memory at load time.
+//
+// The invariant that makes this safe: **in-memory values are always plaintext**,
+// and the enc:v1: prefix is what marks a file value as sealed. Read paths that
+// predate this (core.resolveSourceKey) already unseal, so only the write side and
+// the load-time normalize step are new.
+
+import (
+	"fmt"
+	"log"
+	"os"
+	"strings"
+)
+
+// sealSource seals one source's credentials in place (used by the YAML upsert
+// path, which is a package function and therefore has no Config to borrow a box
+// from).
+func sealSource(s *Source, box *SecretBox) error {
+	if box == nil {
+		return nil
+	}
+	if s.APIKey != "" && !strings.HasPrefix(s.APIKey, encPrefix) {
+		v, err := box.Encrypt(s.APIKey)
+		if err != nil {
+			return err
+		}
+		s.APIKey = v
+	}
+	for k, v := range s.Headers {
+		if v == "" || strings.HasPrefix(v, encPrefix) {
+			continue
+		}
+		e, err := box.Encrypt(v)
+		if err != nil {
+			return err
+		}
+		s.Headers[k] = e
+	}
+	return nil
+}
+
+// normalizeSecrets unseals every credential in the freshly parsed config so the
+// rest of the program only ever sees plaintext. A value without the enc:v1:
+// prefix is left untouched, which keeps hand-written plaintext configs working
+// (and is what a pre-encryption config file looks like).
+//
+// A value that carries the prefix but fails to decrypt is a hard error, not
+// something to paper over: returning the ciphertext (MustDecrypt's behavior)
+// would let the next Save re-seal it and turn one bad value into permanent,
+// compounding corruption. Losing the master key must be loud.
+func (c *Config) normalizeSecrets(box *SecretBox) error {
+	if box == nil {
+		return nil
+	}
+	for i := range c.Sources {
+		s := &c.Sources[i]
+		if strings.HasPrefix(s.APIKey, encPrefix) {
+			v, err := box.Decrypt(s.APIKey)
+			if err != nil {
+				return fmt.Errorf("source %q api_key: %w", s.Name, err)
+			}
+			s.APIKey = v
+		}
+		for k, v := range s.Headers {
+			if strings.HasPrefix(v, encPrefix) {
+				d, err := box.Decrypt(v)
+				if err != nil {
+					return fmt.Errorf("source %q header %q: %w", s.Name, k, err)
+				}
+				s.Headers[k] = d
+			}
+		}
+	}
+	for i := range c.Keys {
+		if strings.HasPrefix(c.Keys[i].Key, encPrefix) {
+			v, err := box.Decrypt(c.Keys[i].Key)
+			if err != nil {
+				return fmt.Errorf("gateway key %q: %w", c.Keys[i].Name, err)
+			}
+			c.Keys[i].Key = v
+		}
+	}
+	return nil
+}
+
+// sealInPlace replaces plaintext credentials with ciphertext for writing. It is
+// deliberately a separate step from Marshal: callers that need the plaintext
+// (auth comparisons, log output, returning a key to the operator who just
+// created it) must not be handed a sealed config by accident.
+func (c *Config) sealInPlace(box *SecretBox) error {
+	if box == nil {
+		return nil
+	}
+	for i := range c.Sources {
+		s := &c.Sources[i]
+		if s.APIKey != "" && !strings.HasPrefix(s.APIKey, encPrefix) {
+			v, err := box.Encrypt(s.APIKey)
+			if err != nil {
+				return err
+			}
+			s.APIKey = v
+		}
+		if len(s.Headers) > 0 {
+			sealed := make(map[string]string, len(s.Headers))
+			for k, v := range s.Headers {
+				if v == "" || strings.HasPrefix(v, encPrefix) {
+					sealed[k] = v
+					continue
+				}
+				e, err := box.Encrypt(v)
+				if err != nil {
+					return err
+				}
+				sealed[k] = e
+			}
+			s.Headers = sealed
+		}
+	}
+	for i := range c.Keys {
+		k := &c.Keys[i]
+		if k.Key != "" && !strings.HasPrefix(k.Key, encPrefix) {
+			v, err := box.Encrypt(k.Key)
+			if err != nil {
+				return err
+			}
+			k.Key = v
+		}
+	}
+	return nil
+}
+
+// unsealAfterWrite restores plaintext after a sealed marshal so the live process
+// keeps working on plaintext values (mirrors Store.persistLocked's dance).
+func (c *Config) unsealAfterWrite(box *SecretBox) {
+	// Save just encrypted every value it can see, so a failure here is
+	// impossible; ignore the error rather than panic in a write path.
+	_ = c.normalizeSecrets(box)
+}
+
+// hasPlaintextSecrets reports whether any credential in the config is still in
+// the clear. Used to decide whether a startup migration write is needed, and to
+// warn (without leaking values) when no master key is available.
+func (c *Config) hasPlaintextSecrets() bool {
+	for _, s := range c.Sources {
+		if s.APIKey != "" && !strings.HasPrefix(s.APIKey, encPrefix) {
+			return true
+		}
+		for _, v := range s.Headers {
+			if v != "" && !strings.HasPrefix(v, encPrefix) {
+				return true
+			}
+		}
+	}
+	for _, k := range c.Keys {
+		if k.Key != "" && !strings.HasPrefix(k.Key, encPrefix) {
+			return true
+		}
+	}
+	return false
+}
+
+// countPlaintextSecrets returns how many credentials are still in the clear, for
+// an operator-facing migration log line that must not print the values.
+func (c *Config) countPlaintextSecrets() int {
+	n := 0
+	for _, s := range c.Sources {
+		if s.APIKey != "" && !strings.HasPrefix(s.APIKey, encPrefix) {
+			n++
+		}
+		for _, v := range s.Headers {
+			if v != "" && !strings.HasPrefix(v, encPrefix) {
+				n++
+			}
+		}
+	}
+	for _, k := range c.Keys {
+		if k.Key != "" && !strings.HasPrefix(k.Key, encPrefix) {
+			n++
+		}
+	}
+	return n
+}
+
+// NormalizeSecretsForRun unseals the loaded config and then seals it back on
+// disk if anything was still in the clear. Order matters: Load() read the file
+// with ciphertext still in place, so the unseal has to happen before the
+// registry (and any Save the startup path performs) sees the values.
+func (c *Config) NormalizeSecretsForRun(box *SecretBox) error {
+	hadPlaintext, err := c.UnsealSecrets(box)
+	if err != nil {
+		return err
+	}
+	return c.SealIfNeeded(hadPlaintext)
+}
+
+// UnsealSecrets decrypts every sealed credential in memory and reports whether
+// the config ON DISK still held plaintext (i.e. whether a sealing write is
+// needed). It never writes; a failed decrypt (wrong master key) is returned so
+// the process refuses to start instead of running with unusable credentials.
+//
+// The return value must be computed BEFORE unsealing and from the disk state,
+// not from memory: after a Save the in-memory values are always plaintext, so a
+// "is anything plaintext?" test run afterwards is unconditionally true and a
+// caller would rewrite the file on every start. That was the actual behaviour -
+// migratePlaintextSecrets() claimed to be idempotent in a comment but rewrote
+// config.yaml on every boot.
+//
+// Callers that consume credentials (the provider registry, key seeding) must
+// unseal FIRST. Seeding compares cfg.Keys[i].Key against the plaintext
+// gateway_keys entries; running it while keys are still ciphertext made the
+// dedupe never match, so every restart appended another copy of the same admin
+// key (production accumulated four).
+func (c *Config) UnsealSecrets(box *SecretBox) (bool, error) {
+	if box == nil {
+		return false, nil
+	}
+	c.AttachSecretBox(box)
+	hadPlaintext := c.hasPlaintextSecrets()
+	if err := c.normalizeSecrets(box); err != nil {
+		return hadPlaintext, err
+	}
+	return hadPlaintext, nil
+}
+
+// SealIfNeeded writes the config back once if hadPlaintext reported that the
+// file still held clear-text credentials. When it is false the file is left
+// untouched, which is what makes startup a no-op for an already-sealed config.
+func (c *Config) SealIfNeeded(hadPlaintext bool) error {
+	if !hadPlaintext || c.box == nil || c.Path == "" {
+		return nil
+	}
+	n := c.countPlaintextSecrets()
+	if err := c.Save(); err != nil {
+		return err
+	}
+	log.Printf("[config] sealed %d plaintext credential(s) in %s", n, c.Path)
+	return nil
+}
+
+// AttachSecretBox wires the encryption box into the config so Save can seal
+// credentials. Kept as an explicit call (rather than a constructor argument) so
+// config.Load stays usable in contexts that have no filesystem secrets (tests,
+// `-check`).
+func (c *Config) AttachSecretBox(box *SecretBox) { c.box = box }
+
+// SecretBox returns the wired encryption box, or nil when none is attached.
+func (c *Config) SecretBox() *SecretBox { return c.box }
+
+// migratePlaintextSecrets seals any credential still in the clear and writes the
+// file once. Kept for callers that attach the box themselves; it decides from
+// the in-memory state, which is why UnsealSecrets + SealIfNeeded (which decide
+// from the on-disk state) are preferred on the startup path.
+func (c *Config) migratePlaintextSecrets() error {
+	if c.box == nil || c.Path == "" {
+		return nil
+	}
+	if !c.hasPlaintextSecrets() {
+		return nil
+	}
+	n := c.countPlaintextSecrets()
+	if err := c.Save(); err != nil {
+		return err
+	}
+	log.Printf("[config] sealed %d plaintext credential(s) in %s", n, c.Path)
+	return nil
+}
+
+// PrintSecrets writes the config's credentials to stdout in the clear and
+// returns an error only when the file cannot be read or the master key does not
+// match. It is the operator counterpart to sealing-at-rest: with keys stored as
+// ciphertext, "which key is this source using" must still be answerable.
+//
+// It deliberately takes a path rather than a *Config so the caller cannot
+// accidentally hand it a config that has already been unsealed in memory, and it
+// never writes anything.
+func PrintSecrets(path string) error {
+	cfg, err := Load(path)
+	if err != nil {
+		return err
+	}
+	box, err := NewSecretBox(cfg.RuntimeFile)
+	if err != nil {
+		return fmt.Errorf("%w (is master.key present and intact?)", err)
+	}
+	if err := cfg.normalizeSecrets(box); err != nil {
+		return err
+	}
+	w := os.Stdout
+	fmt.Fprintf(w, "# %s — %d source(s), %d gateway key(s)\n", path, len(cfg.Sources), len(cfg.Keys))
+	for _, s := range cfg.Sources {
+		fmt.Fprintf(w, "source %-16s api_key=%s\n", s.Name, orNone(s.APIKey))
+		for k, v := range s.Headers {
+			if v != "" {
+				fmt.Fprintf(w, "source %-16s header[%s]=%s\n", s.Name, k, v)
+			}
+		}
+	}
+	for _, k := range cfg.Keys {
+		fmt.Fprintf(w, "key    %-16s role=%-5s %s\n", k.Name, k.Role, k.Key)
+	}
+	fmt.Fprintf(w, "gateway_keys (legacy): %s\n", strings.Join(cfg.GatewayKeys, " "))
+	return nil
+}
+
+func orNone(s string) string {
+	if s == "" {
+		return "(unset)"
+	}
+	return s
+}

internal/config/secret_config_test.go

@@ -0,0 +1,323 @@
+package config
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// writeMasterKey plants a deterministic master.key next to the given runtime
+// file so tests exercise the real box instead of the env-var shortcut.
+func writeMasterKey(t *testing.T, runtimeFile string) *SecretBox {
+	t.Helper()
+	if err := os.WriteFile(filepath.Join(filepath.Dir(runtimeFile), "master.key"),
+		[]byte(strings.Repeat("ab", 32)), 0600); err != nil {
+		t.Fatal(err)
+	}
+	box, err := NewSecretBox(runtimeFile)
+	if err != nil {
+		t.Fatalf("NewSecretBox: %v", err)
+	}
+	return box
+}
+
+func TestSaveSealsCredentialsAndLoadUnseals(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "config.yaml")
+	runtime := filepath.Join(dir, "runtime.json")
+	box := writeMasterKey(t, runtime)
+
+	cfg := &Config{
+		Path:   path,
+		Listen: "127.0.0.1:0",
+		Sources: []Source{{
+			Name:    "up",
+			BaseURL: "http://up/v1",
+			APIKey:  "sk-plaintext-secret",
+			Adapter: "openai",
+			Headers: map[string]string{"X-Extra": "header-secret"},
+			Models:  []Model{{ID: "m", Kind: "chat"}},
+		}},
+		Keys: []GWKey{{Key: "sk-gw-plain", Role: "admin", Name: "admin"}},
+	}
+	cfg.AttachSecretBox(box)
+	if err := cfg.Save(); err != nil {
+		t.Fatalf("Save: %v", err)
+	}
+
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	text := string(raw)
+	if strings.Contains(text, "sk-plaintext-secret") {
+		t.Error("source api_key written in plaintext")
+	}
+	if strings.Contains(text, "header-secret") {
+		t.Error("source header written in plaintext")
+	}
+	if strings.Contains(text, "sk-gw-plain") {
+		t.Error("gateway key written in plaintext")
+	}
+	if n := strings.Count(text, encPrefix); n != 3 {
+		t.Errorf("expected 3 sealed values, found %d", n)
+	}
+
+	// The live in-memory config must still hold plaintext after Save.
+	if cfg.Sources[0].APIKey != "sk-plaintext-secret" {
+		t.Errorf("in-memory api_key mutated by Save: %q", cfg.Sources[0].APIKey)
+	}
+	if cfg.Keys[0].Key != "sk-gw-plain" {
+		t.Errorf("in-memory gateway key mutated by Save: %q", cfg.Keys[0].Key)
+	}
+
+	// A fresh load must hand back plaintext again.
+	loaded, err := Load(path)
+	if err != nil {
+		t.Fatalf("Load: %v", err)
+	}
+	loaded.AttachSecretBox(box)
+	loaded.normalizeSecrets(box)
+	if loaded.Sources[0].APIKey != "sk-plaintext-secret" {
+		t.Errorf("loaded api_key = %q, want plaintext", loaded.Sources[0].APIKey)
+	}
+	if loaded.Sources[0].Headers["X-Extra"] != "header-secret" {
+		t.Errorf("loaded header = %q, want plaintext", loaded.Sources[0].Headers["X-Extra"])
+	}
+	if loaded.Keys[0].Key != "sk-gw-plain" {
+		t.Errorf("loaded gateway key = %q, want plaintext", loaded.Keys[0].Key)
+	}
+}
+
+func TestPlaintextConfigStillLoads(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "config.yaml")
+	// A hand-written, pre-encryption config: no enc:v1: anywhere.
+	body := `listen: 127.0.0.1:0
+sources:
+  - name: legacy
+    base_url: http://legacy/v1
+    api_key: sk-written-by-hand
+    adapter: openai
+    models:
+      - id: m
+        kind: chat
+`
+	if err := os.WriteFile(path, []byte(body), 0644); err != nil {
+		t.Fatal(err)
+	}
+	cfg, err := Load(path)
+	if err != nil {
+		t.Fatalf("Load: %v", err)
+	}
+	box := writeMasterKey(t, filepath.Join(dir, "runtime.json"))
+	cfg.AttachSecretBox(box)
+	cfg.normalizeSecrets(box)
+	if cfg.Sources[0].APIKey != "sk-written-by-hand" {
+		t.Errorf("plaintext config value changed: %q", cfg.Sources[0].APIKey)
+	}
+}
+
+// Sealing a config is a one-time migration: the FIRST run writes, every later
+// start must leave the file alone.
+//
+// This test previously called migratePlaintextSecrets() twice and compared
+// ModTime. That assertion was both weak and wrong:
+//
+//   - migratePlaintextSecrets decides from IN-MEMORY state, and Save() ends by
+//     unsealing memory so the running process keeps working. So every call sees
+//     "plaintext present" and rewrites the file. Comparing ModTime hid this
+//     because two writes inside one filesystem timestamp tick look identical —
+//     the test flaked (~1 in 6) instead of failing.
+//   - it also exercised a function production no longer calls on the startup
+//     path, which now uses UnsealSecrets + SealIfNeeded (they decide from the
+//     ON-DISK state).
+//
+// So the test targets that real path and compares file CONTENT, which cannot be
+// fooled by timestamp granularity.
+func TestSealingIsIdempotentAcrossStarts(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "config.yaml")
+	runtime := filepath.Join(dir, "runtime.json")
+	box := writeMasterKey(t, runtime)
+
+	cfg := &Config{
+		Path:    path,
+		Listen:  "127.0.0.1:0",
+		Sources: []Source{{Name: "up", BaseURL: "http://up/v1", APIKey: "sk-clear", Adapter: "openai", Models: []Model{{ID: "m"}}}},
+	}
+	cfg.AttachSecretBox(box)
+
+	hadPlaintext, err := cfg.UnsealSecrets(box)
+	if err != nil {
+		t.Fatalf("unseal: %v", err)
+	}
+	if !hadPlaintext {
+		t.Fatal("a config written with a clear-text credential must report plaintext")
+	}
+	if err := cfg.SealIfNeeded(hadPlaintext); err != nil {
+		t.Fatalf("seal: %v", err)
+	}
+	first, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(first), encPrefix) {
+		t.Fatal("sealing did not encrypt the value on disk")
+	}
+	// In-memory must be plaintext so the running process keeps working.
+	if cfg.Sources[0].APIKey != "sk-clear" {
+		t.Errorf("in-memory api_key = %q, want plaintext", cfg.Sources[0].APIKey)
+	}
+
+	// Simulate the next start: load from disk, unseal, seal-if-needed. The file
+	// was already sealed, so nothing may be written.
+	cfg2, err := Load(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	cfg2.AttachSecretBox(box)
+	hadPlaintext2, err := cfg2.UnsealSecrets(box)
+	if err != nil {
+		t.Fatalf("second unseal: %v", err)
+	}
+	if hadPlaintext2 {
+		t.Error("a sealed config reported plaintext — SealIfNeeded would rewrite " +
+			"the file on every start")
+	}
+	if err := cfg2.SealIfNeeded(hadPlaintext2); err != nil {
+		t.Fatalf("second seal: %v", err)
+	}
+	second, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(first) != string(second) {
+		t.Error("the second start rewrote an already-sealed config; startup must be a no-op")
+	}
+}
+
+func TestSealedValueDoesNotDoubleEncrypt(t *testing.T) {
+	dir := t.TempDir()
+	box := writeMasterKey(t, filepath.Join(dir, "runtime.json"))
+	cfg := &Config{Path: filepath.Join(dir, "config.yaml"), Sources: []Source{{Name: "a", BaseURL: "http://a"}}}
+	cfg.AttachSecretBox(box)
+
+	once, err := box.Encrypt("sk-x")
+	if err != nil {
+		t.Fatal(err)
+	}
+	cfg.Sources[0].APIKey = once
+	if err := cfg.Save(); err != nil {
+		t.Fatal(err)
+	}
+	got, err := box.Decrypt(cfg.Sources[0].APIKey)
+	if err != nil {
+		t.Fatalf("value became undecryptable: %v", err)
+	}
+	if got != "sk-x" {
+		t.Errorf("round trip = %q, want sk-x", got)
+	}
+}
+
+func TestUpsertSourceInYAMLSealsCredentials(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "config.yaml")
+	box := writeMasterKey(t, filepath.Join(dir, "runtime.json"))
+
+	if err := os.WriteFile(path, []byte("listen: 127.0.0.1:0\nsources: []\n"), 0644); err != nil {
+		t.Fatal(err)
+	}
+	src := Source{Name: "new", BaseURL: "http://new/v1", APIKey: "sk-fresh", Adapter: "openai", Models: []Model{{ID: "m"}}}
+	if err := UpsertSourceInYAML(path, src.Name, src, box); err != nil {
+		t.Fatalf("UpsertSourceInYAML: %v", err)
+	}
+	raw, _ := os.ReadFile(path)
+	if strings.Contains(string(raw), "sk-fresh") {
+		t.Error("newly added source stored its api_key in plaintext")
+	}
+	if !strings.Contains(string(raw), encPrefix) {
+		t.Error("newly added source was not sealed")
+	}
+}
+
+func TestCountPlaintextSecrets(t *testing.T) {
+	cfg := &Config{
+		Sources: []Source{
+			{Name: "a", APIKey: encPrefix + "abc", Headers: map[string]string{"H": encPrefix + "x"}},
+			{Name: "b", APIKey: "sk-in-clear", Headers: map[string]string{"H2": "clear-too"}},
+		},
+		Keys: []GWKey{{Key: "sk-gw-clear"}},
+	}
+	if got := cfg.countPlaintextSecrets(); got != 3 {
+		t.Errorf("countPlaintextSecrets = %d, want 3", got)
+	}
+	if !cfg.hasPlaintextSecrets() {
+		t.Error("hasPlaintextSecrets = false, want true")
+	}
+	cfg.Sources[1].APIKey = encPrefix + "y"
+	cfg.Sources[1].Headers["H2"] = encPrefix + "z"
+	cfg.Keys[0].Key = encPrefix + "k"
+	if cfg.hasPlaintextSecrets() {
+		t.Error("hasPlaintextSecrets = true after sealing everything")
+	}
+}
+
+// TestNormalizeSecretsFailsLoudlyOnWrongMasterKey is the negative case that
+// guards the migration: with a wrong key, a sealed value must NOT be handed
+// back as ciphertext for a later re-seal (that compounds corruption and turns
+// one lost key file into permanently unusable config).
+func TestNormalizeSecretsFailsLoudlyOnWrongMasterKey(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "config.yaml")
+	runtime := filepath.Join(dir, "runtime.json")
+	good := writeMasterKey(t, runtime)
+
+	cfg := &Config{
+		Path:    path,
+		Listen:  "127.0.0.1:0",
+		Sources: []Source{{Name: "up", BaseURL: "http://up/v1", APIKey: "sk-secret", Adapter: "openai", Models: []Model{{ID: "m"}}}},
+	}
+	cfg.AttachSecretBox(good)
+	if err := cfg.Save(); err != nil {
+		t.Fatal(err)
+	}
+	sealedOnce, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// A different master key must be rejected, loudly. NewSecretBox derives the
+	// key path from the RUNTIME file's directory, so the wrong key has to live
+	// in a different directory — otherwise it would read the good master.key.
+	otherDir := t.TempDir()
+	if err := os.WriteFile(filepath.Join(otherDir, "master.key"),
+		[]byte(strings.Repeat("cd", 32)), 0600); err != nil {
+		t.Fatal(err)
+	}
+	badBox, err := NewSecretBox(filepath.Join(otherDir, "runtime.json"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	loaded, err := Load(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	loaded.AttachSecretBox(badBox)
+	if err := loaded.NormalizeSecretsForRun(badBox); err == nil {
+		t.Fatal("expected an error with the wrong master key, got nil")
+	}
+
+	// Crucially: the file must be untouched — no second seal on top.
+	after, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if string(after) != string(sealedOnce) {
+		t.Error("config file was rewritten despite the decrypt failure")
+	}
+	if n := strings.Count(string(after), encPrefix); n != strings.Count(string(sealedOnce), encPrefix) {
+		t.Errorf("sealed count changed %d -> %d (double encryption)", strings.Count(string(sealedOnce), encPrefix), n)
+	}
+}

internal/core/core.go

@@ -7,14 +7,17 @@ import (
 	"crypto/rand"
 	"encoding/hex"
 	"fmt"
+	"log"
 	"os"
 	"path/filepath"
 	"sort"
+	"net/url"
 	"strings"
 	"sync"
 	"sync/atomic"
 	"time"
 
+	"llmsproxy/internal/billing"
 	"llmsproxy/internal/config"
 	"llmsproxy/internal/lua"
 	"llmsproxy/internal/provider"
@@ -29,11 +32,18 @@ type Core struct {
 	mu             sync.Mutex
 	cfg            *config.Config
 	vm             *lua.VM
+	plugins        *lua.Plugins
 	store          *config.Store
 	scheduler      *scheduler.Scheduler
 	registry       *provider.Registry
 	autoChain      atomic.Pointer[scheduler.Chain] // chat AUTO chain
 	autoImageChain atomic.Pointer[scheduler.Chain] // image-generation AUTO chain
+
+	// keyAutoChains caches per-key AUTO chains, keyed by the gateway key.
+	// Entries are rebuilt by rebuildKeyAutoChains whenever sources or key
+	// config change; AutoChainFor reads them without taking c.mu so request
+	// handling stays lock-free like the global chain lookup.
+	keyAutoChains atomic.Pointer[map[string]*scheduler.Chain]
 }
 
 // New builds the core from a config file plus runtime overlay.
@@ -52,7 +62,74 @@ func NewFromConfig(cfg *config.Config) (*Core, error) {
 	if err := c.vm.Start(); err != nil {
 		return nil, fmt.Errorf("lua vm: %w", err)
 	}
+	// Plugins load AFTER the VM is up (a plugin is just another Lua state) but
+	// BEFORE anything can serve traffic, so a plugin's request_end hook sees
+	// the very first request. A plugin that fails to load is logged and skipped
+	// rather than failing startup: plugins are optional extensions, and failing
+	// to boot the gateway because a third-party .lua has a typo would be the
+	// wrong trade.
+	c.plugins = lua.NewPlugins(c.vm, cfg.PluginDir)
+	// Plugin outbound HTTP is scoped to the hosts of configured sources. Set
+	// before LoadDir so a plugin that requests on its first hook already sees
+	// the real allowlist rather than the deny-all placeholder.
+	c.plugins.SetOutboundHosts(sourceHosts(cfg))
+	// Let the plugin layer validate access specs against the keys that exist.
+	// A stored allowlist naming a key that does not exist refuses every caller,
+	// which looks exactly like a broken endpoint.
+	lua.SetKeyNameLookup(func() []string {
+		// ListKeys is the structured view (bare `gateway_keys:` entries are
+		// expanded into records with names); cfg.GatewayKeys is still []string.
+		keys := c.ListKeys()
+		out := make([]string, 0, len(keys))
+		for _, k := range keys {
+			if k.Name != "" {
+				out = append(out, k.Name)
+			}
+		}
+		return out
+	})
+	// Operator overrides win over the plugin's own declared audience.
+	for name, aud := range cfg.PluginVisibility {
+		c.plugins.SetPluginVisibility(name, aud)
+	}
+	if err := c.plugins.SeedBundled(); err != nil {
+		log.Printf("[core] seed bundled plugins: %v", err)
+	}
+	if err := c.plugins.LoadDir(); err != nil {
+		log.Printf("[core] plugin dir: %v", err)
+	}
+	for _, p := range c.plugins.List() {
+		if msg, bad := p["error"]; bad {
+			log.Printf("[core] plugin %v failed to load: %v", p["name"], msg)
+			continue
+		}
+		log.Printf("[core] plugin %v %v loaded (hooks=%v ui=%v)",
+			p["name"], p["version"], p["hooks"], p["ui"])
+	}
+	c.applyBillingDSL()
 	c.store = config.NewStore(cfg.RuntimeFile)
+	// Share one box between the runtime store and config.yaml so a single
+	// master.key seals both files. config.Load left the config holding
+	// ciphertext (if it was sealed); unseal it NOW, before anything reads a
+	// credential.
+	//
+	// ORDER IS LOAD-BEARING. These steps each consume secrets and must run
+	// after the unseal:
+	//   - seedKeys compares cfg.Keys[i].Key against the plaintext
+	//     gateway_keys entries; with ciphertext keys the comparison never
+	//     matched and every restart appended another duplicate admin key
+	//     (production had four copies of the same admin key).
+	//   - rebuildRegistry hands cfg.Sources[i].APIKey to the providers; with
+	//     ciphertext it built every upstream client with "enc:v1:..." as its
+	//     bearer token.
+	// Previously the unseal happened at the END of this function, and things
+	// only appeared to work because seedKeys' Save() unsealed memory as a side
+	// effect. Removing the redundant saves exposed the real ordering bug.
+	cfg.AttachSecretBox(c.store.SecretBox())
+	hadPlaintextSecrets, err := cfg.UnsealSecrets(c.store.SecretBox())
+	if err != nil {
+		return nil, fmt.Errorf("unseal secrets: %w", err)
+	}
 	if err := c.store.Load(); err != nil {
 		return nil, fmt.Errorf("runtime store: %w", err)
 	}
@@ -74,6 +151,12 @@ func NewFromConfig(cfg *config.Config) (*Core, error) {
 	if err := c.seedPresetTemplates(); err != nil {
 		return nil, fmt.Errorf("seed preset templates: %w", err)
 	}
+	// Seal any credential still in the clear in config.yaml. Startup writes
+	// nothing when the file was already sealed (hadPlaintextSecrets is decided
+	// from the on-disk state, before the unseal above).
+	if err := cfg.SealIfNeeded(hadPlaintextSecrets); err != nil {
+		return nil, fmt.Errorf("normalize secrets: %w", err)
+	}
 	return c, nil
 }
 
@@ -192,6 +275,11 @@ func buildRetries(cfg *config.Config) int {
 // VM exposes the Lua adapter runtime.
 func (c *Core) VM() *lua.VM { return c.vm }
 
+// Plugins exposes the loaded plugin set. Never nil once New* has returned, so
+// the gateway can fire stages unconditionally; an unconfigured plugin
+// directory yields an empty registry whose Fire is a no-op.
+func (c *Core) Plugins() *lua.Plugins { return c.plugins }
+
 func (c *Core) Scheduler() *scheduler.Scheduler { return c.scheduler }
 
 func (c *Core) Registry() *provider.Registry { return c.registry }
@@ -199,6 +287,68 @@ func (c *Core) Registry() *provider.Registry { return c.registry }
 // AutoChain returns the current AUTO scheduling chain.
 func (c *Core) AutoChain() *scheduler.Chain { return c.autoChain.Load() }
 
+// AutoChainFor returns the AUTO chain to use for a request authenticated with
+// key. A key that declares its own chain gets that chain; every other key
+// inherits the global chain, so keys created before per-key chains existed
+// keep their current behaviour with no configuration change.
+//
+// It returns (chain, ok). ok is false only when the key has its own chain but
+// it compiled to nothing usable (e.g. every slot names a model that no longer
+// exists) — the caller must surface that rather than silently downgrading the
+// user to the global chain, which would be a confusing, invisible switch.
+func (c *Core) AutoChainFor(key string) (*scheduler.Chain, bool) {
+	if m := c.keyAutoChains.Load(); m != nil {
+		if ch, ok := (*m)[key]; ok {
+			return ch, true
+		}
+	}
+	return c.autoChain.Load(), true
+}
+
+// rebuildKeyAutoChains recompiles every per-key AUTO chain from the current
+// config and provider registry, then swaps the cache in one shot so requests
+// never observe a half-built map. Keys without their own chain are absent
+// from the map and fall back to the global chain.
+func (c *Core) rebuildKeyAutoChains() {
+	m := make(map[string]*scheduler.Chain)
+	for _, k := range c.cfg.Keys {
+		if !k.HasOwnAuto() {
+			continue
+		}
+		m[k.Key] = c.chainForRules(k.Auto, false)
+	}
+	c.keyAutoChains.Store(&m)
+}
+
+// SaveKeyAuto persists one key's own AUTO chain and recompiles the per-key
+// chain cache. Passing an empty list clears the override so the key inherits
+// the global chain again.
+func (c *Core) SaveKeyAuto(key string, entries []config.ModelScope) error {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	idx := -1
+	for i, k := range c.cfg.Keys {
+		if k.Key == key {
+			idx = i
+			break
+		}
+	}
+	if idx < 0 {
+		return fmt.Errorf("key not found")
+	}
+	if err := ValidateScopeQuotas(entries); err != nil {
+		return err
+	}
+	c.cfg.Keys[idx].Auto = cleanScopes(entries)
+	if err := c.saveConfig(); err != nil {
+		return err
+	}
+	// Rebuild the whole cache so a cleared override stops shadowing the
+	// global chain for this key.
+	c.rebuildKeyAutoChains()
+	return nil
+}
+
 // AutoImageChain returns the persisted image-generation AUTO chain snapshot.
 func (c *Core) AutoImageChain() *scheduler.Chain { return c.autoImageChain.Load() }
 
@@ -238,10 +388,15 @@ func (c *Core) FindKey(key string) (config.GWKey, bool) {
 }
 
 // CreateKey builds a new random gateway key and persists it to config.yaml.
+// Quotas live on the model scope entries, so a new key's budget is whatever
+// its scopes carry.
 func (c *Core) CreateKey(name, role string, models []config.ModelScope, note string) (config.GWKey, error) {
 	c.mu.Lock()
 	defer c.mu.Unlock()
 	models = cleanScopes(models)
+	if err := ValidateScopeQuotas(models); err != nil {
+		return config.GWKey{}, err
+	}
 	key := make([]byte, 16)
 	if _, err := rand.Read(key); err != nil {
 		return config.GWKey{}, err
@@ -254,9 +409,7 @@ func (c *Core) CreateKey(name, role string, models []config.ModelScope, note str
 		Note:      note,
 		CreatedAt: time.Now().Unix(),
 	}
-	if rec.Role == "" {
-		rec.Role = "user"
-	}
+	rec.Role = config.NormalizeRole(rec.Role)
 	c.cfg.Keys = append(c.cfg.Keys, rec)
 	if err := c.saveConfig(); err != nil {
 		return config.GWKey{}, err
@@ -264,10 +417,16 @@ func (c *Core) CreateKey(name, role string, models []config.ModelScope, note str
 	return rec, nil
 }
 
-// UpdateKey mutates a key's name/role/model scope and persists it.
+// UpdateKey mutates a key's name/role/model scope and persists it. The scope
+// entries carry their own quotas, so replacing the scope replaces the budgets.
 func (c *Core) UpdateKey(key, name, role string, models []config.ModelScope, note string) (config.GWKey, error) {
 	c.mu.Lock()
 	defer c.mu.Unlock()
+	if models != nil {
+		if err := ValidateScopeQuotas(models); err != nil {
+			return config.GWKey{}, err
+		}
+	}
 	for i, k := range c.cfg.Keys {
 		if k.Key == key {
 			if name != "" {
@@ -439,6 +598,11 @@ func (c *Core) mergedSources() []config.Source {
 }
 
 // resolveSourceKey applies api_key_env and decrypts enc:v1: ciphertext.
+// Config values are already unsealed at startup (Config.NormalizeSecretsForRun),
+// so the decrypt branch is the belt-and-braces path for a source that arrived
+// already sealed through another route. A failure there leaves the ciphertext in
+// place, which makes the upstream reject the key loudly rather than sending an
+// empty Authorization header that might look like a config-less source.
 func (c *Core) resolveSourceKey(s config.Source) config.Source {
 	if s.APIKeyEnv != "" {
 		if v := os.Getenv(s.APIKeyEnv); v != "" {
@@ -447,7 +611,11 @@ func (c *Core) resolveSourceKey(s config.Source) config.Source {
 		return s
 	}
 	if box := c.store.SecretBox(); box != nil && strings.HasPrefix(s.APIKey, "enc:v1:") {
-		s.APIKey = box.MustDecrypt(s.APIKey)
+		if v, err := box.Decrypt(s.APIKey); err == nil {
+			s.APIKey = v
+		} else {
+			log.Printf("[core] source %s: api_key decrypt failed: %v", s.Name, err)
+		}
 	}
 	return s
 }
@@ -470,26 +638,64 @@ func (c *Core) rebuildRegistry() error {
 	}
 	c.buildAutoChain()
 	c.buildAutoImageChain()
+	c.rebuildKeyAutoChains()
+	// A source edit can add or remove an upstream host, which changes what a
+	// plugin is allowed to reach. Keeping this beside the chain rebuilds means
+	// one call site owns "config changed, refresh derived state".
+	c.plugins.SetOutboundHosts(sourceHosts(c.cfg))
 	return nil
 }
 
+// sourceHosts extracts the host of every configured source's base_url — the
+// allowlist for plugin outbound HTTP. Unparseable URLs are skipped rather than
+// failing startup: a malformed base_url is already rejected when the provider is
+// built, and it must not take plugins down with it.
+func sourceHosts(cfg *config.Config) []string {
+	if cfg == nil {
+		return nil
+	}
+	out := make([]string, 0, len(cfg.Sources))
+	for _, s := range cfg.Sources {
+		if s.BaseURL == "" {
+			continue
+		}
+		u, err := url.Parse(s.BaseURL)
+		if err != nil || u.Hostname() == "" {
+			continue
+		}
+		out = append(out, u.Hostname())
+	}
+	return out
+}
+
 // buildAutoChain rebuilds the AUTO chain snapshot from config.yaml rules
 // against the current providers. Slot model ids are normalized to the exact
 // configured spelling (case-insensitive match), otherwise ModelFor would fall
 // back to the source's best chat model and cooldown/quota bookkeeping would
 // key on a name that never matches.
 func (c *Core) buildAutoChain() {
+	c.autoChain.Store(c.chainForRules(c.cfg.Auto, false))
+}
+
+// chainForRules compiles a chat AUTO chain from a rule list. kindFilter is
+// false for the chat chain (image models are skipped) and true for the image
+// chain (only image models participate). Per-key chains always compile as
+// chat chains: a user's key configures the models *they* talk to.
+func (c *Core) chainForRules(rules []config.ModelScope, imageOnly bool) *scheduler.Chain {
 	prov := func(model, source string) scheduler.Provider {
 		p := c.registry.ProviderForSlot(model, source)
 		if p == nil {
 			return nil
 		}
-		if m := p.ModelByID(model); m != nil && m.Kind == "image" {
+		m := p.ModelByID(model)
+		if imageOnly && (m == nil || m.Kind != "image") {
+			return nil
+		}
+		if !imageOnly && m != nil && m.Kind == "image" {
 			return nil
 		}
 		return p
 	}
-	rules := c.cfg.Auto
 	sr := make([]scheduler.Rule, 0, len(rules))
 	for _, e := range rules {
 		model, source := e.Model, e.Source
@@ -498,7 +704,11 @@ func (c *Core) buildAutoChain() {
 				model = exact
 			}
 			if m := p.ModelByID(model); m != nil && m.Kind == "image" {
-				continue // image-kind models never join the chat AUTO chain
+				if !imageOnly {
+					continue // image-kind models never join the chat AUTO chain
+				}
+			} else if imageOnly {
+				continue
 			}
 			if source == "" {
 				source = p.Name()
@@ -513,7 +723,7 @@ func (c *Core) buildAutoChain() {
 			Hours:  e.Hours,
 		})
 	}
-	c.autoChain.Store(scheduler.BuildChain(sr, prov))
+	return scheduler.BuildChain(sr, prov)
 }
 
 // buildAutoImageChain rebuilds the image-generation AUTO chain snapshot.
@@ -655,7 +865,7 @@ func (c *Core) AddSource(src config.Source) error {
 	if err := normalizeSource(&src); err != nil {
 		return err
 	}
-	if err := config.UpsertSourceInYAML(c.cfg.Path, src.Name, src); err != nil {
+	if err := config.UpsertSourceInYAML(c.cfg.Path, src.Name, src, c.cfg.SecretBox()); err != nil {
 		return err
 	}
 	// Update in-memory Sources so mergedSources() finds the entry.
@@ -752,7 +962,88 @@ func normalizeSource(s *config.Source) error {
 
 // Close releases resources.
 func (c *Core) Close() {
+	// Plugin state must be flushed BEFORE the VM stops. The saver's final write
+	// reads each plugin's Lua tables; once vm.Stop() has closed those states the
+	// read finds nothing and the last interval of accumulation is lost — which
+	// is the exact failure this persistence was added to prevent.
+	if c.plugins != nil {
+		c.plugins.Close()
+	}
 	if c.vm != nil {
 		c.vm.Stop()
 	}
 }
+
+// applyBillingDSL compiles the configured billing profile and injects it into
+// the billing plugin.
+//
+// Configured pricing replaces the hand-written JSON blob that used to be PUT
+// through the state API. That path was used twice in production and both times
+// it "succeeded" while being wrong: once without the required `prices` wrapper
+// (silently wiping the accumulated totals), once with a peak-window shape the
+// plugin does not read (peak traffic billed at off-peak rates, no error). A
+// typed config with load-time validation turns both mistakes into startup
+// errors naming the file and field.
+//
+// A DSL with no profile, or one that fails to compile, leaves the plugin's
+// prices untouched: billing keeps running on whatever it had, and the problem
+// is logged rather than becoming a startup failure — the gateway must forward
+// even when its cost accounting is misconfigured.
+func (c *Core) applyBillingDSL() {
+	ps := c.plugins
+	if ps == nil || ps.Count() == 0 {
+		return
+	}
+	// The billing plugin may simply not be installed; a DSL without it is a
+	// no-op, not an error.
+	var have bool
+	for _, row := range ps.List() {
+		if row["name"] == "billing" && row["loaded"] == true {
+			have = true
+			break
+		}
+	}
+	if !have {
+		return
+	}
+	dsl := c.cfg.BillingDSL
+	if dsl == nil || len(dsl.Profiles) == 0 {
+		return
+	}
+	if err := dsl.Validate(); err != nil {
+		log.Printf("[billing] config invalid, profile not applied: %v", err)
+		return
+	}
+	profile := dsl.Resolve(dsl.Active)
+	if profile == nil {
+		log.Printf("[billing] no billing profile resolved (active=%q)", dsl.Active)
+		return
+	}
+	prices, err := billing.Compile(profile, c.cfg.Sources)
+	if err != nil {
+		log.Printf("[billing] profile %q not applied: %v", profile.ID, err)
+		return
+	}
+	if err := ps.SetState("billing", map[string]interface{}{"prices": prices}); err != nil {
+		log.Printf("[billing] applying profile %q: %v", profile.ID, err)
+		return
+	}
+	log.Printf("[billing] profile %q applied (%d rules, %d sources)", profile.ID, len(profile.Rules), len(c.cfg.Sources))
+}
+
+// ValidateScopeQuotas checks every scope entry's caps before they are stored.
+// A typo in a period must be rejected at write time rather than silently
+// becoming a never-resetting budget — the opposite of what was typed.
+func ValidateScopeQuotas(entries []config.ModelScope) error {
+	for _, e := range entries {
+		if err := (config.KeyQuota{
+			TokenQuota: e.TokenQuota,
+			ReqQuota:   e.ReqQuota,
+			Period:     e.Period,
+			Hours:      e.Hours,
+		}).Validate(); err != nil {
+			return fmt.Errorf("model %q: %w", e.Model, err)
+		}
+	}
+	return nil
+}

internal/core/key_auto_concurrency_test.go

@@ -0,0 +1,189 @@
+package core
+
+import (
+	"fmt"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"llmsproxy/internal/config"
+)
+
+// AutoChainFor sits on the request hot path and reads keyAutoChains with no
+// lock, while SaveKeyAuto / rebuildRegistry replace that map from the admin
+// path. A "call it twice" test never interleaves those, so it would pass even
+// if the map were mutated in place. These tests run both sides at once under
+// -race and assert what readers actually observe.
+//
+// The property: a reader sees a chain that is either the key's own or the
+// global one — never a torn or empty one. If rebuildKeyAutoChains built the
+// map in place instead of swapping a fresh pointer, readers would sometimes
+// see a partially-filled map and silently fall back to the global chain, i.e.
+// a user quietly losing their configured chain under load.
+
+// concurrentChainConfig builds N keys, half with their own chain pointing at a
+// distinct model so a fallback to the global chain is distinguishable.
+func concurrentChainConfig(t *testing.T, n int) *config.Config {
+	t.Helper()
+	cfg := perKeyTestConfig(t)
+	cfg.Keys = nil
+	cfg.GatewayKeys = []string{"sk-gw-admin"}
+	cfg.Auto = []config.ModelScope{{Model: "gpt-4o", Source: "s1", Tier: 1}}
+	for i := 0; i < n; i++ {
+		k := config.GWKey{
+			Key:  fmt.Sprintf("sk-gw-k%03d", i),
+			Role: "user",
+			Name: fmt.Sprintf("user%d", i),
+		}
+		if i%2 == 0 {
+			k.Auto = []config.ModelScope{{Model: "gpt-4o-mini", Source: "s1", Tier: 1}}
+		}
+		cfg.Keys = append(cfg.Keys, k)
+	}
+	return cfg
+}
+
+func TestAutoChainForConcurrentWithSaveKeyAuto(t *testing.T) {
+	cfg := concurrentChainConfig(t, 40)
+	c := newTestCore(t, cfg)
+
+	var stop atomic.Bool
+	var wg sync.WaitGroup
+	var bad atomic.Int64
+
+	for r := 0; r < 8; r++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for !stop.Load() {
+				for i := 0; i < 40; i++ {
+					key := fmt.Sprintf("sk-gw-k%03d", i)
+					ch, ok := c.AutoChainFor(key)
+					if !ok || ch == nil || len(ch.Tiers) == 0 {
+						bad.Add(1)
+						continue
+					}
+					// Legal values are this key's own chain model (either
+					// side of the writer's flip) or the global chain's. The
+					// assertion is about the legal set, not a fixed value:
+					// the writers legitimately change the chain underneath.
+					got := ch.Tiers[0].Slots[0].Model
+					if got != "gpt-4o" && got != "gpt-4o-mini" {
+						bad.Add(1)
+					}
+				}
+			}
+		}()
+	}
+
+	for w := 0; w < 2; w++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			model := "gpt-4o-mini"
+			for i := 0; i < 25 && !stop.Load(); i++ {
+				for j := 0; j < 40; j += 2 {
+					if err := c.SaveKeyAuto(fmt.Sprintf("sk-gw-k%03d", j), []config.ModelScope{
+						{Model: model, Source: "s1", Tier: 1},
+					}); err != nil {
+						bad.Add(1)
+						return
+					}
+				}
+				if model == "gpt-4o-mini" {
+					model = "gpt-4o"
+				} else {
+					model = "gpt-4o-mini"
+				}
+			}
+		}()
+	}
+
+	time.Sleep(400 * time.Millisecond)
+	stop.Store(true)
+	wg.Wait()
+
+	if n := bad.Load(); n > 0 {
+		t.Fatalf("%d reads saw a chain outside {own, global} — the cache is "+
+			"being swapped in place instead of atomically", n)
+	}
+}
+
+// The registry rebuild path (AddSource / RemoveSource) also replaces the
+// cache, so readers must keep a usable chain throughout.
+//
+// Churned sources use a "churn" prefix on purpose: deleting a source a chain
+// references legitimately empties that chain (BuildChain drops unresolvable
+// slots), and asserting otherwise would be testing a fiction rather than the
+// code.
+func TestAutoChainForConcurrentWithSourceChanges(t *testing.T) {
+	cfg := concurrentChainConfig(t, 20)
+	c := newTestCore(t, cfg)
+
+	var stop atomic.Bool
+	var wg sync.WaitGroup
+	var bad atomic.Int64
+
+	for r := 0; r < 6; r++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for !stop.Load() {
+				for i := 0; i < 20; i++ {
+					ch, ok := c.AutoChainFor(fmt.Sprintf("sk-gw-k%03d", i))
+					if !ok || ch == nil || len(ch.Tiers) == 0 {
+						bad.Add(1)
+						continue
+					}
+					if got := ch.Tiers[0].Slots[0].Model; got != "gpt-4o" && got != "gpt-4o-mini" {
+						bad.Add(1)
+					}
+				}
+			}
+		}()
+	}
+
+	wg.Add(1)
+	go func() {
+		defer wg.Done()
+		for n := 0; n < 30 && !stop.Load(); n++ {
+			name := fmt.Sprintf("churn%d", n)
+			_ = c.AddSource(config.Source{
+				Name: name, BaseURL: "http://127.0.0.1:9/v1", Adapter: "openai",
+				Models: []config.Model{{ID: "m-" + name, Priority: 10}}},
+			)
+			_ = c.RemoveSource(name)
+		}
+	}()
+
+	time.Sleep(500 * time.Millisecond)
+	stop.Store(true)
+	wg.Wait()
+
+	if n := bad.Load(); n > 0 {
+		t.Fatalf("%d reads lost the chain while unrelated sources were churned", n)
+	}
+}
+
+// Many distinct keys: the lookup must stay a map hit and must not degrade into
+// returning the global chain as the key count grows.
+func TestAutoChainForManyKeys(t *testing.T) {
+	cfg := concurrentChainConfig(t, 500)
+	c := newTestCore(t, cfg)
+
+	for i := 0; i < 500; i++ {
+		key := fmt.Sprintf("sk-gw-k%03d", i)
+		ch, _ := c.AutoChainFor(key)
+		if ch == nil || len(ch.Tiers) == 0 {
+			t.Fatalf("key %s lost its chain with 500 keys configured", key)
+		}
+		want := "gpt-4o"
+		if i%2 == 0 {
+			want = "gpt-4o-mini"
+		}
+		if got := ch.Tiers[0].Slots[0].Model; got != want {
+			t.Fatalf("key %s got %q, want %q", key, got, want)
+		}
+	}
+}

internal/core/key_auto_test.go

@@ -0,0 +1,221 @@
+package core
+
+import (
+	"path/filepath"
+	"testing"
+
+	"llmsproxy/internal/config"
+	"llmsproxy/internal/scheduler"
+)
+
+// Per-key AUTO chains let an admin give one user their own scheduling chain
+// while every other key keeps the global one. These tests pin the three
+// properties that make that safe:
+//
+//  1. a key WITH its own chain gets it;
+//  2. a key WITHOUT one inherits the global chain (backward compatible);
+//  3. the override is actually persisted, so it survives a restart.
+//
+// A test that only checked (1) would pass even if AutoChainFor ignored the key
+// and always returned the global chain whenever the two happened to be equal,
+// so (2) uses deliberately *different* chains.
+
+func perKeyTestConfig(t *testing.T) *config.Config {
+	t.Helper()
+	dir := t.TempDir()
+	return &config.Config{
+		Path:         filepath.Join(dir, "config.yaml"),
+		AdapterDir:   filepath.Join(dir, "adapters"),
+		RuntimeFile:  filepath.Join(dir, "runtime.json"),
+		Listen:       "127.0.0.1:0",
+		DefaultModel: "AUTO",
+		GatewayKeys:  []string{"sk-gw-admin"},
+		Sources: []config.Source{
+			{Name: "s1", BaseURL: "http://127.0.0.1:1/v1", Adapter: "openai",
+				Models: []config.Model{{ID: "gpt-4o", Priority: 10}, {ID: "gpt-4o-mini", Priority: 5}}},
+		},
+		// Global chain points at gpt-4o.
+		Auto: []config.ModelScope{{Model: "gpt-4o", Source: "s1", Tier: 1}},
+		Keys: []config.GWKey{
+			{Key: "sk-gw-admin", Role: "admin"},
+			{Key: "sk-gw-inherits", Role: "user", Name: "inherits"},
+			// This key's own chain points at a DIFFERENT model, so returning
+			// the global chain by mistake is detectable.
+			{Key: "sk-gw-own", Role: "user", Name: "own",
+				Auto: []config.ModelScope{{Model: "gpt-4o-mini", Source: "s1", Tier: 1}}},
+		},
+	}
+}
+
+func firstSlotModel(t *testing.T, c *Core, key string) string {
+	t.Helper()
+	ch, _ := c.AutoChainFor(key)
+	if ch == nil || len(ch.Tiers) == 0 {
+		t.Fatalf("key %s: no chain tiers", key)
+	}
+	slots := ch.Tiers[0].Slots
+	if len(slots) == 0 {
+		t.Fatalf("key %s: tier 1 has no slots", key)
+	}
+	return slots[0].Model
+}
+
+// A key that declares its own chain must be scheduled by it, not the global.
+func TestAutoChainForUsesKeyOwnChain(t *testing.T) {
+	c := newTestCore(t, perKeyTestConfig(t))
+
+	if got := firstSlotModel(t, c, "sk-gw-own"); got != "gpt-4o-mini" {
+		t.Fatalf("key with own chain must use it, got %q (want gpt-4o-mini)", got)
+	}
+	// Sanity: the global chain really is different, so the assertion above
+	// cannot pass by accident.
+	if got := firstSlotModel(t, c, "sk-gw-admin"); got != "gpt-4o" {
+		t.Fatalf("admin must use global chain, got %q (want gpt-4o)", got)
+	}
+}
+
+// A key with no own chain inherits the global one. This is what keeps every
+// pre-existing deployment working with no config change.
+func TestAutoChainForInheritsGlobalWhenUnset(t *testing.T) {
+	c := newTestCore(t, perKeyTestConfig(t))
+
+	if got := firstSlotModel(t, c, "sk-gw-inherits"); got != "gpt-4o" {
+		t.Fatalf("key without own chain must inherit global, got %q (want gpt-4o)", got)
+	}
+}
+
+// A key that is not in the key table at all (e.g. a seed key) must also
+// inherit the global chain rather than erroring or returning nil.
+func TestAutoChainForUnknownKeyInheritsGlobal(t *testing.T) {
+	c := newTestCore(t, perKeyTestConfig(t))
+
+	ch, ok := c.AutoChainFor("sk-gw-never-configured")
+	if !ok || ch == nil || len(ch.Tiers) == 0 {
+		t.Fatalf("unknown key must fall back to global chain, got ok=%v chain=%v", ok, ch)
+	}
+}
+
+// The per-key chain must survive a save/reload round trip: an admin configuring
+// a chain in the WebUI and restarting the gateway must not silently lose it.
+func TestSaveKeyAutoPersistsAndApplies(t *testing.T) {
+	cfg := perKeyTestConfig(t)
+	c := newTestCore(t, cfg)
+
+	if err := c.SaveKeyAuto("sk-gw-inherits", []config.ModelScope{
+		{Model: "gpt-4o-mini", Source: "s1", Tier: 1},
+	}); err != nil {
+		t.Fatalf("SaveKeyAuto: %v", err)
+	}
+	if got := firstSlotModel(t, c, "sk-gw-inherits"); got != "gpt-4o-mini" {
+		t.Fatalf("saved chain must apply immediately, got %q", got)
+	}
+
+	// Reload from disk exactly like a restart does.
+	reloaded, err := config.Load(cfg.Path)
+	if err != nil {
+		t.Fatalf("reload config: %v", err)
+	}
+	c2 := newTestCore(t, reloaded)
+	if got := firstSlotModel(t, c2, "sk-gw-inherits"); got != "gpt-4o-mini" {
+		t.Fatalf("per-key chain must survive restart, got %q", got)
+	}
+	// The other key must be unaffected by its neighbour's override.
+	if got := firstSlotModel(t, c2, "sk-gw-admin"); got != "gpt-4o" {
+		t.Fatalf("override must not leak to other keys, got %q", got)
+	}
+}
+
+// Clearing the override must return the key to the global chain, both in
+// memory and after a restart. This is what the WebUI's "use global" toggle
+// does, so a stale shadowing entry would silently pin the user to their old
+// chain forever.
+func TestSaveKeyAutoClearRestoresGlobal(t *testing.T) {
+	cfg := perKeyTestConfig(t)
+	c := newTestCore(t, cfg)
+
+	if err := c.SaveKeyAuto("sk-gw-own", nil); err != nil {
+		t.Fatalf("clear: %v", err)
+	}
+	if got := firstSlotModel(t, c, "sk-gw-own"); got != "gpt-4o" {
+		t.Fatalf("cleared key must inherit global again, got %q", got)
+	}
+
+	reloaded, err := config.Load(cfg.Path)
+	if err != nil {
+		t.Fatalf("reload: %v", err)
+	}
+	c2 := newTestCore(t, reloaded)
+	if got := firstSlotModel(t, c2, "sk-gw-own"); got != "gpt-4o" {
+		t.Fatalf("clear must survive restart, got %q", got)
+	}
+}
+
+// Saving an unknown key must fail loudly rather than creating a shadow entry
+// that no request can ever match.
+func TestSaveKeyAutoUnknownKeyFails(t *testing.T) {
+	c := newTestCore(t, perKeyTestConfig(t))
+
+	if err := c.SaveKeyAuto("sk-gw-nope", []config.ModelScope{{Model: "gpt-4o", Source: "s1"}}); err == nil {
+		t.Fatal("SaveKeyAuto on unknown key must return an error")
+	}
+}
+
+// Per-key quota validation must run, exactly like the model scope path: an
+// override is another place an operator can write a bad budget.
+func TestSaveKeyAutoRejectsBadQuota(t *testing.T) {
+	c := newTestCore(t, perKeyTestConfig(t))
+
+	err := c.SaveKeyAuto("sk-gw-inherits", []config.ModelScope{
+		{Model: "gpt-4o", Source: "s1", TokenQuota: -5},
+	})
+	if err == nil {
+		t.Fatal("negative token quota in a per-key auto chain must be rejected")
+	}
+}
+
+// Source changes must recompile per-key chains. A key whose chain names a
+// model that does not exist yet compiles to an empty chain (BuildChain drops
+// slots it cannot resolve — sending the request into a black hole would be
+// worse), and must start working as soon as the source appears.
+func TestKeyAutoChainsRebuildOnSourceChange(t *testing.T) {
+	cfg := perKeyTestConfig(t)
+	cfg.Keys = append(cfg.Keys, config.GWKey{
+		Key: "sk-gw-late", Role: "user", Name: "late",
+		Auto: []config.ModelScope{{Model: "new-model", Source: "s2", Tier: 1}},
+	})
+	c := newTestCore(t, cfg)
+
+	// s2 does not exist yet, so the chain has no usable slot.
+	if ch, _ := c.AutoChainFor("sk-gw-late"); chainSlotsOf(ch) != 0 {
+		t.Fatalf("chain naming an unknown model must compile empty, got %d slots",
+			chainSlotsOf(ch))
+	}
+	// The other keys are unaffected by the broken one.
+	if got := firstSlotModel(t, c, "sk-gw-admin"); got != "gpt-4o" {
+		t.Fatalf("broken per-key chain must not affect others, got %q", got)
+	}
+
+	// Add the source; the per-key chain must pick it up without a restart.
+	cfg.Sources = append(cfg.Sources, config.Source{
+		Name: "s2", BaseURL: "http://127.0.0.1:2/v1", Adapter: "openai",
+		Models: []config.Model{{ID: "new-model", Priority: 10}},
+	})
+	if err := c.Reload(); err != nil {
+		t.Fatalf("Reload: %v", err)
+	}
+	if got := firstSlotModel(t, c, "sk-gw-late"); got != "new-model" {
+		t.Fatalf("per-key chain must resolve after the source is added, got %q", got)
+	}
+}
+
+// chainSlotsOf counts usable slots, tolerating a nil chain.
+func chainSlotsOf(ch *scheduler.Chain) int {
+	if ch == nil {
+		return 0
+	}
+	n := 0
+	for _, tn := range ch.Tiers {
+		n += len(tn.Slots)
+	}
+	return n
+}

internal/core/startup_secrets_test.go

@@ -0,0 +1,178 @@
+package core
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+)
+
+// Startup must be idempotent, and every credential consumer must see PLAINTEXT.
+//
+// Two bugs shared one root cause: NewFromConfig unsealed the config at the END,
+// after the steps that read credentials had already run.
+//
+//  1. seedKeys() deduped cfg.Keys[i].Key against the plaintext gateway_keys
+//     entries. With ciphertext keys the comparison never matched, so every
+//     restart appended another copy of the same admin key — production reached
+//     four identical admin keys.
+//  2. rebuildRegistry() handed cfg.Sources[i].APIKey to provider.New, so with
+//     ciphertext it built every upstream client with "enc:v1:..." as its bearer
+//     token and every upstream call would 401.
+//
+// They masked each other: seedKeys' Save() unsealed memory as a side effect, so
+// bug 2 was invisible until the redundant saves were removed. These tests drive
+// the real entry point (core.New -> config.Load off disk), because building a
+// fresh in-memory config per attempt hides both bugs — that is exactly how the
+// first draft of this test failed to reproduce anything.
+
+// writeSealedInstall creates a config on disk the way a real install looks
+// after its first run: credentials already sealed, gateway_keys still plaintext.
+func writeSealedInstall(t *testing.T, dir string) string {
+	t.Helper()
+	path := filepath.Join(dir, "config.yaml")
+	cfg := &config.Config{
+		Path:         path,
+		AdapterDir:   filepath.Join(dir, "adapters"),
+		RuntimeFile:  filepath.Join(dir, "runtime.json"),
+		Listen:       "127.0.0.1:0",
+		DefaultModel: "AUTO",
+		GatewayKeys:  []string{"sk-gw-seed-me"},
+		Sources: []config.Source{{
+			Name: "up", BaseURL: "http://127.0.0.1:1/v1", APIKey: "sk-upstream-secret",
+			Adapter: "openai", Models: []config.Model{{ID: "gpt-4o", Priority: 10}},
+		}},
+	}
+	c, err := NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("initial install: %v", err)
+	}
+	c.Close()
+
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(raw), "enc:v1:") {
+		t.Fatalf("fixture is wrong: on-disk config holds no sealed credential, "+
+			"so this test cannot detect ciphertext leaking into consumers:\n%s", raw)
+	}
+	return path
+}
+
+// writeSealedInstallNoSeed is writeSealedInstall without gateway_keys, so no
+// key is ever seeded and no startup Save() can unseal the config as a side
+// effect. That isolation is what makes the ciphertext leak observable.
+func writeSealedInstallNoSeed(t *testing.T, dir string) string {
+	t.Helper()
+	path := filepath.Join(dir, "config.yaml")
+	cfg := &config.Config{
+		Path:         path,
+		AdapterDir:   filepath.Join(dir, "adapters"),
+		RuntimeFile:  filepath.Join(dir, "runtime.json"),
+		Listen:       "127.0.0.1:0",
+		DefaultModel: "AUTO",
+		Sources: []config.Source{{
+			Name: "up", BaseURL: "http://127.0.0.1:1/v1", APIKey: "sk-upstream-secret",
+			Adapter: "openai", Models: []config.Model{{ID: "gpt-4o", Priority: 10}},
+		}},
+	}
+	c, err := NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("initial install: %v", err)
+	}
+	c.Close()
+	raw, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(raw), "enc:v1:") {
+		t.Fatalf("fixture is wrong: no sealed credential on disk:\n%s", raw)
+	}
+	return path
+}
+
+// countKeysWithSecret loads the config, unseals it, and counts keys whose
+// plaintext equals one of the gateway_keys entries.
+func countKeysWithSecret(t *testing.T, path string) (total, matching int) {
+	t.Helper()
+	cfg, err := config.Load(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	box, err := config.NewSecretBox(cfg.RuntimeFile)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if _, err := cfg.UnsealSecrets(box); err != nil {
+		t.Fatal(err)
+	}
+	want := map[string]bool{}
+	for _, g := range cfg.GatewayKeys {
+		want[g] = true
+	}
+	for _, k := range cfg.Keys {
+		total++
+		if want[k.Key] {
+			matching++
+		}
+	}
+	return total, matching
+}
+
+// Restarting must not grow the key list. Before the fix this went 2 -> 3 -> 4.
+func TestRestartDoesNotDuplicateSeededKeys(t *testing.T) {
+	dir := t.TempDir()
+	path := writeSealedInstall(t, dir)
+
+	if _, n := countKeysWithSecret(t, path); n != 1 {
+		t.Fatalf("after install: %d keys carry the seeded secret, want 1", n)
+	}
+
+	for i := 2; i <= 4; i++ {
+		c, err := New(path) // production path: reads the file
+		if err != nil {
+			t.Fatalf("restart %d: %v", i, err)
+		}
+		c.Close()
+		total, n := countKeysWithSecret(t, path)
+		if n != 1 {
+			t.Fatalf("restart %d: %d keys carry the seeded secret (total %d), want "+
+				"exactly 1 — seedKeys is comparing ciphertext against plaintext and "+
+				"appending a duplicate on every start", i, n, total)
+		}
+	}
+}
+
+// Every provider must hold the real upstream credential, not its ciphertext.
+//
+// This deliberately runs with gateway_keys EMPTY. The buggy order was masked
+// whenever seedKeys had work to do, because its Save() unsealed memory as a
+// side effect — providers then happened to see plaintext. With no key to seed,
+// no Save runs, and the ciphertext reaches the registry directly. An earlier
+// version of this test kept gateway_keys populated and was insensitive: it
+// passed even with the bug present.
+func TestProvidersGetPlaintextCredentials(t *testing.T) {
+	dir := t.TempDir()
+	path := writeSealedInstallNoSeed(t, dir)
+
+	c, err := New(path)
+	if err != nil {
+		t.Fatalf("restart: %v", err)
+	}
+	defer c.Close()
+
+	for _, p := range c.registry.Providers() {
+		key := p.Config().APIKey
+		if strings.HasPrefix(key, "enc:v1:") {
+			t.Errorf("provider %q holds CIPHERTEXT api_key %q — every upstream call "+
+				"would 401", p.Name(), key)
+		}
+		if key != "sk-upstream-secret" {
+			t.Errorf("provider %q api_key = %q, want the plaintext upstream secret",
+				p.Name(), key)
+		}
+	}
+}

internal/gateway/api.go

@@ -3,6 +3,7 @@ package gateway
 import (
 	"encoding/csv"
 	"encoding/json"
+	"fmt"
 	"io"
 	"log"
 	"net/http"
@@ -78,6 +79,177 @@ type sourcePayload struct {
 	MaxTokens     int                    `json:"max_tokens"`
 	MaxConcurrent int                    `json:"max_concurrent"`
 	RPM           int                    `json:"rpm"` // optional requests-per-minute cap, 0 = unlimited
+	// ModelIDs is an additive-only shorthand: when set, the named models are
+	// merged into an existing source instead of Models replacing the list.
+	// It exists because "add one model" is the most common scripted edit and a
+	// full Models list cannot be written without reading the source first.
+	ModelIDs []string `json:"model_ids,omitempty"`
+
+	// The four fields below are POINTERS, and that is the whole point.
+	//
+	// POST/PUT is an upsert that rewrites the whole source, so any field the
+	// payload cannot express is silently reset to its zero value. That already
+	// destroyed api_key once (fixed with resolveAPIKey) and would equally
+	// destroy proxy_url, api_key_env, timeout and queue_timeout.
+	//
+	// A plain string/duration cannot distinguish "the caller did not mention
+	// this field" from "the caller asked for the empty value", and only the
+	// former may inherit. So:
+	//
+	//	nil   → field absent from the request: keep the current value
+	//	&"x"  → present: store exactly "x" (including "" to clear it)
+	//
+	// api_key deliberately keeps its older "empty means inherit" rule rather
+	// than gaining a pointer: that rule is already published (v1.7.6) and
+	// scripts depend on it. Changing it now would let a script that echoes an
+	// empty api_key erase a live credential, which is the exact failure this
+	// whole area exists to prevent.
+	ProxyURL     *string `json:"proxy_url,omitempty"`
+	APIKeyEnv    *string `json:"api_key_env,omitempty"`
+	Timeout      *string `json:"timeout,omitempty"`       // duration string, e.g. "300s"
+	QueueTimeout *string `json:"queue_timeout,omitempty"` // duration string, e.g. "90s"
+}
+
+// optionalSourceFields carries the Source fields a partial payload must not
+// reset, each as a pointer so "absent" and "explicitly empty" stay distinct.
+//
+//	nil  → absent from the request: keep the current value
+//	&""  → present: clear it
+//	&"x" → present: store "x"
+//
+// The alternative — a plain value with "empty means inherit" — cannot express
+// "clear this field", so emptying the proxy input in the UI would silently keep
+// the old proxy. It is exactly why api_key is NOT modelled this way: for a
+// credential, failing to keep the old value is worse than failing to clear it,
+// and that rule is already published (v1.7.6). Two fields with opposite failure
+// modes therefore get opposite rules, and both are spelled out here.
+type optionalSourceFields struct {
+	ProxyURL     *string
+	APIKeyEnv    *string
+	Timeout      *string
+	QueueTimeout *string
+}
+
+// resolve overlays the four optional fields onto payload.
+//
+// A nil pointer means the request did not mention the field, so it keeps the
+// value already stored in cur (nil cur = a source being created, where the
+// payload's own zero value is correct). A non-nil pointer always wins, including
+// when it points at the empty string, which is how a UI form clears a field.
+//
+// Only these four fields are overlaid. Everything else in payload is taken as
+// sent: an upsert that inherited the whole record would make it impossible to
+// change anything.
+func (o optionalSourceFields) resolve(payload config.Source, cur *config.Source) (config.Source, error) {
+	// cur is dereferenced exactly once here so the create path (nil) cannot
+	// panic on the per-field lookups below.
+	var storedProxyURL, storedKeyEnv string
+	var storedTimeout, storedQueueTimeout time.Duration
+	if cur != nil {
+		storedProxyURL, storedKeyEnv = cur.ProxyURL, cur.APIKeyEnv
+		storedTimeout, storedQueueTimeout = cur.Timeout, cur.QueueTimeout
+	}
+	pick := func(p *string, stored string) string {
+		if p != nil {
+			return *p
+		}
+		return stored
+	}
+	payload.ProxyURL = pick(o.ProxyURL, storedProxyURL)
+	payload.APIKeyEnv = pick(o.APIKeyEnv, storedKeyEnv)
+	if err := overlayDuration(&payload.Timeout, o.Timeout, "timeout", storedTimeout); err != nil {
+		return payload, err
+	}
+	if err := overlayDuration(&payload.QueueTimeout, o.QueueTimeout, "queue_timeout", storedQueueTimeout); err != nil {
+		return payload, err
+	}
+	return payload, nil
+}
+
+// overlayDuration applies the absent/present rule to one duration field: a nil
+// pointer keeps stored, a set pointer replaces it (with "" / "0s" clearing).
+func overlayDuration(dst *time.Duration, p *string, field string, stored time.Duration) error {
+	if p == nil {
+		*dst = stored
+		return nil
+	}
+	d, err := parseOptionalDuration(*p, field)
+	if err != nil {
+		return err
+	}
+	*dst = d
+	return nil
+}
+
+// parseOptionalDuration parses a duration written as a Go duration string.
+// An empty string clears the field back to the config default (0), which is
+// what a UI form submitting an empty timeout box should mean. Parsing happens
+// only for fields the request actually mentions, so a typo can never surface
+// as a silent reset of something else.
+func parseOptionalDuration(v, field string) (time.Duration, error) {
+	v = strings.TrimSpace(v)
+	if v == "" {
+		return 0, nil
+	}
+	d, err := time.ParseDuration(v)
+	if err != nil {
+		return 0, fmt.Errorf("%s %q is not a duration (use e.g. 120s, 5m, 1h)", field, v)
+	}
+	if d < 0 {
+		return 0, fmt.Errorf("%s must be >= 0 (got %s)", field, v)
+	}
+	return d, nil
+}
+
+// keepExistingAPIKey is the mask a client sends when it means "keep the
+// current credential". POST /api/sources is an upsert, so a script that edits
+// one field and echoes a placeholder back would otherwise persist that
+// placeholder as the live api_key — which fails closed as 401 on the next
+// request, long after the script exited successfully. See TestSourceUpsertKeepsAPIKey.
+const keepExistingAPIKey = "__KEEP__"
+
+// resolveAPIKey returns the credential to store, inheriting the current one
+// when the payload does not supply a new one.
+//
+// Two distinct "no new credential" signals exist and they must not collapse:
+//   - the explicit mask keepExistingAPIKey — always inherit.
+//   - the empty string — inherit only for a source that already has a
+//     credential. Empty is a legitimate value for a *new* credential-less
+//     source (dshcodebuddy uses "dsh-local-noauth", local servers use ""),
+//     so a fresh source with an empty key must stay empty rather than inherit
+//     nothing and fail differently.
+func (g *Gateway) resolveAPIKey(name, want string, exists bool) string {
+	if want == keepExistingAPIKey {
+		return g.currentAPIKey(name)
+	}
+	if exists && want == "" {
+		return g.currentAPIKey(name)
+	}
+	return want
+}
+
+func (g *Gateway) currentAPIKey(name string) string {
+	if s := g.sourceByName(name); s != nil {
+		return s.APIKey
+	}
+	return ""
+}
+
+// sourceByName returns a live view of one source, or nil when it does not
+// exist. It goes through core.Sources() so the key is already unsealed and
+// defaults applied, matching what the routes actually use.
+func (g *Gateway) sourceByName(name string) *config.Source {
+	for _, s := range g.core.Sources() {
+		if s.Name == name {
+			ss := s
+			return &ss
+		}
+	}
+	return nil
+}
+
+func (g *Gateway) sourceExists(name string) bool {
+	return g.sourceByName(name) != nil
 }
 
 func (g *Gateway) handleSourcesAPI(w http.ResponseWriter, r *http.Request) {
@@ -91,17 +263,34 @@ func (g *Gateway) handleSourcesAPI(w http.ResponseWriter, r *http.Request) {
 	switch r.Method {
 	case http.MethodGet:
 		writeJSON(w, http.StatusOK, map[string]interface{}{"sources": g.core.Sources()})
-	case http.MethodPost:
+	case http.MethodPost, http.MethodPut:
 		var p sourcePayload
 		body, _ := io.ReadAll(r.Body)
 		if err := json.Unmarshal(body, &p); err != nil {
 			writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
 			return
 		}
+		// PUT addresses the source by path; the body name is only a fallback so
+		// a single-field edit does not have to echo the name back. A body that
+		// disagrees with the path is a caller bug, not something to silently
+		// resolve — it usually means the wrong source is about to be written.
+		if r.Method == http.MethodPut {
+			if path == "" {
+				writeError(w, http.StatusBadRequest, "invalid_request", "source name required in path")
+				return
+			}
+			if p.Name != "" && p.Name != path {
+				writeError(w, http.StatusBadRequest, "invalid_request",
+					"source name in body ("+p.Name+") does not match path ("+path+")")
+				return
+			}
+			p.Name = path
+		}
+		exists := g.sourceExists(p.Name)
 		src := config.Source{
 			Name:          p.Name,
 			BaseURL:       p.BaseURL,
-			APIKey:        p.APIKey,
+			APIKey:        g.resolveAPIKey(p.Name, p.APIKey, exists),
 			Adapter:       p.Adapter,
 			Endpoint:      p.Endpoint,
 			ImageEndpoint: p.ImageEndpoint,
@@ -113,6 +302,44 @@ func (g *Gateway) handleSourcesAPI(w http.ResponseWriter, r *http.Request) {
 			MaxConcurrent: p.MaxConcurrent,
 			RPM:           p.RPM,
 		}
+		// The four upsert-fragile fields are applied last: a nil pointer means
+		// "not in the request", so they are inherited from the stored source
+		// instead of being reset to the payload's zero value. This runs whether
+		// or not the request mentioned them, because resolve() itself decides
+		// per field — passing them unconditionally keeps the rule in one place.
+		src, err := optionalSourceFields{
+			ProxyURL:     p.ProxyURL,
+			APIKeyEnv:    p.APIKeyEnv,
+			Timeout:      p.Timeout,
+			QueueTimeout: p.QueueTimeout,
+		}.resolve(src, g.sourceByName(p.Name))
+		if err != nil {
+			writeError(w, http.StatusBadRequest, "source_error", err.Error())
+			return
+		}
+		// model_ids is additive: "add these models" is the common scripted edit
+		// and it must not require reading (and echoing) the whole list back.
+		// A request that omits models entirely is therefore a pure add, not a
+		// request to empty the list -- that distinction is the whole point.
+		if len(p.ModelIDs) > 0 {
+			if len(src.Models) == 0 {
+				if existing := g.sourceByName(p.Name); existing != nil {
+					src.Models = append(src.Models, existing.Models...)
+				}
+			}
+			have := map[string]bool{}
+			for _, m := range src.Models {
+				have[m.ID] = true
+			}
+			for _, id := range p.ModelIDs {
+				id = strings.TrimSpace(id)
+				if id == "" || have[id] {
+					continue
+				}
+				have[id] = true
+				src.Models = append(src.Models, config.Model{ID: id, Kind: "chat"})
+			}
+		}
 		if err := g.core.AddSource(src); err != nil {
 			writeError(w, http.StatusBadRequest, "source_error", err.Error())
 			return
@@ -206,6 +433,55 @@ func (g *Gateway) handleStatsAPI(w http.ResponseWriter, r *http.Request) {
 		}
 	}
 	key := exportKey(r)
+	// ?period=day|week|month|all switches the whole payload to a calendar
+	// window (UTC) aggregated off the audit files, instead of the
+	// since-process-start totals. The CSV exports below are unaffected: they
+	// take an explicit from/to range and stream, so a period selector there
+	// would only be a second way to spell the same bounds.
+	if p := periodFromQuery(r.URL.Query()); p != PeriodAll {
+		if !ValidPeriod(p) {
+			writeError(w, http.StatusBadRequest, "bad_period",
+				"period must be one of day, week, month, all")
+			return
+		}
+		out := g.stats.PeriodSnapshot(p, key, time.Now())
+		// The dashboard paints its records table from this payload's `records`
+		// field (paintRecords(st.records)), so a period view that omits the key
+		// renders an empty table no matter how much traffic the window had. The
+		// aggregate above is period-scoped; the record list is the same
+		// newest-first page the scroll handler pulls from /api/stats/records,
+		// which is deliberately not period-filtered (the cursor, not a date
+		// range, is its state) — that asymmetry is what the paging code
+		// already assumes.
+		page := g.stats.AuditPage("", limit, key)
+		writeJSON(w, http.StatusOK, map[string]interface{}{
+			"period":    out.Period,
+			"from":      out.From,
+			"total":     out.Total,
+			"by_key":    out.ByKey,
+			"by_model":  out.Models,
+			"by_source": out.Srcs,
+			"by_status": out.Status,
+			"buckets":   out.Bucket,
+			"truncated": out.Truncated,
+			// "Active requests" is a live gauge (requests in flight right now),
+			// not a period aggregate. Omitting it from the period response made
+			// the KPI read `st.active || 0` → 0 on every day/week/month view while
+			// the all-time view showed the truth, which read as "the gateway is
+			// idle" precisely when it was busiest. A calendar window cannot
+			// filter it, so it travels with every response unfiltered.
+			"active": g.stats.Active(),
+			"records":   page.Records,
+			// The paging cursor has to travel with the first page, otherwise
+			// the scroll handler has nothing to page from and the table is
+			// stuck at one screen forever.
+			"next_cursor": page.Next,
+			"has_more":    page.HasMore,
+			"rotated":     page.Rotated,
+			"key_names":   g.keyNamesFor(),
+		})
+		return
+	}
 	if r.URL.Query().Get("export") == "csv" {
 		from, _ := strconv.ParseInt(r.URL.Query().Get("from"), 10, 64)
 		to, _ := strconv.ParseInt(r.URL.Query().Get("to"), 10, 64)
@@ -313,14 +589,39 @@ func (g *Gateway) handleStatsAPI(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	snap := g.stats.Snapshot(limit, key)
-	keyNames := map[string]string{}
-	for _, k := range g.core.ListKeys() {
-		keyNames[keyID(k.Key)] = k.Name
-	}
-	snap["key_names"] = keyNames
+	snap["key_names"] = g.keyNamesFor()
+	// Records come from the audit files via the same AuditPage the scroll
+	// handler uses, not from Snapshot's in-memory ring. Two reasons:
+	//
+	//  - Ordering. The ring is oldest-first, AuditPage is newest-first. The
+	//    first screen and the paged continuation have to agree, or scrolling
+	//    splices new rows in above old ones.
+	//  - Depth. The ring holds maxRecs entries (a few hundred); the audit
+	//    files hold everything, so "all" actually means all and the cursor
+	//    can keep paging.
+	//
+	// It also carries next_cursor, without which the table can never load a
+	// second page.
+	page := g.stats.AuditPage("", limit, key)
+	snap["records"] = page.Records
+	snap["next_cursor"] = page.Next
+	snap["has_more"] = page.HasMore
+	snap["rotated"] = page.Rotated
 	writeJSON(w, http.StatusOK, snap)
 }
 
+// keyNamesFor is the masked-id -> display-name map every stats payload needs.
+// It is keyed by keyID (the mask), not the raw key, because that is what the
+// aggregate rows carry — building it in one place stops the period branch and
+// the lifetime branch from drifting apart.
+func (g *Gateway) keyNamesFor() map[string]string {
+	names := map[string]string{}
+	for _, k := range g.core.ListKeys() {
+		names[keyID(k.Key)] = k.Name
+	}
+	return names
+}
+
 // handleStatsRecordsAPI pages the request records straight off the audit files.
 // The dashboard loads only its first screen and asks for the next page as the
 // user scrolls, so neither side holds the full history: the server keeps no

internal/gateway/apiv1.go

@@ -0,0 +1,474 @@
+package gateway
+
+// Agent-facing management API.
+//
+// The Web UI has always driven the gateway through /api/*, so the management
+// surface already exists. What it lacked was anything an agent could *rely* on:
+// no way to discover the contract, no single call that answers "what is the
+// current state", and per-endpoint response shapes that make scripting brittle.
+//
+// This file adds a versioned, self-describing facade under /api/v1 without
+// touching the existing endpoints, so the Web UI keeps working unchanged while
+// agents get a stable contract:
+//
+//	GET    /api/v1                      — machine-readable index of every endpoint
+//	GET    /api/v1/overview             — one call: sources + auto chain + keys + health
+//	GET    /api/v1/sources              — sources, credentials masked
+//	GET    /api/v1/sources/{name}       — one source
+//	GET    /api/v1/auto                 — scheduling chain + live slot states
+//	GET    /api/v1/keys                 — key metadata (never the secret)
+//	GET    /api/v1/models               — every model id the gateway can route
+//	GET    /api/v1/health               — per-source health, no auth needed beyond the key
+//
+// Auth is the same gateway key as everywhere else (Authorization: Bearer, or
+// ?api_key=, or the gw_key cookie). Read endpoints accept any role; writes
+// still require admin, enforced by the same middleware the UI goes through.
+
+import (
+	"errors"
+	"net/http"
+	"net/url"
+	"sort"
+	"strings"
+
+	"llmsproxy/internal/config"
+)
+
+// apiV1Routes dispatches /api/v1/*. Kept separate from routes() so the original
+// switch stays readable and the versioned surface can grow on its own.
+func (g *Gateway) apiV1Routes(w http.ResponseWriter, r *http.Request) {
+	path := strings.TrimPrefix(r.URL.Path, "/api/v1")
+	path = strings.Trim(path, "/")
+
+	switch {
+	case path == "":
+		g.apiV1Index(w, r)
+	case path == "overview":
+		g.apiV1Overview(w, r)
+	case path == "health":
+		g.apiV1Health(w, r)
+	case path == "models":
+		g.apiV1Models(w, r)
+	case path == "sources":
+		if r.Method != http.MethodGet {
+			writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "use GET")
+			return
+		}
+		writeJSON(w, http.StatusOK, map[string]interface{}{"sources": maskSources(g.core.Sources())})
+	case strings.HasPrefix(path, "sources/"):
+		name, err := decodePathSegment(strings.TrimPrefix(path, "sources/"))
+		if err != nil {
+			writeError(w, http.StatusBadRequest, "invalid_request", err.Error())
+			return
+		}
+		if r.Method != http.MethodGet {
+			writeError(w, http.StatusMethodNotAllowed, "method_not_allowed",
+				"use PUT/DELETE on /api/sources/{name} to modify a source")
+			return
+		}
+		for _, s := range g.core.Sources() {
+			if s.Name == name {
+				// ?reveal=credentials is admin-only and is what the Web UI's
+				// edit dialog uses: a form that round-trips a source must be
+				// able to show the current key, otherwise saving an unrelated
+				// field would blank it. Everything else stays masked.
+				if r.URL.Query().Get("reveal") == "credentials" {
+					if reqRole(r.Context()) != "admin" {
+						writeError(w, http.StatusForbidden, "forbidden",
+							"admin role required to reveal credentials")
+						return
+					}
+					// The edit form round-trips this response, so it needs the credential in the
+					// clear AND the durations — see maskSourceFields.
+					writeJSON(w, http.StatusOK, map[string]interface{}{"source": maskSourceFields(s, true, true)})
+					return
+				}
+				writeJSON(w, http.StatusOK, map[string]interface{}{"source": maskSource(s)})
+				return
+			}
+		}
+		writeError(w, http.StatusNotFound, "not_found", "no such source: "+name)
+	case path == "auto":
+		if r.Method != http.MethodGet {
+			writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "use GET")
+			return
+		}
+		writeJSON(w, http.StatusOK, map[string]interface{}{
+			"rules":       g.core.AutoRules(),
+			"image_rules": g.core.AutoImageRules(),
+			"states":      g.core.AutoSlotStates(),
+		})
+	case path == "keys":
+		if reqRole(r.Context()) != "admin" {
+			writeError(w, http.StatusForbidden, "forbidden", "admin role required")
+			return
+		}
+		if r.Method != http.MethodGet {
+			writeError(w, http.StatusMethodNotAllowed, "method_not_allowed",
+				"use POST/PUT/DELETE on /api/keys to manage keys")
+			return
+		}
+		keys := g.core.ListKeys()
+		out := make([]map[string]interface{}, 0, len(keys))
+		for _, k := range keys {
+			out = append(out, map[string]interface{}{
+				"name":       k.Name,
+				"role":       k.Role,
+				"models":     k.Models,
+				"note":       k.Note,
+				"created_at": k.CreatedAt,
+				"seed":       k.Seed,
+				// Quotas live on the scope entries (k.Models), echoed above;
+				// there is deliberately no key-wide total.
+				// The secret itself is never echoed. An operator that needs it
+				// already has it from creation time or from config.yaml.
+				"key_prefix": maskKey(k.Key),
+			})
+		}
+		writeJSON(w, http.StatusOK, map[string]interface{}{"keys": out})
+	default:
+		g.apiV1Index(w, r)
+	}
+}
+
+// apiV1Index is the discovery document: it names every endpoint with its
+// method, auth requirement and purpose, so an agent does not have to scrape the
+// HTML to find out what it can call.
+func (g *Gateway) apiV1Index(w http.ResponseWriter, r *http.Request) {
+	type ep struct {
+		Method      string `json:"method"`
+		Path        string `json:"path"`
+		Auth        string `json:"auth"`
+		Summary     string `json:"summary"`
+		WriteEffect string `json:"writes,omitempty"`
+	}
+	index := map[string]interface{}{
+		"api_version": "v1",
+		"description": "llmsproxy management API. Authenticate with a gateway key: " +
+			"'Authorization: Bearer <key>', '?api_key=<key>', or the gw_key cookie. " +
+			"Read endpoints accept any role; mutations require an admin key.",
+		"discovery": []ep{
+			{Method: "GET", Path: "/api/v1", Auth: "any", Summary: "this document"},
+			{Method: "GET", Path: "/api/v1/overview", Auth: "any",
+				Summary: "current state in one call: sources, auto chain, key count, source health"},
+			{Method: "GET", Path: "/api/v1/health", Auth: "any",
+				Summary: "per-source health snapshot only"},
+			{Method: "GET", Path: "/api/v1/models", Auth: "any",
+				Summary: "every model id the gateway can route, grouped by owning source"},
+			{Method: "GET", Path: "/api/v1/sources", Auth: "any",
+				Summary: "all sources with credentials masked"},
+			{Method: "GET", Path: "/api/v1/sources/{name}", Auth: "any", Summary: "one source"},
+			{Method: "GET", Path: "/api/v1/auto", Auth: "any",
+				Summary: "AUTO scheduling chain and live per-slot state"},
+			{Method: "GET", Path: "/api/v1/keys", Auth: "admin",
+				Summary: "gateway key metadata; secrets are never returned"},
+
+			{Method: "GET", Path: "/api/sources", Auth: "admin", Summary: "raw source list (includes api_key)"},
+			{Method: "POST", Path: "/api/sources", Auth: "admin", Summary: "add or replace a source",
+				WriteEffect: "writes config.yaml (api_key sealed at rest)"},
+			{Method: "PUT", Path: "/api/sources/{name}", Auth: "admin", Summary: "update one source",
+				WriteEffect: "writes config.yaml (api_key sealed at rest)"},
+			{Method: "DELETE", Path: "/api/sources/{name}", Auth: "admin", Summary: "delete a source",
+				WriteEffect: "writes config.yaml"},
+
+			{Method: "GET", Path: "/api/auto", Auth: "any", Summary: "scheduling chain (readable by any role)"},
+			{Method: "PUT", Path: "/api/auto", Auth: "admin", Summary: "replace the scheduling chain",
+				WriteEffect: "writes config.yaml (the AUTO chain lives here)"},
+
+			{Method: "GET", Path: "/api/keys", Auth: "admin", Summary: "gateway keys"},
+			{Method: "POST", Path: "/api/keys", Auth: "admin", Summary: "create a gateway key",
+				WriteEffect: "writes config.yaml"},
+			{Method: "PUT", Path: "/api/keys/{key}", Auth: "admin",
+				Summary:     "update a gateway key (name, role, model scopes and their per-model quotas)",
+				WriteEffect: "writes config.yaml"},
+			{Method: "DELETE", Path: "/api/keys/{key}", Auth: "admin",
+				Summary:     "delete a gateway key — the path segment is the KEY itself, not its name",
+				WriteEffect: "writes config.yaml"},
+
+			{Method: "GET", Path: "/api/status", Auth: "any", Summary: "per-source health detail"},
+			{Method: "GET", Path: "/api/stats", Auth: "any", Summary: "usage aggregates"},
+			{Method: "GET", Path: "/api/stats/records", Auth: "any", Summary: "paged request records"},
+			{Method: "GET", Path: "/api/adapters", Auth: "admin", Summary: "installed Lua adapters"},
+			{Method: "POST", Path: "/api/sources/discover", Auth: "admin",
+				Summary: "probe a base_url+api_key for its model catalog",
+				// Probing is read-only with respect to configuration: it writes
+				// nothing, which is why it is safe to call repeatedly while the
+				// operator is still typing a key.
+				WriteEffect: "none",
+			},
+			{Method: "POST", Path: "/api/sources/discover/apply", Auth: "admin",
+				Summary: "probe and write the source in one step",
+				// Requires replace_models=true when the source already exists,
+				// so re-running discovery cannot silently discard a hand-tuned
+				// model list. An empty api_key keeps the stored credential.
+				WriteEffect: "creates or replaces a source and its model list",
+			},
+			{Method: "GET", Path: "/api/source_templates", Auth: "admin", Summary: "source templates"},
+			{Method: "POST", Path: "/v1/chat/completions", Auth: "any", Summary: "OpenAI-compatible inference"},
+		},
+		"conventions": map[string]interface{}{
+			"errors":      "{ \"error\": { \"type\": <code>, \"message\": <text> } }",
+			"path_escape": "URL-encode source and key names; {name} is a single path segment",
+			"idempotency": "POST /api/sources and PUT /api/sources/{name} both upsert by name; " +
+				"PUT takes the name from the path and rejects a body name that disagrees",
+			"partial_update": "api_key may be omitted or sent as the literal \"__KEEP__\" to inherit the " +
+				"current credential; model_ids adds models to the existing list instead of replacing it, " +
+				"so a one-field edit never needs to read the source first",
+			"optional_fields": "proxy_url, api_key_env, timeout and queue_timeout use presence semantics: " +
+				"OMITTED from the request keeps the stored value, present (even as \"\") overwrites it. " +
+				"The upsert rewrites the whole source, so without this a one-field edit would silently reset " +
+				"them — and api_key_env in particular is only visible as a credential failure on the NEXT " +
+				"upstream call. api_key deliberately keeps the older \"empty means inherit\" rule instead, " +
+				"because losing a credential breaks the source while losing a proxy only changes its route.",
+			"durations": "timeout and queue_timeout are Go duration strings (\"300s\", \"5m\", \"1h\"); " +
+				"the empty string clears them back to the config defaults",
+			"config_truth": "all configuration lives in config.yaml; API writes are persisted immediately",
+			"credentials": "credentials are masked by default. GET /api/v1/sources/{name}?reveal=credentials " +
+				"returns them in the clear and is admin-only — the Web UI edit dialog uses it, because a form " +
+				"that round-trips a source must show the current key or saving another field would blank it.",
+		},
+	}
+	writeJSON(w, http.StatusOK, index)
+}
+
+// apiV1Overview is the "what is the current state" call an agent makes first.
+// One round trip instead of five.
+func (g *Gateway) apiV1Overview(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodGet {
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "use GET")
+		return
+	}
+	sources := g.core.Sources()
+	names := make([]string, 0, len(sources))
+	modelCount := 0
+	for _, s := range sources {
+		names = append(names, s.Name)
+		modelCount += len(s.Models)
+	}
+	keys := g.core.ListKeys()
+	adminCount, userCount := 0, 0
+	for _, k := range keys {
+		if k.Role == "admin" {
+			adminCount++
+		} else {
+			userCount++
+		}
+	}
+	writeJSON(w, http.StatusOK, map[string]interface{}{
+		"api_version": "v1",
+		"summary": map[string]interface{}{
+			"source_count":      len(sources),
+			"model_count":       modelCount,
+			"gateway_key_count": len(keys),
+			"admin_key_count":   adminCount,
+			"user_key_count":    userCount,
+			"auto_slots":        len(g.core.AutoRules()),
+		},
+		"sources":    maskSources(sources),
+		"auto":       g.core.AutoRules(),
+		"auto_image": g.core.AutoImageRules(),
+		"health":     g.sourceHealthBrief(),
+		"caller": map[string]interface{}{
+			"role": reqRole(r.Context()),
+			"key":  maskKey(reqKey(r.Context())),
+		},
+	})
+}
+
+func (g *Gateway) apiV1Health(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodGet {
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "use GET")
+		return
+	}
+	writeJSON(w, http.StatusOK, map[string]interface{}{"sources": g.sourceHealthBrief()})
+}
+
+// apiV1Models lists routable model ids per source, which is what an agent needs
+// to build a valid request — /v1/models flattens them and hides the owner.
+func (g *Gateway) apiV1Models(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodGet {
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "use GET")
+		return
+	}
+	bySource := map[string][]string{}
+	var all []string
+	for _, s := range g.core.Sources() {
+		for _, m := range s.Models {
+			bySource[s.Name] = append(bySource[s.Name], m.ID)
+			all = append(all, m.ID)
+		}
+	}
+	sort.Strings(all)
+	all = dedupeStrings(all)
+	writeJSON(w, http.StatusOK, map[string]interface{}{
+		"count":     len(all),
+		"models":    all,
+		"by_source": bySource,
+		"note":      "request a model as \"<source>:<model>\" to pin one source, or the bare id to let the gateway choose",
+	})
+}
+
+// ---- helpers ----
+
+// maskSources returns sources with credentials replaced by a presence marker.
+// The UI still uses /api/sources (which returns real values, because its edit
+// form round-trips them); this is the safe view for programmatic callers.
+func maskSources(srcs []config.Source) []map[string]interface{} {
+	out := make([]map[string]interface{}, 0, len(srcs))
+	for _, s := range srcs {
+		out = append(out, maskSource(s))
+	}
+	return out
+}
+
+func maskSource(s config.Source) map[string]interface{} {
+	return maskSourceFields(s, false, false)
+}
+
+// maskSourceFields builds the API view of a source.
+//
+// revealCredentials swaps the masked api_key for the real one — admin-only, and
+// used by the WebUI edit dialog, whose form has to round-trip the whole source
+// or saving an unrelated field would blank the key.
+//
+// exposeDurations adds the two durations as strings, matching what the write
+// path accepts. It rides along with the credential reveal because both exist for
+// the same reason: config.Source tags Timeout/QueueTimeout `json:"-"`, so a
+// plain marshal of the struct omits them. A form that cannot SEE the stored
+// timeout would clear it on every save, since the timeout box is always sent.
+func maskSourceFields(s config.Source, revealCredentials, exposeDurations bool) map[string]interface{} {
+	key := maskKey(s.APIKey)
+	if revealCredentials {
+		key = s.APIKey
+	}
+	m := map[string]interface{}{
+		"name":           s.Name,
+		"base_url":       s.BaseURL,
+		"adapter":        s.Adapter,
+		"endpoint":       s.Endpoint,
+		"image_endpoint": s.ImageEndpoint,
+		"api_key":        key,
+		"api_key_set":    s.APIKey != "",
+		"models":         s.Models,
+		"headers":        maskHeaders(s.Headers),
+		"proxy_url":      s.ProxyURL,
+		"api_key_env":    s.APIKeyEnv,
+		"meta":           s.Meta,
+		"temperature":    s.Temperature,
+		"max_tokens":     s.MaxTokens,
+		"max_concurrent": s.MaxConcurrent,
+		"rpm":            s.RPM,
+	}
+	if exposeDurations {
+		m["timeout"] = s.Timeout.String()
+		m["queue_timeout"] = s.QueueTimeout.String()
+	}
+	return m
+}
+
+func maskKey(k string) string {
+	if k == "" {
+		return ""
+	}
+	if strings.HasPrefix(k, "enc:v1:") {
+		return "(sealed)"
+	}
+	if len(k) <= 10 {
+		return k[:2] + "…"
+	}
+	return k[:6] + "…" + k[len(k)-4:]
+}
+
+func maskHeaders(h map[string]string) map[string]string {
+	if h == nil {
+		return nil
+	}
+	out := make(map[string]string, len(h))
+	for k, v := range h {
+		out[k] = maskKey(v)
+	}
+	return out
+}
+
+// sourceHealthBrief reuses the same registry status the UI's /api/status shows,
+// so the agent view and the UI view cannot drift apart. Recent-traffic counters
+// come from the same stats window, because a source that is actually serving
+// traffic must never look down just because a probe was rate-limited.
+func (g *Gateway) sourceHealthBrief() []map[string]interface{} {
+	sts := g.core.Registry().Status()
+	recent := g.stats.SourceRecent(300)
+	avgs := g.stats.SourceAverages(300)
+	for i := range sts {
+		if v, ok := recent[sts[i].Name]; ok {
+			sts[i].RecentOK = v[0]
+			sts[i].RecentErr = v[1]
+		}
+		if a, ok := avgs[sts[i].Name]; ok {
+			sts[i].AvgFirstByteMs = a.AvgFirstByteMs
+			sts[i].AvgTokPerS = a.AvgTokPerS
+		}
+	}
+	out := make([]map[string]interface{}, 0, len(sts))
+	for _, s := range sts {
+		row := map[string]interface{}{
+			"name":           s.Name,
+			"adapter":        s.Adapter,
+			"healthy":        s.Healthy,
+			"available":      s.Available,
+			"live_available": s.LiveAvailable,
+			"model_count":    len(s.Models),
+		}
+		if s.LastError != "" {
+			row["last_error"] = s.LastError
+		}
+		if s.LastChecked > 0 {
+			row["last_checked"] = s.LastChecked
+		}
+		if s.RecentOK > 0 || s.RecentErr > 0 {
+			row["recent_ok"] = s.RecentOK
+			row["recent_err"] = s.RecentErr
+		}
+		if s.AvgFirstByteMs > 0 {
+			row["avg_first_byte_ms"] = s.AvgFirstByteMs
+		}
+		if s.AvgTokPerS > 0 {
+			row["avg_tok_per_s"] = s.AvgTokPerS
+		}
+		if s.FailCount > 0 {
+			row["fail_count"] = s.FailCount
+		}
+		if s.Permanent {
+			row["permanent"] = true
+		}
+		out = append(out, row)
+	}
+	return out
+}
+
+// decodePathSegment URL-decodes one path segment and rejects an empty result,
+// so a name containing a slash cannot be silently mis-resolved.
+func decodePathSegment(seg string) (string, error) {
+	s, err := url.PathUnescape(seg)
+	if err != nil {
+		return "", err
+	}
+	if s == "" {
+		return "", errEmptySegment
+	}
+	return s, nil
+}
+
+func dedupeStrings(in []string) []string {
+	out := in[:0]
+	var last string
+	for i, s := range in {
+		if i == 0 || s != last {
+			out = append(out, s)
+		}
+		last = s
+	}
+	return out
+}
+
+// errEmptySegment marks a path like /api/v1/sources/ with no name after it.
+var errEmptySegment = errors.New("empty path segment")

internal/gateway/apiv1_test.go

@@ -0,0 +1,299 @@
+package gateway
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+)
+
+// v1Gateway builds a gateway with one source and two keys, and returns the
+// secrets so a test can assert they never appear in a response.
+func v1Gateway(t *testing.T) (*Gateway, string, string) {
+	t.Helper()
+	g := newTestGateway(t, config.Source{
+		Name:    "up",
+		BaseURL: "http://up.example/v1",
+		APIKey:  "sk-up-secret",
+		Adapter: "openai",
+		Models:  []config.Model{{ID: "m1", Kind: "chat"}},
+	})
+	// The shared helper only seeds one admin key; add a user key through the
+	// core so role-gated endpoints have something to reject.
+	if _, err := g.core.CreateKey("agent", "user", nil, "test agent key"); err != nil {
+		t.Fatalf("CreateKey: %v", err)
+	}
+	rec := doReq(t, g, http.MethodGet, "/api/keys", "")
+	var doc struct {
+		Keys []config.GWKey `json:"keys"`
+	}
+	if err := json.Unmarshal(rec.Body.Bytes(), &doc); err != nil {
+		t.Fatalf("decode keys: %v", err)
+	}
+	userKey := ""
+	for _, k := range doc.Keys {
+		if k.Role == "user" {
+			userKey = k.Key
+		}
+	}
+	if userKey == "" {
+		t.Fatal("no user key was created")
+	}
+	return g, "sk-test", userKey
+}
+
+func decodeJSON(t *testing.T, body string, v interface{}) {
+	t.Helper()
+	if err := json.Unmarshal([]byte(body), v); err != nil {
+		t.Fatalf("decode %s: %v", body, err)
+	}
+}
+
+func TestAPIV1IndexIsDiscoverable(t *testing.T) {
+	g, _, _ := v1Gateway(t)
+	rec := doReq(t, g, http.MethodGet, "/api/v1", "")
+	if rec.Code != http.StatusOK {
+		t.Fatalf("status = %d, want 200: %s", rec.Code, rec.Body.String())
+	}
+	var doc struct {
+		APIVersion  string `json:"api_version"`
+		Description string `json:"description"`
+		Discovery   []struct {
+			Method  string `json:"method"`
+			Path    string `json:"path"`
+			Auth    string `json:"auth"`
+			Summary string `json:"summary"`
+		} `json:"discovery"`
+		Conventions map[string]interface{} `json:"conventions"`
+	}
+	decodeJSON(t, rec.Body.String(), &doc)
+	if doc.APIVersion != "v1" {
+		t.Errorf("api_version = %q", doc.APIVersion)
+	}
+	if doc.Description == "" {
+		t.Error("index should carry a usage description")
+	}
+	if len(doc.Discovery) == 0 {
+		t.Fatal("discovery list is empty")
+	}
+	want := map[string]bool{
+		"/api/v1/overview": false, "/api/v1/sources": false,
+		"/api/v1/auto": false, "/api/v1/models": false,
+		"/api/v1/health": false, "/api/v1/keys": false,
+		"/api/sources": false, "/api/auto": false, "/v1/chat/completions": false,
+	}
+	for _, e := range doc.Discovery {
+		if _, ok := want[e.Path]; ok {
+			want[e.Path] = true
+		}
+		if e.Method == "" || e.Path == "" || e.Auth == "" || e.Summary == "" {
+			t.Errorf("incomplete discovery entry: %+v", e)
+		}
+	}
+	for p, found := range want {
+		if !found {
+			t.Errorf("discovery is missing %s", p)
+		}
+	}
+	if doc.Conventions["errors"] == nil {
+		t.Error("conventions should document the error shape")
+	}
+}
+
+func TestAPIV1OverviewSummarisesState(t *testing.T) {
+	g, _, _ := v1Gateway(t)
+	rec := doReq(t, g, http.MethodGet, "/api/v1/overview", "")
+	if rec.Code != http.StatusOK {
+		t.Fatalf("status = %d: %s", rec.Code, rec.Body.String())
+	}
+	var doc struct {
+		Summary struct {
+			SourceCount     int `json:"source_count"`
+			ModelCount      int `json:"model_count"`
+			GatewayKeyCount int `json:"gateway_key_count"`
+		} `json:"summary"`
+		Sources []map[string]interface{} `json:"sources"`
+		Auto    []map[string]interface{} `json:"auto"`
+		Health  []map[string]interface{} `json:"health"`
+		Caller  struct {
+			Role string `json:"role"`
+		} `json:"caller"`
+	}
+	decodeJSON(t, rec.Body.String(), &doc)
+	if doc.Summary.SourceCount != 1 {
+		t.Errorf("source_count = %d, want 1", doc.Summary.SourceCount)
+	}
+	if doc.Summary.ModelCount != 1 {
+		t.Errorf("model_count = %d, want 1", doc.Summary.ModelCount)
+	}
+	if doc.Summary.GatewayKeyCount < 2 {
+		t.Errorf("gateway_key_count = %d, want >= 2", doc.Summary.GatewayKeyCount)
+	}
+	if len(doc.Sources) != 1 || doc.Sources[0]["name"] != "up" {
+		t.Errorf("sources = %+v", doc.Sources)
+	}
+	if doc.Caller.Role != "admin" {
+		t.Errorf("caller.role = %q, want admin", doc.Caller.Role)
+	}
+}
+
+func TestAPIV1NeverEchoesSecrets(t *testing.T) {
+	g, admin, user := v1Gateway(t)
+	for _, path := range []string{"/api/v1", "/api/v1/overview", "/api/v1/sources", "/api/v1/sources/up", "/api/v1/keys", "/api/v1/models", "/api/v1/health", "/api/v1/auto"} {
+		rec := doReq(t, g, http.MethodGet, path, "")
+		if rec.Code != http.StatusOK {
+			t.Fatalf("%s status = %d: %s", path, rec.Code, rec.Body.String())
+		}
+		if strings.Contains(rec.Body.String(), "sk-up-secret") {
+			t.Errorf("%s leaked the source api_key", path)
+		}
+		if admin != "" && strings.Contains(rec.Body.String(), admin) {
+			t.Errorf("%s leaked the admin gateway key", path)
+		}
+		if strings.Contains(rec.Body.String(), user) {
+			t.Errorf("%s leaked the user gateway key", path)
+		}
+	}
+	// Masking must still be useful: it says a key is configured.
+	rec := doReq(t, g, http.MethodGet, "/api/v1/sources/up", "")
+	if !strings.Contains(rec.Body.String(), `"api_key_set":true`) {
+		t.Error("masked source should report that a key is configured")
+	}
+}
+
+func TestAPIV1KeysRequiresAdminAndMasks(t *testing.T) {
+	g, _, user := v1Gateway(t)
+	// Authenticate as the user key by swapping the shared helper's header.
+	req := newAuthedRequest(t, http.MethodGet, "/api/v1/keys", user)
+	rec := serveViaHandler(t, g, req)
+	if rec.Code != http.StatusForbidden {
+		t.Errorf("user role on /api/v1/keys = %d, want 403: %s", rec.Code, rec.Body.String())
+	}
+}
+
+func TestAPIV1SourcesNotFound(t *testing.T) {
+	g, _, _ := v1Gateway(t)
+	rec := doReq(t, g, http.MethodGet, "/api/v1/sources/nope", "")
+	if rec.Code != http.StatusNotFound {
+		t.Errorf("status = %d, want 404: %s", rec.Code, rec.Body.String())
+	}
+}
+
+func TestAPIV1RejectsWrites(t *testing.T) {
+	g, _, _ := v1Gateway(t)
+	for _, m := range []string{http.MethodPost, http.MethodPut, http.MethodDelete, http.MethodPatch} {
+		rec := doReq(t, g, m, "/api/v1/sources", `{}`)
+		if rec.Code == http.StatusOK {
+			t.Errorf("%s /api/v1/sources returned 200; the v1 read facade must not mutate", m)
+		}
+	}
+	// And the source must be untouched.
+	rec := doReq(t, g, http.MethodGet, "/api/v1/sources", "")
+	if !strings.Contains(rec.Body.String(), `"name":"up"`) {
+		t.Error("the source disappeared after a rejected write")
+	}
+}
+
+func TestAPIV1ModelsGroupsBySource(t *testing.T) {
+	g, _, _ := v1Gateway(t)
+	rec := doReq(t, g, http.MethodGet, "/api/v1/models", "")
+	if rec.Code != http.StatusOK {
+		t.Fatalf("status = %d", rec.Code)
+	}
+	var doc struct {
+		Count    int                 `json:"count"`
+		Models   []string            `json:"models"`
+		BySource map[string][]string `json:"by_source"`
+	}
+	decodeJSON(t, rec.Body.String(), &doc)
+	if doc.Count != 1 || doc.Models[0] != "m1" {
+		t.Errorf("models = %+v", doc.Models)
+	}
+	if len(doc.BySource["up"]) != 1 || doc.BySource["up"][0] != "m1" {
+		t.Errorf("by_source = %+v", doc.BySource)
+	}
+}
+
+func TestAPIV1RequiresAuthentication(t *testing.T) {
+	g, _, _ := v1Gateway(t)
+	for _, path := range []string{"/api/v1", "/api/v1/overview", "/api/v1/sources", "/api/v1/health", "/api/v1/models"} {
+		req := newRequest(http.MethodGet, path, "")
+		rec := serveViaHandler(t, g, req)
+		if rec.Code != http.StatusUnauthorized {
+			t.Errorf("%s without a key = %d, want 401", path, rec.Code)
+		}
+	}
+}
+
+func TestMaskKeyHidesTheSecret(t *testing.T) {
+	cases := map[string]string{
+		"":                   "",
+		"ab":                 "ab…",
+		"sk-gw-abcdefghijkl": "sk-gw-…ijkl",
+		"enc:v1:AAAA":        "(sealed)",
+	}
+	for in, want := range cases {
+		if got := maskKey(in); got != want {
+			t.Errorf("maskKey(%q) = %q, want %q", in, got, want)
+		}
+	}
+	// The masked form must not reveal the middle of the secret.
+	full := maskKey("sk-gw-1234567890abcdef")
+	if strings.Contains(full, "4567890abcdef") {
+		t.Errorf("maskKey leaked the middle: %q", full)
+	}
+}
+
+// newRequest builds an unauthenticated request.
+func newRequest(method, path, body string) *http.Request {
+	req, _ := http.NewRequest(method, path, strings.NewReader(body))
+	if body != "" {
+		req.Header.Set("Content-Type", "application/json")
+	}
+	return req
+}
+
+// newAuthedRequest builds a request carrying the given bearer key.
+func newAuthedRequest(t *testing.T, method, path, key string) *http.Request {
+	t.Helper()
+	req := newRequest(method, path, "")
+	req.Header.Set("Authorization", "Bearer "+key)
+	return req
+}
+
+// serveViaHandler pushes a request through the full handler chain (auth included).
+func serveViaHandler(t *testing.T, g *Gateway, req *http.Request) *httptest.ResponseRecorder {
+	t.Helper()
+	rec := httptest.NewRecorder()
+	g.Handler().ServeHTTP(rec, req)
+	return rec
+}
+
+// TestAPIV1RevealRequiresAdmin: the only way to read a key in the clear is the
+// explicit admin-only reveal. Without the role check this endpoint would hand
+// every source credential to any valid (even user-scoped) key.
+func TestAPIV1RevealRequiresAdmin(t *testing.T) {
+	g, _, user := v1Gateway(t)
+	rec := serveViaHandler(t, g, newAuthedRequest(t, http.MethodGet, "/api/v1/sources/up?reveal=credentials", user))
+	if rec.Code != http.StatusForbidden {
+		t.Errorf("user reveal = %d, want 403: %s", rec.Code, rec.Body.String())
+	}
+	if strings.Contains(rec.Body.String(), "sk-up-secret") {
+		t.Error("a forbidden reveal still leaked the key")
+	}
+	rec = doReq(t, g, http.MethodGet, "/api/v1/sources/up?reveal=credentials", "")
+	if rec.Code != http.StatusOK {
+		t.Fatalf("admin reveal = %d: %s", rec.Code, rec.Body.String())
+	}
+	if !strings.Contains(rec.Body.String(), "sk-up-secret") {
+		t.Error("admin reveal should return the key in the clear (the edit form needs it)")
+	}
+	// The masked default view must still hide it on the same path.
+	rec = doReq(t, g, http.MethodGet, "/api/v1/sources/up", "")
+	if strings.Contains(rec.Body.String(), "sk-up-secret") {
+		t.Error("the default view leaked the key")
+	}
+}

internal/gateway/billing_rules_api.go

@@ -0,0 +1,282 @@
+package gateway
+
+import (
+	"encoding/json"
+	"net/http"
+
+	"llmsproxy/internal/billing"
+	"llmsproxy/internal/config"
+)
+
+// Billing rules API.
+//
+//	GET  /api/plugins/billing/rules    the active profile's rules + all profiles
+//	PUT  /api/plugins/billing/rules    replace the whole DSL and re-inject prices
+//	POST /api/plugins/billing/rules    { profile, rule } append one rule
+//	DELETE /api/plugins/billing/rules  { profile, url } remove one rule
+//
+// Why a separate endpoint instead of the generic plugin state: prices are
+// plugin state, but they are also a durable, reviewable CONFIGURATION. Routing
+// them through /state would let an admin key PUT arbitrary plugin state and
+// silently reset the accumulated accounting. Here the server owns the shape:
+// it validates the DSL, writes it back to config.yaml, recompiles, and hands
+// the plugin its prices table. The operator edits rules; the plugin keeps
+// numbers. The two are never mixed in one payload.
+//
+// Editing means "what the operator typed is what config.yaml holds". A
+// billing mistake found next week must be traceable to a reviewable file, not
+// to a blob in the plugin's state sidecar.
+func (g *Gateway) handleBillingRules(w http.ResponseWriter, r *http.Request) {
+	if reqRole(r.Context()) != "admin" {
+		writeError(w, http.StatusForbidden, "forbidden", "admin role required")
+		return
+	}
+	switch r.Method {
+	case http.MethodGet:
+		g.getBillingRules(w)
+	case http.MethodPut:
+		g.putBillingRules(w, r)
+	case http.MethodPost:
+		g.addBillingRule(w, r)
+	case http.MethodDelete:
+		g.delBillingRule(w, r)
+	default:
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "use GET/PUT/POST/DELETE")
+	}
+}
+
+// billingRulesPayload is the UI-facing view: the profiles as declared, plus
+// which one is active and the source URLs they can match — the editor needs
+// the URL list to offer suggestions, and an operator typing a URL that matches
+// no source is exactly the silent-no-pricing failure this feature exists to
+// remove.
+type billingRulesPayload struct {
+	DSL      *config.BillingDSL `json:"billing"`
+	Active   string             `json:"active"`
+	URLs     []string           `json:"urls"`
+	Warnings []string           `json:"warnings"`
+}
+
+func (g *Gateway) billingRulesPayload() billingRulesPayload {
+	out := billingRulesPayload{}
+	cfg := g.core.Config()
+	if cfg != nil && cfg.BillingDSL != nil {
+		// Copy so the response cannot be mutated back through the pointer.
+		cp := *cfg.BillingDSL
+		out.DSL = &cp
+		if p := cp.Resolve(""); p != nil {
+			out.Active = p.ID
+		}
+	}
+	for _, s := range g.sourceURLs() {
+		out.URLs = append(out.URLs, s)
+	}
+	out.Warnings = g.billingRuleWarnings()
+	return out
+}
+
+func (g *Gateway) sourceURLs() []string {
+	cfg := g.core.Config()
+	if cfg == nil {
+		return nil
+	}
+	seen := map[string]bool{}
+	var out []string
+	for _, s := range cfg.Sources {
+		if s.BaseURL == "" || seen[s.BaseURL] {
+			continue
+		}
+		seen[s.BaseURL] = true
+		out = append(out, s.BaseURL)
+	}
+	return out
+}
+
+// billingRuleWarnings reports rules that match no configured source. This is
+// the silent killer of hand-written price tables: the rule parses, validates,
+// and prices nothing at all, so every request on that URL lands in unpriced.
+// Saying so out loud is the difference between a five-second fix and an
+// afternoon wondering why the bill is zero.
+func (g *Gateway) billingRuleWarnings() []string {
+	cfg := g.core.Config()
+	if cfg == nil || cfg.BillingDSL == nil {
+		return nil
+	}
+	urls := map[string]bool{}
+	for _, u := range g.sourceURLs() {
+		urls[u] = true
+	}
+	var warns []string
+	for _, p := range cfg.BillingDSL.Profiles {
+		for _, r := range p.Rules {
+			if r.URL == "*" || urls[r.URL] {
+				continue
+			}
+			warns = append(warns, "profile "+p.ID+": rule url "+r.URL+" matches no configured source")
+		}
+	}
+	return warns
+}
+
+func (g *Gateway) getBillingRules(w http.ResponseWriter) {
+	writeJSON(w, http.StatusOK, g.billingRulesPayload())
+}
+
+func (g *Gateway) putBillingRules(w http.ResponseWriter, r *http.Request) {
+	var body struct {
+		Billing *config.BillingDSL `json:"billing"`
+	}
+	if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+		writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+		return
+	}
+	if body.Billing == nil {
+		writeError(w, http.StatusBadRequest, "invalid_request", "billing is required")
+		return
+	}
+	g.applyBillingDSLUpdate(w, body.Billing)
+}
+
+func (g *Gateway) addBillingRule(w http.ResponseWriter, r *http.Request) {
+	var body struct {
+		Profile string             `json:"profile"`
+		Rule    config.BillingRule `json:"rule"`
+	}
+	if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+		writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+		return
+	}
+	cfg := g.core.Config()
+	if cfg == nil || cfg.BillingDSL == nil || len(cfg.BillingDSL.Profiles) == 0 {
+		writeError(w, http.StatusBadRequest, "no_profile", "no billing profiles configured; create one first")
+		return
+	}
+	next := cloneBillingDSL(cfg.BillingDSL)
+	idx := profileIndex(next, body.Profile)
+	if idx < 0 {
+		writeError(w, http.StatusBadRequest, "unknown_profile", "no such profile: "+body.Profile)
+		return
+	}
+	p := &next.Profiles[idx]
+	// Replace rather than append when the URL is already declared: two rules
+	// for one URL would silently make the first unreachable (first match wins),
+	// which is the exact ambiguity an editor should not be able to create.
+	replaced := false
+	for i := range p.Rules {
+		if p.Rules[i].URL == body.Rule.URL {
+			p.Rules[i] = body.Rule
+			replaced = true
+			break
+		}
+	}
+	if !replaced {
+		p.Rules = append(p.Rules, body.Rule)
+	}
+	g.applyBillingDSLUpdate(w, next)
+}
+
+func (g *Gateway) delBillingRule(w http.ResponseWriter, r *http.Request) {
+	var body struct {
+		Profile string `json:"profile"`
+		URL     string `json:"url"`
+	}
+	if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+		writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+		return
+	}
+	cfg := g.core.Config()
+	if cfg == nil || cfg.BillingDSL == nil {
+		writeError(w, http.StatusBadRequest, "no_profile", "no billing profiles configured")
+		return
+	}
+	next := cloneBillingDSL(cfg.BillingDSL)
+	idx := profileIndex(next, body.Profile)
+	if idx < 0 {
+		writeError(w, http.StatusBadRequest, "unknown_profile", "no such profile: "+body.Profile)
+		return
+	}
+	p := &next.Profiles[idx]
+	out := p.Rules[:0]
+	found := false
+	for _, rule := range p.Rules {
+		if rule.URL == body.URL {
+			found = true
+			continue
+		}
+		out = append(out, rule)
+	}
+	if !found {
+		writeError(w, http.StatusNotFound, "not_found", "no rule for url "+body.URL)
+		return
+	}
+	p.Rules = out
+	g.applyBillingDSLUpdate(w, next)
+}
+
+// applyBillingDSLUpdate is the single write path: validate, persist to
+// config.yaml, recompile, inject, and only then report success. If the config
+// cannot be written the change is NOT applied in memory either — a rules editor
+// that says "saved" and loses the edit on the next restart is worse than one
+// that refuses.
+func (g *Gateway) applyBillingDSLUpdate(w http.ResponseWriter, next *config.BillingDSL) {
+	if err := next.Validate(); err != nil {
+		writeError(w, http.StatusBadRequest, "invalid_rules", err.Error())
+		return
+	}
+	cfg := g.core.Config()
+	if cfg == nil {
+		writeError(w, http.StatusInternalServerError, "no_config", "no config loaded")
+		return
+	}
+	// Assign BEFORE saving. The previous order saved first and assigned after,
+	// so every rule edit reported success while writing a config.yaml with no
+	// billing section at all — the operator's edit vanished on restart and
+	// nothing in the API response said so.
+	prev := cfg.BillingDSL
+	cfg.BillingDSL = next
+
+	// Compile before persisting: a profile that cannot be turned into prices
+	// must not reach the config file, or the next restart fails to load the
+	// very rules the editor just accepted.
+	prices, err := billing.CompileOpts(next.Resolve(next.Active), cfg.Sources, true)
+	if err != nil {
+		cfg.BillingDSL = prev // no half-applied state
+		writeError(w, http.StatusBadRequest, "compile_failed", err.Error())
+		return
+	}
+	if err := cfg.Save(); err != nil {
+		cfg.BillingDSL = prev
+		writeError(w, http.StatusInternalServerError, "persist_failed", err.Error())
+		return
+	}
+	ps := g.core.Plugins()
+	if ps != nil {
+		if err := ps.SetState("billing", map[string]interface{}{"prices": prices}); err != nil {
+			writeError(w, http.StatusBadRequest, "plugin_error", err.Error())
+			return
+		}
+	}
+	p := g.billingRulesPayload()
+	writeJSON(w, http.StatusOK, map[string]interface{}{
+		"ok": true, "billing": p.DSL, "active": p.Active, "warnings": p.Warnings,
+	})
+}
+
+func cloneBillingDSL(d *config.BillingDSL) *config.BillingDSL {
+	out := &config.BillingDSL{Active: d.Active}
+	out.Profiles = make([]config.BillingProfile, len(d.Profiles))
+	copy(out.Profiles, d.Profiles)
+	return out
+}
+
+func profileIndex(d *config.BillingDSL, id string) int {
+	if id == "" {
+		return -1
+	}
+	for i, p := range d.Profiles {
+		if p.ID == id {
+			return i
+		}
+	}
+	return -1
+}

internal/gateway/billing_rules_api_test.go

@@ -0,0 +1,281 @@
+package gateway
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"sort"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+)
+
+// rulesGateway needs a real billing plugin (the API injects prices into it)
+// AND two sources whose base_urls the rules will match, so it cannot reuse
+// either existing helper: gatewayWithBilling has no sources, and
+// newTestGateway loads no plugins.
+func rulesGateway(t *testing.T) *Gateway {
+	t.Helper()
+	g := gatewayWithBilling(t)
+	cfg := g.core.Config()
+	cfg.Sources = []config.Source{
+		{Name: "localzen", BaseURL: "https://free.example.com/v1"},
+		{Name: "commandcode", BaseURL: "https://api.commandcode.ai/v1"},
+	}
+	return g
+}
+
+func rulesDo(t *testing.T, g *Gateway, method, body string) *httptest.ResponseRecorder {
+	t.Helper()
+	return doReq(t, g, method, "/api/plugins/billing/rules", body)
+}
+
+// TestBillingRulesCRUD covers the whole editor loop against the real HTTP
+// surface: read, add, replace-by-URL, delete. The replace-by-URL case is the
+// one worth pinning — two rules for one URL would make the first unreachable
+// (first match wins) without any error, which is exactly the kind of silent
+// ambiguity an editor must not be able to create.
+func TestBillingRulesCRUD(t *testing.T) {
+	g := rulesGateway(t)
+	seed := `{"billing":{"active":"p1","profiles":[{"id":"p1","rules":[
+      {"url":"https://free.example.com/v1","mode":"free"}]}]}}`
+	if rr := rulesDo(t, g, http.MethodPut, seed); rr.Code != http.StatusOK {
+		t.Fatalf("seed rules = %d: %s", rr.Code, rr.Body.String())
+	}
+
+	// Add a token-priced rule for the second URL.
+	add := `{"profile":"p1","rule":{"url":"https://api.commandcode.ai/v1","mode":"token",
+      "models":{"deepseek-v4.1-flash":{"prompt":"0.15","completion":"0.60"}}}}`
+	if rr := rulesDo(t, g, http.MethodPost, add); rr.Code != http.StatusOK {
+		t.Fatalf("add rule = %d: %s", rr.Code, rr.Body.String())
+	}
+	var got billingRulesPayload
+	if err := json.Unmarshal(rulesDo(t, g, http.MethodGet, "").Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if len(got.DSL.Profiles[0].Rules) != 2 {
+		t.Fatalf("after add there are %d rules, want 2", len(got.DSL.Profiles[0].Rules))
+	}
+
+	// Adding the SAME url again must replace, not append.
+	if rr := rulesDo(t, g, http.MethodPost, `{"profile":"p1","rule":{"url":"https://api.commandcode.ai/v1","mode":"free"}}`); rr.Code != http.StatusOK {
+		t.Fatalf("re-add same url = %d: %s", rr.Code, rr.Body.String())
+	}
+	if err := json.Unmarshal(rulesDo(t, g, http.MethodGet, "").Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if n := len(got.DSL.Profiles[0].Rules); n != 2 {
+		t.Fatalf("★ re-adding an existing url appended instead of replacing: %d rules", n)
+	}
+	var mode string
+	for _, r := range got.DSL.Profiles[0].Rules {
+		if r.URL == "https://api.commandcode.ai/v1" {
+			mode = r.Mode
+		}
+	}
+	if mode != "free" {
+		t.Errorf("re-added rule mode = %q, want free (the replacement must win)", mode)
+	}
+
+	// Delete by URL.
+	if rr := rulesDo(t, g, http.MethodDelete, `{"profile":"p1","url":"https://api.commandcode.ai/v1"}`); rr.Code != http.StatusOK {
+		t.Fatalf("delete rule = %d: %s", rr.Code, rr.Body.String())
+	}
+	if err := json.Unmarshal(rulesDo(t, g, http.MethodGet, "").Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if len(got.DSL.Profiles[0].Rules) != 1 {
+		t.Errorf("after delete there are %d rules, want 1", len(got.DSL.Profiles[0].Rules))
+	}
+}
+
+// TestBillingRulesRejectUnknownURLWithWarning is the silent-failure guard: a
+// rule whose URL matches no source prices nothing, so every request on it
+// falls into unpriced. The API must say so instead of accepting it silently.
+func TestBillingRulesRejectUnknownURLWithWarning(t *testing.T) {
+	g := rulesGateway(t)
+	if rr := rulesDo(t, g, http.MethodPut, `{"billing":{"active":"p","profiles":[{"id":"p",
+      "rules":[{"url":"https://typo.example.com/v1","mode":"free"}]}]}}`); rr.Code != http.StatusOK {
+		t.Fatalf("seed = %d: %s", rr.Code, rr.Body.String())
+	}
+	var got billingRulesPayload
+	if err := json.Unmarshal(rulesDo(t, g, http.MethodGet, "").Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if len(got.Warnings) == 0 {
+		t.Fatal("★ a rule matching no configured source was accepted with no warning — " +
+			"it will price nothing and every request will be recorded as unpriced")
+	}
+	if !strings.Contains(got.Warnings[0], "typo.example.com") {
+		t.Errorf("warning does not name the offending url: %v", got.Warnings)
+	}
+	// The editor needs the real URL list to offer suggestions.
+	if len(got.URLs) != 2 {
+		t.Errorf("urls offered to the editor = %v, want the 2 configured base_urls", got.URLs)
+	}
+}
+
+// TestBillingRulesInvalidIsRejectedNotPersisted checks the write path refuses
+// bad input and leaves the previous rules in place — an editor that clears the
+// table on a typo loses the price table.
+func TestBillingRulesInvalidIsRejectedNotPersisted(t *testing.T) {
+	g := rulesGateway(t)
+	seed := `{"billing":{"active":"p","profiles":[{"id":"p","rules":[
+      {"url":"https://free.example.com/v1","mode":"free"}]}]}}`
+	if rr := rulesDo(t, g, http.MethodPut, seed); rr.Code != http.StatusOK {
+		t.Fatalf("seed = %d: %s", rr.Code, rr.Body.String())
+	}
+	// An unknown mode must be refused.
+	if rr := rulesDo(t, g, http.MethodPost, `{"profile":"p","rule":{"url":"https://x/v1","mode":"banana"}}`); rr.Code != http.StatusBadRequest {
+		t.Errorf("invalid mode accepted with %d, want 400", rr.Code)
+	}
+	var got billingRulesPayload
+	if err := json.Unmarshal(rulesDo(t, g, http.MethodGet, "").Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if len(got.DSL.Profiles[0].Rules) != 1 {
+		t.Errorf("★ a rejected rule was persisted anyway: %d rules", len(got.DSL.Profiles[0].Rules))
+	}
+	// An unparseable price must be refused too, not silently priced at zero.
+	if rr := rulesDo(t, g, http.MethodPost, `{"profile":"p","rule":{"url":"https://y/v1","mode":"token",
+      "models":{"m":{"prompt":"free","completion":"0.6"}}}}`); rr.Code != http.StatusBadRequest {
+		t.Errorf("non-numeric price accepted with %d, want 400", rr.Code)
+	}
+}
+
+// TestBillingRulesPersistAndReachThePlugin is the point of the whole
+// endpoint, and mutation verification showed the earlier CRUD tests missed it
+// twice: dropping cfg.Save() and dropping the SetState injection both left
+// every test green. Both failures are invisible in a GET — the response echoes
+// the in-memory rules either way — so this test checks the two consequences
+// that actually matter:
+//
+//  1. persistence: the rule must be in the config file on disk, because an
+//     edit that evaporates on restart is a lie the UI told the operator;
+//  2. effect: the billing plugin must now PRICE the source (its published
+//     prices table has a non-zero entry), because a rule that is stored but
+//     never injected prices nothing at all.
+func TestBillingRulesPersistAndReachThePlugin(t *testing.T) {
+	g := rulesGateway(t)
+	cfg := g.core.Config()
+
+	seed := `{"billing":{"active":"p","profiles":[{"id":"p","rules":[
+      {"url":"https://free.example.com/v1","mode":"free"},
+      {"url":"https://api.commandcode.ai/v1","mode":"token",
+       "models":{"deepseek-v4.1-flash":{"prompt":"0.15","completion":"0.60"}}}]}]}}`
+	if rr := rulesDo(t, g, http.MethodPut, seed); rr.Code != http.StatusOK {
+		t.Fatalf("seed = %d: %s", rr.Code, rr.Body.String())
+	}
+
+	// 1. persisted to the file the process loaded from.
+	raw, err := os.ReadFile(cfg.Path)
+	if err != nil {
+		t.Fatalf("read config: %v", err)
+	}
+	if !strings.Contains(string(raw), "commandcode.ai") {
+		t.Errorf("★ the rule is not in config.yaml on disk — an edit that does not "+
+			"survive a restart:\n%s", raw)
+	}
+	reloaded, err := config.Load(cfg.Path)
+	if err != nil {
+		t.Fatalf("reload saved config: %v", err)
+	}
+	if reloaded.BillingDSL == nil || len(reloaded.BillingDSL.Profiles) == 0 {
+		t.Fatalf("saved config has no billing DSL: %s", raw)
+	}
+	// A round-trip through YAML must not lose the price strings.
+	found := false
+	for _, r := range reloaded.BillingDSL.Profiles[0].Rules {
+		if strings.Contains(r.URL, "commandcode") {
+			found = true
+			if m, ok := r.Models["deepseek-v4.1-flash"]; !ok || m.Prompt != "0.15" {
+				t.Errorf("price did not survive the YAML round-trip: %+v", m)
+			}
+		}
+	}
+	if !found {
+		t.Errorf("the token rule is missing after reload:\n%s", raw)
+	}
+
+	// 2. reached the plugin. Prices live in a separate Lua field from `state`,
+	// and the GET /state endpoint returns the published state together with
+	// the prices sidecar — which is exactly what the plugin's own UI reads.
+	state := g.core.Plugins().State("billing")
+	if state == nil {
+		t.Fatalf("billing plugin published no state")
+	}
+	rr := doReq(t, g, http.MethodGet, "/api/plugins/billing/state", "")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("GET state = %d: %s", rr.Code, rr.Body.String())
+	}
+	var envelope struct {
+		Prices map[string]interface{} `json:"prices"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &envelope); err != nil {
+		t.Fatalf("decode state envelope: %v", err)
+	}
+	prices := envelope.Prices
+	if prices == nil {
+		t.Fatalf("state payload carries no prices")
+	}
+	sources, ok := prices["sources"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("prices has no sources map: %#v", prices)
+	}
+	cc, ok := sources["commandcode"]
+	if !ok {
+		t.Fatalf("★ the rule was saved but never injected: prices has %v, "+
+			"no commandcode entry. Every request on it stays unpriced.", keysOf(sources))
+	}
+	if cc == nil {
+		t.Error("commandcode price entry is nil")
+	}
+	// The free rule must price localzen, and the two must not be conflated.
+	if _, ok := sources["localzen"]; !ok {
+		t.Errorf("the free rule did not reach the plugin: %v", keysOf(sources))
+	}
+	// The injected price must be per-token, i.e. 0.15 USD/M => 1.5e-7.
+	// Asserting the exact magnitude catches a unit error (per-million vs
+	// per-token), which would be off by a factor of a million and still look
+	// like "a number".
+	models, ok := cc.(map[string]interface{})["models"].(map[string]interface{})
+	if ok {
+		if m, ok := models["deepseek-v4.1-flash"].(map[string]interface{}); ok {
+			p, _ := m["prompt"].(float64)
+			if p <= 0 || p > 1e-6 {
+				t.Errorf("injected prompt price = %v, want per-token (~1.5e-7); "+
+					"a per-million value here would over-bill by 1e6x", p)
+			}
+		}
+	}
+}
+
+func keysOf(m map[string]interface{}) []string {
+	out := make([]string, 0, len(m))
+	for k := range m {
+		out = append(out, k)
+	}
+	sort.Strings(out)
+	return out
+}
+
+// TestBillingRulesRequireAdmin keeps price rewriting behind the admin role,
+// like every other plugin write: prices are configuration, not user data.
+func TestBillingRulesRequireAdmin(t *testing.T) {
+	g := rulesGateway(t)
+	rec, err := g.core.CreateKey("viewer", "user", nil, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	userKey := rec.Key
+	for _, m := range []string{http.MethodPut, http.MethodPost, http.MethodDelete} {
+		req, _ := http.NewRequest(m, "/api/plugins/billing/rules", strings.NewReader(`{}`))
+		req.Header.Set("Authorization", "Bearer "+userKey)
+		req.Header.Set("Content-Type", "application/json")
+		if rr := newRecorderFor(t, g, req); rr.Code != http.StatusForbidden {
+			t.Errorf("user %s rules = %d, want 403", m, rr.Code)
+		}
+	}
+}

internal/gateway/chat.go

@@ -7,11 +7,14 @@ import (
 	"fmt"
 	"log"
 	"net/http"
+	"regexp"
+	"strconv"
 	"strings"
 	"sync/atomic"
 	"time"
 
 	"llmsproxy/internal/config"
+	"llmsproxy/internal/lua"
 	"llmsproxy/internal/provider"
 	"llmsproxy/internal/scheduler"
 	"llmsproxy/internal/types"
@@ -28,6 +31,9 @@ type chatRequest struct {
 	ToolChoice      interface{}            `json:"tool_choice,omitempty"`
 	DisableThinking bool                   `json:"disable_thinking"`
 	ExtraBody       map[string]interface{} `json:"extra_body,omitempty"`
+
+	// PromptCacheKey 是 OpenAI 原生的缓存键；部分客户端用它携带会话标识。
+	PromptCacheKey string `json:"prompt_cache_key,omitempty"`
 }
 
 // ChatCompletion is the non-streaming OpenAI response object.
@@ -38,6 +44,10 @@ type ChatCompletion struct {
 	Model   string            `json:"model"`
 	Choices []ChatChoice      `json:"choices"`
 	Usage   *types.TokenUsage `json:"usage,omitempty"`
+	// Cost is the upstream-reported charge for this request, passed through
+	// verbatim (OpenCode reports it as a decimal string, always "0" on the
+	// flat-rate Go subscription). Absent when the upstream reports nothing.
+	Cost string `json:"cost,omitempty"`
 }
 
 type ChatChoice struct {
@@ -62,6 +72,10 @@ type ChatChunk struct {
 	// Usage is sent in the final chunk of a stream (empty choices) so
 	// OpenAI-compatible clients can read token usage.
 	Usage *types.TokenUsage `json:"usage,omitempty"`
+	// Cost is the upstream-reported charge for this request, passed through
+	// verbatim (OpenCode reports it as a decimal string). Emitted on the
+	// terminal chunk, mirroring OpenCode's own {"choices":[],"cost":"0"}.
+	Cost string `json:"cost,omitempty"`
 }
 
 type ChunkChoice struct {
@@ -115,13 +129,15 @@ func (g *Gateway) resolveCands(ctx context.Context, req *chatRequest) ([]*provid
 }
 
 // filterCandsByModels keeps only providers exposing at least one model of the
-// scope (used for user keys with a restricted model scope). An "AUTO" scope
-// entry means the key is allowed to use any model. Scope entries with a
-// Source pinned to a specific upstream narrow the candidates to that source
+// scope (used for user keys with a restricted model scope). Scope entries with
+// a Source pinned to a specific upstream narrow the candidates to that source
 // for the matching model.
+//
+// An "AUTO" scope entry only allows the AUTO routing mode; it does NOT grant
+// access to specific models.
 func filterCandsByModels(cands []*provider.Provider, allow []config.ModelScope) []*provider.Provider {
 	for _, m := range allow {
-		if m.Model == "" || strings.EqualFold(m.Model, "AUTO") {
+		if m.Model == "" {
 			return cands
 		}
 	}
@@ -152,20 +168,18 @@ func filterCandsByModels(cands []*provider.Provider, allow []config.ModelScope)
 	return out
 }
 
-// intersectModels restricts a model list to the scope (preserving order). An
-// "AUTO" scope entry grants every model.
+// intersectModels restricts a model list to the scope (preserving order).
+// An "AUTO" scope entry only allows the AUTO routing mode; it does NOT grant
+// every model.
 func intersectModels(models []string, allow []config.ModelScope) []string {
 	allowed := make(map[string]bool, len(allow))
-	any := false
 	for _, m := range allow {
-		if m.Model == "" || strings.EqualFold(m.Model, "AUTO") {
-			any = true
-			break
+		if m.Model == "" {
+			return models
+		}
+		if !strings.EqualFold(m.Model, "AUTO") {
+			allowed[m.Model] = true
 		}
-		allowed[m.Model] = true
-	}
-	if any {
-		return models
 	}
 	out := make([]string, 0, len(models))
 	seen := map[string]bool{}
@@ -179,48 +193,110 @@ func intersectModels(models []string, allow []config.ModelScope) []string {
 }
 
 // checkModelScope validates the effective model against the key's model scope
-// and token quota. Returns an error message when rejected. A scope entry with
-// model "AUTO" grants all models; its quota caps the key's total tokens.
+// and token quota. Returns an error message when rejected.
+//
+// A scope entry with model "AUTO" only allows requests where the effective
+// model is AUTO (the routing mode). It does NOT grant access to specific model
+// ids — that requires an explicit scope entry for the model.
+// checkModelScope keeps its string-returning signature for callers that only
+// need to know whether the request may proceed.
 func (g *Gateway) checkModelScope(ctx context.Context, model string) string {
+	if q := g.checkQuota(ctx, model); q != nil {
+		return q.msg
+	}
+	return ""
+}
+
+// quotaRejection is a quota verdict: the message plus whether the client
+// should retry. A spent quota is a rate limit (429 + Retry-After), not a
+// permission failure (403): a client that sees 403 gives up on the key, while
+// one that sees 429 with a retry hint waits and resumes when the window rolls
+// over.
+type quotaRejection struct {
+	msg   string
+	retry int64 // seconds until the window resets; 0 = unknown
+}
+
+func (q *quotaRejection) Error() string { return q.msg }
+
+// checkQuota validates the effective model against the key's model scope and
+// that entry's quota. Returns nil when the request may proceed.
+//
+// Quotas are per scope entry, never key-wide: a model whose budget is spent
+// is refused on its own while the key's other models keep working. The verdict
+// carries the remaining seconds of the reset window so a spent budget answers
+// 429 + Retry-After (come back when it rolls over) instead of 403 (which reads
+// as "this key may never use this model" and makes clients give up).
+func (g *Gateway) checkQuota(ctx context.Context, model string) *quotaRejection {
 	allow := g.allowedModels(ctx)
 	if allow == nil {
-		return ""
-	}
-	for _, sc := range allow {
-		if sc.Model != "" && strings.EqualFold(sc.Model, "AUTO") {
-			if sc.TokenQuota > 0 {
-				used := g.scopeTokens(ctx, sc)
-				if used >= sc.TokenQuota {
-					return fmt.Sprintf("token quota exceeded (%d/%d)", used, sc.TokenQuota)
-				}
-			}
-			return ""
-		}
+		return nil
 	}
 	for _, sc := range allow {
 		if sc.Model != model {
 			continue
 		}
+		win := AutoPeriodSeconds(sc.Period, sc.Hours)
+		k := keyID(reqKey(ctx))
 		if sc.TokenQuota > 0 {
-			used := g.scopeTokens(ctx, sc)
-			if used >= sc.TokenQuota {
-				return fmt.Sprintf("token quota exceeded for %q (%d/%d)", model, used, sc.TokenQuota)
+			if used := g.scopeTokens(ctx, sc); used >= sc.TokenQuota {
+				return &quotaRejection{
+					msg:   fmt.Sprintf("token quota exceeded for %q (%d/%d)", model, used, sc.TokenQuota),
+					retry: AutoSecondsToReset(sc.Period, sc.Hours),
+				}
 			}
 		}
-		return ""
+		if sc.ReqQuota > 0 {
+			if used := g.stats.KeyWindowReqs(k, win); used >= sc.ReqQuota {
+				return &quotaRejection{
+					msg:   fmt.Sprintf("request quota exceeded for %q (%d/%d)", model, used, sc.ReqQuota),
+					retry: AutoSecondsToReset(sc.Period, sc.Hours),
+				}
+			}
+		}
+		return nil
 	}
-	return fmt.Sprintf("model %q is not allowed for this key", model)
+	// A scope that simply does not list the model is a different problem from
+	// a quota being exhausted, and the message has to say which: telling an
+	// operator "not allowed for this key" when they configured a quota sends
+	// them to the wrong setting. AUTO is checked here before the chain is even
+	// consulted, so this is the message that surfaces for AUTO.
+	if isAuto(model) {
+		return &quotaRejection{msg: fmt.Sprintf(
+			"model %q is not in this key's model scope, so it cannot use AUTO; "+
+				"add %q to the key's models or remove the scope restriction", model, model)}
+	}
+	return &quotaRejection{msg: fmt.Sprintf("model %q is not allowed for this key", model)}
 }
 
-// scopeTokens returns the tokens a scope entry has consumed within its reset
-// window (total for AUTO / per model otherwise).
+// quotaWindowSuffix describes a quota's reset window for an error message, so
+// a rejected caller can tell a permanent block from one that clears in an hour.
+func quotaWindowSuffix(period string, hours int64) string {
+	switch {
+	case period == "hour":
+		return ", resets hourly"
+	case period == "week":
+		return ", resets weekly"
+	case period == "month":
+		return ", resets monthly"
+	case period == "nhour" && hours > 1:
+		return fmt.Sprintf(", resets every %dh", hours)
+	}
+	return ""
+} // scopeTokens returns the tokens this key consumed on the scope entry's model
+// within its reset window, isolated per key. For an AUTO entry the cap covers
+// everything the key routed through AUTO; for a model entry it covers that
+// model only.
+//
+// It reads the per-key hourly buckets rather than the key-blind model
+// buckets, so one key's usage can never exhaust another's quota.
 func (g *Gateway) scopeTokens(ctx context.Context, sc config.ModelScope) int64 {
 	k := keyID(reqKey(ctx))
-	if sc.Model != "" && strings.EqualFold(sc.Model, "AUTO") {
-		return g.stats.KeyTokens(k)
-	}
 	win := AutoPeriodSeconds(sc.Period, sc.Hours)
-	return g.stats.WindowTokens(sc.Model, sc.Source, win)
+	if sc.Model != "" && strings.EqualFold(sc.Model, "AUTO") {
+		return g.stats.KeyWindowTokens(k, win)
+	}
+	return g.stats.KeyWindowModelTokens(k, sc.Model, sc.Source, win)
 }
 
 // hasScopeModel reports whether a model (possibly with a "source-model" /
@@ -229,18 +305,41 @@ func (g *Gateway) scopeTokens(ctx context.Context, sc config.ModelScope) int64 {
 // source that actually serves the bare model (via Registry.EffectiveModel), so
 // model ids that themselves contain separators (e.g. "deepseek-v4-flash-free")
 // are never corrupted (P10-2).
+//
+// A scope entry with model "AUTO" only matches the literal AUTO routing mode;
+// it does NOT grant access to specific model ids.
 func (g *Gateway) hasScopeModel(list []config.ModelScope, s string) bool {
 	if r := g.core.Registry(); r != nil {
 		s = r.EffectiveModel(s)
 	}
 	for _, x := range list {
-		if x.Model == s || (x.Model != "" && strings.EqualFold(x.Model, "AUTO")) {
+		if x.Model == s {
 			return true
 		}
 	}
 	return false
 }
 
+// writeScopeReject answers a model-scope or quota rejection. A spent quota is
+// 429 (rate_limit_exceeded) with Retry-After, so a client waits and resumes
+// after the reset; a model the key may not use stays 403 (model_not_allowed),
+// because retrying cannot help.
+func (g *Gateway) writeReject(w http.ResponseWriter, q *quotaRejection) {
+	if q.retry > 0 {
+		w.Header().Set("Retry-After", strconv.FormatInt(q.retry, 10))
+		writeError(w, http.StatusTooManyRequests, "rate_limit_exceeded", q.msg)
+		return
+	}
+	// no window to wait for: the cap is either permanent or key-wide with no
+	// period. "rate_limit_exceeded" still says "come back after the operator
+	// raises the cap", which 403 would not.
+	if strings.Contains(q.msg, "quota exceeded") {
+		writeError(w, http.StatusTooManyRequests, "rate_limit_exceeded", q.msg)
+		return
+	}
+	writeError(w, http.StatusForbidden, "model_not_allowed", q.msg)
+}
+
 func (g *Gateway) resolveByModel(model string) ([]*provider.Provider, string) {
 	if isAuto(model) {
 		return g.core.Registry().Resolve("AUTO"), ""
@@ -287,6 +386,8 @@ func (g *Gateway) handleChat(w http.ResponseWriter, r *http.Request) {
 		writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
 		return
 	}
+	// 客户端自带的会话标识（若有）比推导出来的更准，见 clientSessionFromRequest。
+	clientSession := clientSessionFromRequest(r, req.PromptCacheKey)
 	if len(req.Messages) == 0 {
 		writeError(w, http.StatusBadRequest, "invalid_request", "messages is required")
 		return
@@ -295,14 +396,28 @@ func (g *Gateway) handleChat(w http.ResponseWriter, r *http.Request) {
 	if model == "" {
 		model = g.core.DefaultModel()
 	}
+	// request_start fires for EVERY chat request, on both the AUTO and the
+	// direct path, and it fires BEFORE the quota / model-scope gates on
+	// purpose: a plugin that counts volume or audits traffic must also see the
+	// requests the gateway rejected, otherwise "requests accepted" would be all
+	// it could ever report. It sits after authentication (so the key and role in
+	// the payload are real) and after the messages check (a body with no
+	// messages is not a chat request at all).
+	//
+	// Calling it here rather than inside each branch is what keeps the two paths
+	// honest: an earlier version called it only from the AUTO branch, so every
+	// direct (model-pinned) request silently skipped it. That was caught by
+	// TestHooksFireOnRealDirectChat, not by reading the code.
+	g.fireStart(r.Context(), &req, "chat", model, len(req.Messages), len(req.Tools))
 	if isAuto(model) {
-		chain := g.core.AutoChain()
+		// The key's own chain wins; keys without one inherit the global chain.
+		chain, _ := g.core.AutoChainFor(reqKey(r.Context()))
 		if chain == nil || len(chain.Tiers) == 0 {
 			writeError(w, http.StatusServiceUnavailable, "no_provider", "no auto slot configured")
 			return
 		}
-		if msg := g.checkModelScope(r.Context(), "AUTO"); msg != "" {
-			writeError(w, http.StatusForbidden, "model_not_allowed", msg)
+		if q := g.checkQuota(r.Context(), "AUTO"); q != nil {
+			g.writeReject(w, q)
 			return
 		}
 		ctx := r.Context()
@@ -317,6 +432,7 @@ func (g *Gateway) handleChat(w http.ResponseWriter, r *http.Request) {
 			ToolChoice:      req.ToolChoice,
 			DisableThinking: req.DisableThinking,
 			ExtraBody:       req.ExtraBody,
+			ClientSession:   clientSession,
 		}
 		rec := &Req{
 			Key:  keyID(reqKey(ctx)),
@@ -346,6 +462,11 @@ func (g *Gateway) handleChat(w http.ResponseWriter, r *http.Request) {
 			return
 		}
 	}
+	// Note: a key whose scope excludes AUTO is rejected earlier, by
+	// checkQuota(ctx, "AUTO") above, which produces the same message. That
+	// ordering matters — checkQuota runs before the chain is consulted, so it
+	// is the check an AUTO request actually hits. Duplicating it here would be
+	// dead code.
 	cands, effective := g.resolveCands(r.Context(), &req)
 	if len(cands) == 0 {
 		writeError(w, http.StatusNotFound, "model_not_found", fmt.Sprintf("model %q is not configured", model))
@@ -354,8 +475,8 @@ func (g *Gateway) handleChat(w http.ResponseWriter, r *http.Request) {
 	if effective == "" {
 		effective = firstModel(cands[0])
 	}
-	if msg := g.checkModelScope(r.Context(), effective); msg != "" {
-		writeError(w, http.StatusForbidden, "model_not_allowed", msg)
+	if q := g.checkQuota(r.Context(), effective); q != nil {
+		g.writeReject(w, q)
 		return
 	}
 	ctx := r.Context()
@@ -372,6 +493,7 @@ func (g *Gateway) handleChat(w http.ResponseWriter, r *http.Request) {
 		ToolChoice:      req.ToolChoice,
 		DisableThinking: req.DisableThinking,
 		ExtraBody:       req.ExtraBody,
+		ClientSession:   clientSession,
 	}
 	rec := &Req{
 		Key:    keyID(reqKey(ctx)),
@@ -517,21 +639,115 @@ func upstreamErrStatus(err error) int {
 	return http.StatusBadGateway
 }
 
+// overflowMarkers 匹配上游「上下文超窗」类措辞。
+//
+// 上游写法五花八门，且**不在 pi 客户端的识别列表里**。pi 靠
+// @earendil-works/pi-ai 的 OVERFLOW_PATTERNS 判断超窗并据此触发压缩重试，
+// 而 justworker 返回的是「请精简对话历史…(Context window is full…)」——
+// 与那 25 条正则一条都不匹配，于是 pi 既不压缩也不重试，只把它当成一条
+// 普通上游错误。这里把可识别的超窗措辞归一化成 pi 一定认得的标记。
+var overflowMarkers = []*regexp.Regexp{
+	regexp.MustCompile(`(?i)context[ _-]?window is full`),
+	regexp.MustCompile(`(?i)context[_ ]length[_ ]exceeded`),
+	regexp.MustCompile(`(?i)exceeds? the context window`),
+	regexp.MustCompile(`(?i)maximum context length`),
+	regexp.MustCompile(`(?i)reduce the length of the messages`),
+	regexp.MustCompile(`(?i)too many tokens`),
+	regexp.MustCompile(`(?i)token limit exceeded`),
+	regexp.MustCompile(`请精简对话历史`),
+	regexp.MustCompile(`上下文(长度)?超(出|限)`),
+	regexp.MustCompile(`对话历史过长`),
+}
+
+// overflowCanonical 命中 pi 的 /context[_ ]length[_ ]exceeded/i。
+//
+// 注意：这个前缀只是必要条件，不是充分条件——pi 还会先用
+// NON_OVERFLOW_PATTERNS 排除整条消息，见 overflowClientMessage。
+const overflowCanonical = "context_length_exceeded"
+
+// looksLikeOverflow 判断错误文本是否属于上下文超窗。
+func looksLikeOverflow(s string) bool {
+	for _, re := range overflowMarkers {
+		if re.MatchString(s) {
+			return true
+		}
+	}
+	return false
+}
+
 // clientUpstreamErr collapses upstream failure details into a short message
 // for the client: per-tier bodies (WAF HTML pages, quota payloads, ...) stay
 // in rec.Err / the stats API and the server log instead of the response.
 // Per-tier one-line reasons are kept (quota/cooling skips carry no error body
 // and are the actionable part); each is capped so HTML dumps can't leak.
+//
+// 超窗会被归一化成「干净」的 overflowCanonical 消息，见 overflowClientMessage。
 func clientUpstreamErr(err error) string {
 	log.Printf("[gateway] upstream failure surfaced to client: %v", err)
+	if looksLikeOverflow(err.Error()) {
+		return overflowClientMessage(err)
+	}
+	return upstreamErrSummary(err)
+}
+
+// overflowHint 是给客户端的超窗说明，刻意只用 pi 认得的措辞。
+const overflowHint = "context window is full; reduce the length of the messages"
+
+// clientSessionFromRequest 取客户端自带的会话标识，拿不到时返回 ""。
+//
+// 通用 OpenAI 客户端默认不带会话 id，但 pi 支持：只要 provider 的 compat 里打开
+// sendSessionAffinityHeaders，pi 就会把**平台真实的会话 id**（uuidv7，整个会话
+// 恒定）放到 x-session-affinity / x-client-request-id 上（sessionAffinityFormat
+// 为 openrouter 时是 x-session-id，为 openai 时是 session_id）。
+// 有了它就不必再从首条 user 消息推导会话指纹（那只在客户端不发会话 id 时才作为
+// 退路，且历史压缩后会漂移）。prompt_cache_key 是 OpenAI 原生的缓存键，
+// 客户端若带也一并当会话用。
+//
+// 刻意不采纳 x-client-request-id：名字含 request，部分客户端每请求都换，
+// 拿它当会话会让上游缓存永不命中。pi 总会同时发 x-session-affinity，够用。
+func clientSessionFromRequest(r *http.Request, bodyKey string) string {
+	for _, h := range []string{"X-Session-Affinity", "X-Session-Id", "Session-Id"} {
+		if v := strings.TrimSpace(r.Header.Get(h)); v != "" {
+			return v
+		}
+	}
+	return strings.TrimSpace(bodyKey)
+}
+
+// overflowClientMessage 为超窗失败生成「干净」的客户端消息。
+//
+// 为什么不能沿用 upstreamErrSummary：pi 的 isContextOverflow 先查
+// NON_OVERFLOW_PATTERNS（/rate limit/、/too many requests/、Bedrock 前缀），
+// 一旦命中就直接判为「非超窗」——**哪怕消息里已经有 context_length_exceeded**，
+// pi 也不会压缩重试。而 AUTO 链的失败消息天生是多 tier 原因的拼接：
+// 超窗 tier（gozen 400 maximum context length）常与配额/限流 tier
+// （429 token plan exhausted、cooling、no free slot）同时出现。
+// 把明细原样带出去，等于让一条限流 tier 的措辞反过来封杀超窗识别。
+// 所以这里只保留「超窗」措辞 + 是哪个源超的窗，其余一律不带。
+func overflowClientMessage(err error) string {
+	var ce *scheduler.ChainErr
+	if errors.As(err, &ce) {
+		for _, t := range ce.Tiers {
+			if looksLikeOverflow(t.Err.Error()) {
+				return fmt.Sprintf("%s: %s (%s/%s)", overflowCanonical, overflowHint, t.Source, t.Model)
+			}
+		}
+	}
+	return fmt.Sprintf("%s: %s", overflowCanonical, overflowHint)
+}
+
+// upstreamErrSummary 把上游失败压成一行短消息。
+func upstreamErrSummary(err error) string {
 	var ce *scheduler.ChainErr
 	if errors.As(err, &ce) {
 		parts := make([]string, 0, len(ce.Tiers)+len(ce.Skipped))
 		for _, t := range ce.Tiers {
-			parts = append(parts, types.OneLine(fmt.Sprintf("%s/%s: %v", t.Source, t.Model, t.Err), 80))
+			// 160 而不是 80：短诊断词（"Context window is full"）常落在尾部，
+			// 80 字节按字节截断正好会把它切掉，超窗就再也认不出来。
+			parts = append(parts, types.OneLine(fmt.Sprintf("%s/%s: %v", t.Source, t.Model, t.Err), 160))
 		}
 		for _, sk := range ce.Skipped {
-			parts = append(parts, types.OneLine(sk, 80))
+			parts = append(parts, types.OneLine(sk, 160))
 		}
 		msg := strings.Join(parts, "; ")
 		if len(msg) > 300 {
@@ -573,6 +789,18 @@ func recordChatUsage(rec *Req, req *types.ChatRequest, resp *types.UnifiedRespon
 	// Cache accounting: whichever source format the adapter normalized into
 	// (prompt_tokens_details.cached_tokens or legacy prompt_cache_hit_tokens),
 	// read it back so the request record carries the hit/miss split.
+	// reasoning_tokens is a SUBSET of completion_tokens and is billed at the
+	// completion rate by every provider measured, so it must NOT be added to
+	// Compl. Recording it separately is what lets a price rule split thinking
+	// from output for the providers that do price them differently, and what
+	// keeps the billing plugin from guessing.
+	//
+	// Leaving it unrecorded is worse than useless: the adapters already parse it
+	// and forward it (opencodego/gemini/anthropic all do), so the number exists
+	// in the response and is thrown away at the one place that could keep it.
+	if d := resp.TokenUsage.CompletionTokensDetails; d != nil && d.ReasoningTokens > 0 {
+		rec.Reasoning = int64(d.ReasoningTokens)
+	}
 	if d := resp.TokenUsage.PromptTokensDetails; d != nil && d.CachedTokens > 0 {
 		rec.CacheHit = int64(d.CachedTokens)
 	}
@@ -606,6 +834,7 @@ func writeChatCompletion(w http.ResponseWriter, resp *types.UnifiedResponse, mod
 		Created: time.Now().Unix(),
 		Model:   modelName,
 		Choices: []ChatChoice{{Index: 0, Message: msg, FinishReason: resp.FinishReason}},
+		Cost:    resp.Cost,
 	}
 	if resp.TokenUsage.Total > 0 || resp.TokenUsage.Prompt > 0 || resp.TokenUsage.Completion > 0 {
 		out.Usage = &resp.TokenUsage
@@ -630,6 +859,7 @@ func (g *Gateway) singleChat(w http.ResponseWriter, ctx context.Context, cands [
 	recordChatUsage(rec, req, resp)
 	rec.Source = usedSrc
 	rec.Model = usedModel
+	g.fireRouted(ctx, "chat", usedSrc, usedModel, -1, false)
 	// Non-streaming: the whole response arrives at once, so TTFB equals
 	// the total latency.
 	rec.FirstByteMs = rec.LatMs
@@ -637,7 +867,18 @@ func (g *Gateway) singleChat(w http.ResponseWriter, ctx context.Context, cands [
 	writeChatCompletion(w, resp, effective)
 }
 
-// writeRec records a finished request (audit + aggregates).
+// writeRec records a finished request (audit + aggregates) and fires the
+// plugin request_end stage.
+//
+// This is the ONE place every request passes through on its way out, which is
+// what makes it the right hook point: the four entry points (single/stream ×
+// direct/auto) all funnel here, so a plugin sees each request exactly once with
+// its final accounting. Firing earlier would miss the streamed ones (their
+// numbers are only known once the stream finishes), and firing in each entry
+// point would mean four call sites to keep in sync.
+//
+// Hooks run AFTER the record is written: a plugin must not be able to delay or
+// lose the audit trail, and a plugin that throws is contained by Fire.
 func (g *Gateway) writeRec(rec *Req) {
 	if rec == nil {
 		return
@@ -646,6 +887,194 @@ func (g *Gateway) writeRec(rec *Req) {
 		rec.Time = time.Now().UnixMilli()
 	}
 	g.stats.Record(*rec)
+	g.fireEnd(rec)
+}
+
+// fireStart dispatches the plugin request_start stage: the request has been
+// parsed and authorized but no upstream slot has been chosen yet, so `source`
+// is empty. A plugin that only wants volume/acceptance counts can subscribe
+// here and stay out of the per-request hot path entirely.
+func (g *Gateway) fireStart(ctx context.Context, req *chatRequest, kind, model string, msgs, tools int) {
+	ps := g.core.Plugins()
+	if ps == nil || ps.Count() == 0 {
+		return
+	}
+	ps.Fire(lua.StageRequestStart, map[string]interface{}{
+		"stage":          string(lua.StageRequestStart),
+		"type":           kind,
+		"model":          model,
+		"key":            keyID(reqKey(ctx)),
+		"role":           reqRole(ctx),
+		"source":         "",
+		"stream":         req.Stream,
+		"messages_count": msgs,
+		"tools_count":    tools,
+		"ts":             time.Now().Unix(),
+	})
+}
+
+// fireImageStart dispatches request_start for /v1/images/generations.
+//
+// It is a separate function rather than a call to fireStart with a nil
+// chatRequest because the image body has no messages and no tools: passing
+// zeroes through a struct built for chat would invite someone to read a field
+// that simply does not exist on this path.
+func (g *Gateway) fireImageStart(ctx context.Context, model string) {
+	ps := g.core.Plugins()
+	if ps == nil || ps.Count() == 0 {
+		return
+	}
+	ps.Fire(lua.StageRequestStart, map[string]interface{}{
+		"stage":          string(lua.StageRequestStart),
+		"type":           "image",
+		"model":          model,
+		"key":            keyID(reqKey(ctx)),
+		"role":           reqRole(ctx),
+		"source":         "",
+		"stream":         false,
+		"messages_count": 0,
+		"tools_count":    0,
+		"ts":             time.Now().Unix(),
+	})
+}
+
+// fireRouted dispatches the plugin routed stage once a (source, model) slot has
+// been selected. tier is the AUTO tier index, or -1 on the direct path, so a
+// plugin can tell "this came from tier 1" from "this bypassed the chain".
+func (g *Gateway) fireRouted(ctx context.Context, kind, source, model string, tier int, stream bool) {
+	ps := g.core.Plugins()
+	if ps == nil || ps.Count() == 0 {
+		return
+	}
+	ps.Fire(lua.StageRouted, map[string]interface{}{
+		"stage":  string(lua.StageRouted),
+		"type":   kind,
+		"source": source,
+		"model":  model,
+		"key":    keyID(reqKey(ctx)),
+		"tier":   tier,
+		"stream": stream,
+		"ts":     time.Now().Unix(),
+	})
+}
+
+// chainTraceSink adapts a scheduler TraceSink into the plugin chain_step stage.
+//
+// It returns nil when no plugin is loaded, so the scheduler's emit() does a
+// single nil check per event and the AUTO hot path pays nothing on a gateway
+// with no plugins.
+//
+// The events are also accumulated into walk so request_end can carry a compact
+// summary: a plugin that only listens to request_end still learns that a
+// degradation happened, which is the common case for a dashboard that does not
+// want to subscribe to a high-frequency stage.
+func (g *Gateway) chainTraceSink(ctx context.Context, kind string, walk *[]map[string]interface{}) scheduler.TraceSink {
+	ps := g.core.Plugins()
+	if ps == nil || ps.Count() == 0 {
+		return nil
+	}
+	key := keyID(reqKey(ctx))
+	return func(ev scheduler.TraceEvent) {
+		payload := map[string]interface{}{
+			"stage":   string(lua.StageChainStep),
+			"kind":    string(ev.Kind),
+			"type":    kind,
+			"key":     key,
+			"tier":    ev.Tier,
+			"attempt": ev.Attempt,
+		}
+		if ev.Source != "" {
+			payload["source"] = ev.Source
+		}
+		if ev.Model != "" {
+			payload["model"] = ev.Model
+		}
+		if ev.Reason != "" {
+			payload["reason"] = ev.Reason
+		}
+		if ev.Err != "" {
+			payload["error"] = ev.Err
+		}
+		if walk != nil {
+			// Keep the summary bounded: a pathological chain could emit many
+			// steps, and request_end's payload is written to the audit trail.
+			if len(*walk) < maxWalkSummary {
+				*walk = append(*walk, map[string]interface{}{
+					"kind": string(ev.Kind), "tier": ev.Tier,
+					"source": ev.Source, "model": ev.Model, "reason": ev.Reason,
+				})
+			}
+		}
+		ps.Fire(lua.StageChainStep, payload)
+	}
+}
+
+// maxWalkSummary caps how many chain steps request_end carries, so a long
+// degradation cannot inflate every audit record.
+const maxWalkSummary = 12
+
+// tierServed returns the AUTO tier that actually served the request, or -1 when
+// the walk is empty (a direct request) or ended without a selection (total
+// failure). It is the single most useful number for "why did my expensive tier
+// not get used".
+func tierServed(walk []map[string]interface{}) int {
+	for i := len(walk) - 1; i >= 0; i-- {
+		if k, _ := walk[i]["kind"].(string); k == string(scheduler.TraceSelected) {
+			if t, ok := walk[i]["tier"].(int); ok {
+				return t
+			}
+		}
+	}
+	return -1
+}
+
+// fireEnd dispatches the plugin request_end stage for one finished request.
+func (g *Gateway) fireEnd(rec *Req) {
+	ps := g.core.Plugins()
+	if ps == nil || ps.Count() == 0 {
+		return
+	}
+	payload := map[string]interface{}{
+		"stage":             string(lua.StageRequestEnd),
+		"type":              rec.Type,
+		"model":             rec.Model,
+		"source":            rec.Source,
+		"key":               rec.Key,
+		"ok":                rec.OK,
+		"status":            rec.Status,
+		"latency_ms":        rec.LatMs,
+		"first_byte_ms":     rec.FirstByteMs,
+		"prompt_tokens":     rec.Prompt,
+		"completion_tokens": rec.Compl,
+		"cache_hit_tokens":  rec.CacheHit,
+		"cache_miss_tokens": rec.CacheMiss,
+		// Whether UPSTREAM reported a cache number at all. A plugin cannot
+		// infer this from cache_hit_tokens alone: zero hits because nothing was
+		// cached and zero hits because the provider never reports caching are
+		// the same value, and they mean opposite things when you are checking
+		// whether a cache discount is doing anything. The audit record already
+		// carried this (rec.CacheReported); the plugin payload did not, so any
+		// plugin-level cache accounting had to guess.
+		"cache_reported": rec.CacheReported,
+		// A SUBSET of completion_tokens, not an extra. Billing plugins that
+		// add it to completion would double-charge every thinking token.
+		"reasoning_tokens": rec.Reasoning,
+		"image_count":      rec.ImageCount,
+		"error":            rec.Err,
+		"time":             rec.Time,
+		// chain_walk: the AUTO tier-by-tier trace, when the request went
+		// through the chain. Empty for a direct request and for a gateway with
+		// no plugins loaded. Absent rather than empty so a plugin can tell
+		// "no chain" from "chain with no degradation".
+		"degraded":    len(rec.Walk) > 1,
+		"chain_walk":  rec.Walk,
+		"tier_served": tierServed(rec.Walk),
+	}
+	// The merged result is intentionally discarded: request_end is the last
+	// stage, so there is nobody downstream to read a plugin's additions. Plugins
+	// that need to publish derived numbers (the billing plugin) do it in their
+	// OWN state and expose them through the /api/plugins/<name>/state endpoint.
+	ps.Fire(lua.StageRequestEnd, payload)
 }
 
 // mergeUsage combines token usage across stream chunks additively. Some
@@ -725,11 +1154,15 @@ func (g *Gateway) pumpStream(w http.ResponseWriter, rec *Req, chunks <-chan type
 	// source's status-page latency average.
 	rec.FirstByteMs = time.Since(t0).Milliseconds()
 	var lastUsage *types.TokenUsage
+	var lastCost string
 	lastFinish := ""
 	for ck := range chunks {
 		if ck.Usage != nil {
 			lastUsage = mergeUsage(lastUsage, ck.Usage)
 		}
+		if ck.Cost != "" {
+			lastCost = ck.Cost
+		}
 		chunk := ChatChunk{
 			ID: id, Object: "chat.completion.chunk", Created: created, Model: modelName,
 		}
@@ -766,23 +1199,23 @@ func (g *Gateway) pumpStream(w http.ResponseWriter, rec *Req, chunks <-chan type
 	// Final usage chunk (OpenAI standard: empty choices + usage before [DONE]).
 	// Prefer the upstream's exact usage if the stream carried it; fall back to
 	// the gateway's estimate otherwise.
+	// Write the upstream's exact numbers back onto the audit record. Without
+	// this the streamed path kept only the per-chunk byte estimate, so the same
+	// request recorded ~1.4x its real completion tokens (measured: upstream 100,
+	// audit 145) while the non-streaming path recorded 100. Two paths, two
+	// different numbers for one request is a reporting bug, not a rounding one.
+	if lastUsage != nil {
+		if lastUsage.Prompt > 0 {
+			rec.Prompt = int64(lastUsage.Prompt)
+		}
+		if lastUsage.Completion > 0 {
+			rec.Compl = int64(lastUsage.Completion)
+		}
+	}
 	var tut *types.TokenUsage
 	if lastUsage != nil {
 		tut = lastUsage
-		// Write the upstream's cache accounting back onto the request record
-		// so the audit trail carries the hit/miss split for streaming too.
-		if d := tut.PromptTokensDetails; d != nil {
-			rec.CacheReported = true
-			if d.CachedTokens > 0 {
-				rec.CacheHit = int64(d.CachedTokens)
-			}
-		} else if tut.PromptCacheHit > 0 {
-			rec.CacheHit = int64(tut.PromptCacheHit)
-			rec.CacheReported = true
-		}
-		if tut.PromptCacheMiss > 0 {
-			rec.CacheMiss = int64(tut.PromptCacheMiss)
-		}
+		applyStreamUsageDetails(rec, lastUsage)
 	} else if rec.Prompt+rec.Compl > 0 {
 		u := types.TokenUsage{
 			Prompt:     int(rec.Prompt),
@@ -796,6 +1229,14 @@ func (g *Gateway) pumpStream(w http.ResponseWriter, rec *Req, chunks <-chan type
 			ID: id, Object: "chat.completion.chunk", Created: created, Model: modelName,
 			Choices: []ChunkChoice{},
 			Usage:   tut,
+			Cost:    lastCost,
+		})
+	} else if lastCost != "" {
+		// Cost arrived without any usage (OpenCode sends it on its own frame).
+		send(ChatChunk{
+			ID: id, Object: "chat.completion.chunk", Created: created, Model: modelName,
+			Choices: []ChunkChoice{},
+			Cost:    lastCost,
 		})
 	}
 	fmt.Fprintf(w, "data: [DONE]\n\n")
@@ -827,6 +1268,7 @@ func (g *Gateway) streamChat(w http.ResponseWriter, ctx context.Context, cands [
 	// failover it differs from the first candidate). Direct streams previously
 	// discarded it.
 	rec.Source = usedSrc
+	g.fireRouted(ctx, "stream", usedSrc, rec.Model, -1, true)
 	rec.Prompt = estimatePromptTokens(req)
 	g.pumpStream(w, rec, chunks, effective, t0)
 }
@@ -840,8 +1282,11 @@ func (g *Gateway) streamChat(w http.ResponseWriter, ctx context.Context, cands [
 func (g *Gateway) singleChatAuto(w http.ResponseWriter, ctx context.Context, chain *scheduler.Chain, req *types.ChatRequest, rec *Req, quotaExhausted func(*scheduler.Slot) bool) {
 	rec.LatMs = 0
 	t0 := time.Now()
-	resp, usedSrc, usedModel, err := g.core.Scheduler().ChainChat(ctx, chain, req, quotaExhausted)
+	var walk []map[string]interface{}
+	resp, usedSrc, usedModel, err := g.core.Scheduler().ChainChat(ctx, chain, req, quotaExhausted,
+		g.chainTraceSink(ctx, "chat", &walk))
 	rec.LatMs = time.Since(t0).Milliseconds()
+	rec.Walk = walk
 	if err != nil {
 		g.failChat(w, rec, err)
 		g.writeRec(rec)
@@ -852,6 +1297,10 @@ func (g *Gateway) singleChatAuto(w http.ResponseWriter, ctx context.Context, cha
 	recordChatUsage(rec, req, resp)
 	rec.Source = usedSrc
 	rec.Model = usedModel
+	// AUTO has no single tier to report: the chain may have walked several
+	// before this slot served the request, so -2 means "resolved by the chain"
+	// and a plugin can tell that apart from the direct path's -1.
+	g.fireRouted(ctx, "chat", usedSrc, usedModel, -2, false)
 	rec.FirstByteMs = rec.LatMs
 	g.writeRec(rec)
 	writeChatCompletion(w, resp, usedModel)
@@ -870,7 +1319,10 @@ func (g *Gateway) streamChatAuto(w http.ResponseWriter, ctx context.Context, cha
 		rec.LatMs = time.Since(t0).Milliseconds()
 		g.writeRec(rec)
 	}()
-	chunks, usedSrc, usedModel, err := g.core.Scheduler().ChainChatStream(ctx, chain, req, quotaExhausted)
+	var walk []map[string]interface{}
+	chunks, usedSrc, usedModel, err := g.core.Scheduler().ChainChatStream(ctx, chain, req, quotaExhausted,
+		g.chainTraceSink(ctx, "stream", &walk))
+	rec.Walk = walk
 	if err != nil {
 		g.failChat(w, rec, err)
 		return
@@ -879,6 +1331,7 @@ func (g *Gateway) streamChatAuto(w http.ResponseWriter, ctx context.Context, cha
 		rec.Model = usedModel
 	}
 	rec.Source = usedSrc
+	g.fireRouted(ctx, "stream", usedSrc, rec.Model, -2, true)
 	rec.Prompt = estimatePromptTokens(req)
 	g.pumpStream(w, rec, chunks, usedModel, t0)
 }
@@ -901,10 +1354,15 @@ func (g *Gateway) handleImage(w http.ResponseWriter, r *http.Request) {
 	if model == "" {
 		model = g.core.DefaultModel()
 	}
+	// Same rule as the chat path, and for the same reason: an image request is
+	// billable traffic, so a cost plugin must see it. It fires before the
+	// quota/scope gates so rejected image requests are visible too.
+	// messages_count/tools_count are 0: the image request has neither.
+	g.fireImageStart(r.Context(), model)
 	if isAuto(model) {
 		if chain := g.core.AutoImageChain(); chain != nil && len(chain.Tiers) > 0 {
-			if msg := g.checkModelScope(r.Context(), "AUTO"); msg != "" {
-				writeError(w, http.StatusForbidden, "model_not_allowed", msg)
+			if q := g.checkQuota(r.Context(), "AUTO"); q != nil {
+				g.writeReject(w, q)
 				return
 			}
 			done := g.stats.Begin()
@@ -924,9 +1382,13 @@ func (g *Gateway) handleImage(w http.ResponseWriter, r *http.Request) {
 			if usedModel != "" {
 				rec.Model = usedModel // actual image model served, not "AUTO"
 			}
+			g.fireRouted(r.Context(), "image", usedSrc, rec.Model, -2, false)
 			rec.OK = true
 			rec.Status = http.StatusOK
-			rec.Compl = int64(len(resp.ImageData))
+			// Image generation has no token concept. Recording len(ImageData)
+			// (the image COUNT) in completion_tokens mislabels image count as
+			// tokens and feeds it into the token totals; leave it 0.
+			rec.ImageCount = len(resp.ImageData)
 			g.writeRec(rec)
 			writeJSON(w, http.StatusOK, types.ImageGenResponse{
 				Created: time.Now().Unix(),
@@ -950,8 +1412,8 @@ func (g *Gateway) handleImage(w http.ResponseWriter, r *http.Request) {
 		writeError(w, http.StatusServiceUnavailable, "no_provider", "no image source configured")
 		return
 	}
-	if msg := g.checkModelScope(r.Context(), effectiveImageModel(model, cands)); msg != "" {
-		writeError(w, http.StatusForbidden, "model_not_allowed", msg)
+	if q := g.checkQuota(r.Context(), effectiveImageModel(model, cands)); q != nil {
+		g.writeReject(w, q)
 		return
 	}
 	done := g.stats.Begin()
@@ -973,12 +1435,50 @@ func (g *Gateway) handleImage(w http.ResponseWriter, r *http.Request) {
 	if resp.Model != "" {
 		rec.Model = resp.Model // record the actual model served, not the raw request id
 	}
+	g.fireRouted(r.Context(), "image", rec.Source, rec.Model, -1, false)
 	rec.OK = true
 	rec.Status = http.StatusOK
-	rec.Compl = int64(len(resp.ImageData))
+	// Image generation has no token concept — see the AUTO path above.
+	rec.ImageCount = len(resp.ImageData)
 	g.writeRec(rec)
 	writeJSON(w, http.StatusOK, types.ImageGenResponse{
 		Created: time.Now().Unix(),
 		Data:    resp.ImageData,
 	})
 }
+
+// applyStreamUsageDetails mirrors the upstream's reasoning and cache numbers
+// onto a STREAMED request's audit record.
+//
+// It exists as a named function because this was inline in the streaming handler,
+// where it was untestable — and untested it stayed wrong. The non-streaming
+// recorder read completion_tokens_details.reasoning_tokens; this one did not. So
+// a source's streamed requests recorded completion and no reasoning at all while
+// its non-streamed calls recorded both. Nothing failed and no test was red: the
+// audit trail was simply quieter than it should have been.
+//
+// Found by reading the production audit file after wiring a reasoning-heavy
+// source: completion=8166 with reasoning absent, on every streamed request.
+//
+// Reasoning is a SUBSET of completion, never an addition to it. See
+// recordChatUsage for why that matters to billing.
+func applyStreamUsageDetails(rec *Req, u *types.TokenUsage) {
+	if d := u.CompletionTokensDetails; d != nil && d.ReasoningTokens > 0 {
+		rec.Reasoning = int64(d.ReasoningTokens)
+	}
+	// Cache accounting is here for the same reason it is kept apart: "zero hits
+	// because nothing was cached" and "this provider never reports caching" are
+	// the same number and mean opposite things.
+	if d := u.PromptTokensDetails; d != nil {
+		rec.CacheReported = true
+		if d.CachedTokens > 0 {
+			rec.CacheHit = int64(d.CachedTokens)
+		}
+	} else if u.PromptCacheHit > 0 {
+		rec.CacheHit = int64(u.PromptCacheHit)
+		rec.CacheReported = true
+	}
+	if u.PromptCacheMiss > 0 {
+		rec.CacheMiss = int64(u.PromptCacheMiss)
+	}
+}

internal/gateway/chat_reasoning_test.go

@@ -0,0 +1,160 @@
+package gateway
+
+import (
+	"testing"
+
+	"llmsproxy/internal/types"
+)
+
+// TestRecordChatUsageKeepsReasoningTokens pins the reasoning-token plumbing.
+//
+// The adapters already parse completion_tokens_details.reasoning_tokens and
+// forward it (opencodego, gemini, anthropic all do), and types.go has carried
+// the field since it was defined — but nothing ever READ it. The number arrived
+// in the response and was discarded at the one place that could have kept it,
+// so a pricing rule had no way to split thinking from output.
+//
+// The property that matters most: reasoning is a SUBSET of completion. Every
+// provider measured bills it at the completion rate, so a billing plugin that
+// added it to completion_tokens would double-charge every thinking token. The
+// test asserts completion is left alone.
+func TestRecordChatUsageKeepsReasoningTokens(t *testing.T) {
+	rec := &Req{}
+	resp := &types.UnifiedResponse{
+		TokenUsage: types.TokenUsage{
+			Prompt:     100,
+			Completion: 500,
+			CompletionTokensDetails: &types.CompletionTokensDetails{
+				ReasoningTokens: 320,
+			},
+		},
+	}
+	recordChatUsage(rec, &types.ChatRequest{}, resp)
+
+	if rec.Reasoning != 320 {
+		t.Errorf("Reasoning = %d, want 320", rec.Reasoning)
+	}
+	// The subset must not be folded into completion — it is already counted there.
+	if rec.Compl != 500 {
+		t.Errorf("Compl = %d, want 500 (reasoning is a subset, never additive)", rec.Compl)
+	}
+	if rec.Prompt != 100 {
+		t.Errorf("Prompt = %d, want 100", rec.Prompt)
+	}
+}
+
+// TestRecordChatUsageReasoningAbsentStaysZero: "not reported" and "reported as
+// zero" must both leave the field at 0, and neither may invent a number. A
+// plugin reading Reasoning cannot tell them apart, which is why the field is
+// documented as "0 means not reported" rather than "0 means no thinking".
+func TestRecordChatUsageReasoningAbsentStaysZero(t *testing.T) {
+	for _, tc := range []struct {
+		name string
+		u    types.TokenUsage
+	}{
+		{"no details object", types.TokenUsage{Prompt: 10, Completion: 20}},
+		{"details without reasoning", types.TokenUsage{
+			Prompt:                  10,
+			Completion:              20,
+			CompletionTokensDetails: &types.CompletionTokensDetails{},
+		}},
+	} {
+		rec := &Req{}
+		recordChatUsage(rec, &types.ChatRequest{}, &types.UnifiedResponse{TokenUsage: tc.u})
+		if rec.Reasoning != 0 {
+			t.Errorf("%s: Reasoning = %d, want 0", tc.name, rec.Reasoning)
+		}
+		if rec.Compl != 20 {
+			t.Errorf("%s: Compl = %d, want 20", tc.name, rec.Compl)
+		}
+	}
+}
+
+// TestStreamedReasoningIsRecordedOnTheRequest is the streaming counterpart.
+//
+// TestRecordChatUsageKeepsReasoningTokens covers the non-streaming path only,
+// and for a long time the streamed path dropped the number: the non-streaming
+// recorder read TokenUsage.CompletionTokensDetails, while the streaming one read
+// only Prompt and Completion from the terminal usage chunk. (Cache accounting was
+// mirrored; reasoning simply never was.)
+//
+// The symptom is invisible in the response — the client sees a correct answer
+// either way. It only shows in the audit trail, where a reasoning-heavy source's
+// streamed requests recorded completion=8166 and no reasoning at all. Found by
+// reading the production audit file, not by a failing test, which is exactly why
+// the property is pinned here.
+func TestStreamedReasoningIsRecordedOnTheRequest(t *testing.T) {
+	rec := &Req{}
+	applyStreamUsageDetails(rec, &types.TokenUsage{
+		Prompt:     159579,
+		Completion: 751,
+		Total:      160330,
+		CompletionTokensDetails: &types.CompletionTokensDetails{
+			ReasoningTokens: 512,
+		},
+	})
+	if rec.Reasoning != 512 {
+		t.Errorf("streamed Reasoning = %d, want 512 — a streamed request records no "+
+			"reasoning, so a pricing rule that splits thinking from output silently "+
+			"bills it as ordinary output", rec.Reasoning)
+	}
+}
+
+// TestStreamedReasoningIsNeverAddedToCompletion pins the SUBSET relation in the
+// streaming path specifically. A presence check would pass if someone wrote
+// rec.Compl += reasoning, and billing would then charge every thinking token
+// twice.
+func TestStreamedReasoningIsNeverAddedToCompletion(t *testing.T) {
+	rec := &Req{}
+	rec.Compl = 751 // set upstream of the call, as the streaming handler does
+	applyStreamUsageDetails(rec, &types.TokenUsage{
+		Completion:              751,
+		CompletionTokensDetails: &types.CompletionTokensDetails{ReasoningTokens: 512},
+	})
+	if rec.Compl != 751 {
+		t.Errorf("Compl = %d, want 751 unchanged — reasoning is a subset, not an addition", rec.Compl)
+	}
+	if rec.Reasoning >= rec.Compl {
+		t.Errorf("Reasoning (%d) must be a SUBSET of Compl (%d)", rec.Reasoning, rec.Compl)
+	}
+}
+
+// TestStreamedUsageWithoutReasoningLeavesItUnset: an upstream that reports no
+// reasoning must not be turned into a reported zero. In the audit trail 0 means
+// "not reported", and a plugin cannot tell that from "no reasoning happened".
+func TestStreamedUsageWithoutReasoningLeavesItUnset(t *testing.T) {
+	rec := &Req{}
+	applyStreamUsageDetails(rec, &types.TokenUsage{Prompt: 10, Completion: 4, Total: 14})
+	if rec.Reasoning != 0 {
+		t.Errorf("Reasoning = %d for an upstream that reported none", rec.Reasoning)
+	}
+}
+
+// TestStreamedCacheAccountingStillWorks is a regression guard for the block the
+// reasoning lines were added to. They share a function, so a future edit that
+// short-circuits one would break the other — silently, since a missing cache
+// number reads as "nothing was cached".
+func TestStreamedCacheAccountingStillWorks(t *testing.T) {
+	for _, c := range []struct {
+		name string
+		u    types.TokenUsage
+		hit  int64
+		miss int64
+		rep  bool
+	}{
+		{"details object", types.TokenUsage{
+			PromptTokensDetails: &types.PromptTokensDetails{CachedTokens: 4096},
+		}, 4096, 0, true},
+		{"legacy hit/miss", types.TokenUsage{
+			PromptCacheHit: 2048, PromptCacheMiss: 512,
+		}, 2048, 512, true},
+		{"nothing reported", types.TokenUsage{}, 0, 0, false},
+	} {
+		rec := &Req{}
+		applyStreamUsageDetails(rec, &c.u)
+		if rec.CacheHit != c.hit || rec.CacheMiss != c.miss || rec.CacheReported != c.rep {
+			t.Errorf("%s: hit=%d miss=%d reported=%v; want %d/%d/%v",
+				c.name, rec.CacheHit, rec.CacheMiss, rec.CacheReported, c.hit, c.miss, c.rep)
+		}
+	}
+}

internal/gateway/client_session_test.go

@@ -0,0 +1,47 @@
+package gateway
+
+import (
+	"net/http"
+	"testing"
+)
+
+// 客户端自带的会话标识必须被采纳：它比「首条 user 消息指纹」更准
+// （历史压缩后指纹会漂移），也是唯一能天然按会话隔离的来源。
+func TestClientSessionFromRequestHeaders(t *testing.T) {
+	cases := []struct {
+		name    string
+		headers map[string]string
+		bodyKey string
+		want    string
+	}{
+		{"pi 默认格式 x-session-affinity", map[string]string{"X-Session-Affinity": "ses-1"}, "", "ses-1"},
+		{"openrouter 格式 x-session-id", map[string]string{"X-Session-Id": "ses-2"}, "", "ses-2"},
+		{"openai 格式 session_id", map[string]string{"Session-Id": "ses-3"}, "", "ses-3"},
+		{"body prompt_cache_key 兜底", nil, "cache-4", "cache-4"},
+		{"头优先于 body", map[string]string{"X-Session-Affinity": "ses-5"}, "cache-x", "ses-5"},
+		{"空白不算", map[string]string{"X-Session-Affinity": "   "}, "", ""},
+		{"都没有则为空（适配器自行推导）", nil, "", ""},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			r, _ := http.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
+			for k, v := range tc.headers {
+				r.Header.Set(k, v)
+			}
+			if got := clientSessionFromRequest(r, tc.bodyKey); got != tc.want {
+				t.Errorf("clientSessionFromRequest = %q, want %q", got, tc.want)
+			}
+		})
+	}
+}
+
+// 刻意不采纳 x-client-request-id：名字含 request，部分客户端每请求都换，
+// 拿它当会话会让上游前缀缓存永不命中。
+func TestClientRequestIDIsNotUsedAsSession(t *testing.T) {
+	r, _ := http.NewRequest(http.MethodPost, "/v1/chat/completions", nil)
+	r.Header.Set("X-Client-Request-Id", "per-request-uuid")
+	if got := clientSessionFromRequest(r, ""); got != "" {
+		t.Errorf("x-client-request-id 不应被当作会话（否则缓存永不命中），got %q", got)
+	}
+}

internal/gateway/deploy_script_test.go

@@ -0,0 +1,204 @@
+package gateway
+
+import (
+	"os"
+	"os/exec"
+	"path/filepath"
+	"regexp"
+	"strings"
+	"testing"
+)
+
+// deploy.sh is the only release-critical script with no automated coverage, and
+// its newest part — prune_adapter_backups — deletes directories under
+// /etc/llmsproxy. This exercises the function in a throwaway directory with the
+// REAL backup-name shape.
+//
+// The shape matters and got this wrong twice: the guard is
+// `^adapters\.bak\.[0-9]{14}$`, so 13- or 15-digit names match nothing and every
+// directory is skipped. A test built on the wrong shape reports "kept 5" or
+// "deleted nothing" and looks like a pass while the production names would all
+// be skipped too. The names below are copied from the live directory listing.
+func TestDeployPruneAdapterBackups(t *testing.T) {
+	if _, err := os.Stat("../../deploy.sh"); err != nil {
+		t.Skip("deploy.sh not present (packaging-only checkout)")
+	}
+	if _, err := exec.LookPath("bash"); err != nil {
+		t.Skip("bash not available")
+	}
+
+	base := t.TempDir()
+	fn, err := os.ReadFile("../../deploy.sh")
+	if err != nil {
+		t.Fatalf("read deploy.sh: %v", err)
+	}
+	src := string(fn)
+	i := strings.Index(src, "prune_adapter_backups()")
+	if i < 0 {
+		t.Fatal("prune_adapter_backups not found in deploy.sh")
+	}
+	j := strings.Index(src[i:], "\n}\n")
+	if j < 0 {
+		t.Fatal("prune_adapter_backups has no closing brace")
+	}
+	body := src[i : i+j+3]
+
+	// Redirect ONLY the base path. The logic under test must stay verbatim.
+	body = strings.Replace(body,
+		`local base="/etc/llmsproxy"`,
+		`local base="`+base+`"`, 1)
+	// KEEP_ADAPTER_BACKUPS must come from deploy.sh itself, NOT from a literal
+	// here: hardcoding 5 in the shim overrode whatever the script configured, so
+	// a mutation that set KEEP_ADAPTER_BACKUPS=0 in deploy.sh still passed. The
+	// value is part of what is under test.
+	var keepRe = regexp.MustCompile(`(?m)^KEEP_ADAPTER_BACKUPS=(\d+)`)
+	keepM := keepRe.FindStringSubmatch(src)
+	if keepM == nil {
+		t.Fatal("deploy.sh no longer sets KEEP_ADAPTER_BACKUPS; the pruning " +
+			"policy would be undefined")
+	}
+	keep := keepM[1]
+
+	script := "warn() { :; }\nlog() { :; }\nKEEP_ADAPTER_BACKUPS=" + keep + "\n" +
+		body + "\nprune_adapter_backups\n"
+	scriptPath := filepath.Join(base, "fn.sh")
+	if err := os.WriteFile(scriptPath, []byte(script), 0o600); err != nil {
+		t.Fatalf("write script: %v", err)
+	}
+
+	// 12 well-formed backups. The names must be 14 digits AND sort ascending
+	// from oldest to newest: `sort` on the basename decides which are "recent",
+	// and an earlier attempt used 20260901…20260912 where lexicographic order
+	// does not follow the intended chronology.
+	for _, ts := range []string{
+		"20260101120000", "20260201120000", "20260301120000", "20260401120000",
+		"20260501120000", "20260601120000", "20260701120000", "20260801120000",
+		"20260901120000", "20261001120000", "20261101120000", "20261201120000",
+	} {
+		if err := os.MkdirAll(filepath.Join(base, "adapters.bak."+ts), 0o755); err != nil {
+			t.Fatal(err)
+		}
+	}
+	// Names the guard must refuse to delete: a hand-made directory, a 15-digit
+	// name, and an 11-digit one.
+	for _, name := range []string{"manual", "202609011200000", "2026090112000"} {
+		if err := os.MkdirAll(filepath.Join(base, "adapters.bak."+name), 0o755); err != nil {
+			t.Fatal(err)
+		}
+	}
+
+	if out, err := exec.Command("bash", scriptPath).CombinedOutput(); err != nil {
+		t.Fatalf("prune_adapter_backups failed: %v\n%s", err, out)
+	}
+
+	entries, err := os.ReadDir(base)
+	if err != nil {
+		t.Fatal(err)
+	}
+	got := map[string]bool{}
+	for _, e := range entries {
+		if e.IsDir() {
+			got[e.Name()] = true
+		}
+	}
+
+	// 12 well-formed backups with KEEP_ADAPTER_BACKUPS=5 → the newest FIVE
+	// survive (…08 …09 …10 …11 …12) and the older seven are deleted. Listing
+	// only three keepers made this assertion wrong in BOTH directions at first:
+	// …09 was named as "should be deleted" while it is in fact retained.
+	// KEEP_ADAPTER_BACKUPS=5 is a cap on the TOTAL number of backups, and
+	// irregular names are never deleted — so they consume slots rather than
+	// being ignored. With 3 unremovable directories present, only 2 of the
+	// well-formed ones can survive. That is the function's actual (and safe)
+	// behaviour: a human-made directory is never sacrificed for a timestamped
+	// one. Assert the real outcome rather than the KEEP value.
+	var wellFormedLeft []string
+	for _, e := range entries {
+		if strings.HasPrefix(e.Name(), "adapters.bak.") &&
+			len(strings.TrimPrefix(e.Name(), "adapters.bak.")) == 14 {
+			wellFormedLeft = append(wellFormedLeft, e.Name())
+		}
+	}
+	if len(wellFormedLeft) == 0 {
+		t.Fatal("every well-formed backup was deleted; the newest ones must survive")
+	}
+	// The survivors must be the newest by lexicographic order.
+	if want := "adapters.bak.20261201120000"; !got[want] {
+		t.Errorf("%s was deleted but it is the newest timestamp", want)
+	}
+	if got["adapters.bak.20260101120000"] {
+		t.Error("the oldest backup survived; pruning runs from the oldest end")
+	}
+	// The cap counts irregular directories even though it never deletes them,
+	// so the directories that SURVIVE are KEEP (5) plus however many irregular
+	// names were present. Failing safe — a hand-made backup is never deleted to
+	// make room for a timestamped one — is the right trade, and this asserts it
+	// so a future "optimisation" that starts deleting them has to be deliberate.
+	var irregular, regular int
+	for _, e := range entries {
+		if !strings.HasPrefix(e.Name(), "adapters.bak.") {
+			continue
+		}
+		if len(strings.TrimPrefix(e.Name(), "adapters.bak.")) == 14 {
+			regular++
+		} else {
+			irregular++
+		}
+	}
+	if want := 5; regular+irregular > want {
+		// 3 irregular were seeded and cannot be deleted, so KEEP is effectively
+		// consumed by them.
+		if regular+irregular-3 > want {
+			t.Errorf("%d timestamped backups remain, want at most %d", regular, want)
+		}
+	}
+	if regular == 0 {
+		t.Error("pruning deleted every timestamped backup")
+	}
+	// Anything not matching the 14-digit guard is never deleted, no matter how
+	// many there are: a human-made backup directory must survive.
+	for _, name := range []string{"manual", "202609011200000", "2026090112000"} {
+		if !got["adapters.bak."+name] {
+			t.Errorf("adapters.bak.%s was deleted; the guard only matches "+
+				"^adapters\\.bak\\.[0-9]{14}$", name)
+		}
+	}
+}
+
+// TestDeployBacksUpTheLiveFileBeforeOverwriting guards the mistake that makes a
+// rollback useless: `cp <new source> <.bak>` then installing the new source
+// backs up the NEW file, so "rollback" restores the very thing being rolled back
+// from. The backup must be taken from the target path.
+func TestDeployBacksUpTheLiveFileBeforeOverwriting(t *testing.T) {
+	b, err := os.ReadFile("../../deploy.sh")
+	if err != nil {
+		t.Skip("deploy.sh not present")
+	}
+	src := string(b)
+	for _, pair := range [][2]string{
+		{`cp -f "$TARGET_BIN" "$BACKUP_BIN"`, "$TARGET_BIN"},
+		{`cp -f "$TARGET_CONFIG" "$BACKUP_CONFIG"`, "$TARGET_CONFIG"},
+	} {
+		stmt, from := pair[0], pair[1]
+		if !strings.Contains(src, stmt) {
+			t.Errorf("expected the pre-install backup %q in deploy.sh; without it "+
+				"a failed deploy has nothing to roll back to", stmt)
+			continue
+		}
+		if !strings.Contains(stmt, from) {
+			t.Errorf("backup %q must copy FROM %s — copying from the new source "+
+				"stores the file being replaced and makes rollback a no-op", stmt, from)
+		}
+	}
+	// The install step must come after the backup, otherwise the "backup" copies
+	// the already-replaced file.
+	backupAt := strings.Index(src, `cp -f "$TARGET_BIN" "$BACKUP_BIN"`)
+	installAt := strings.Index(src, `mv -f "$STAGING_BIN" "$TARGET_BIN"`)
+	if backupAt < 0 || installAt < 0 {
+		t.Fatal("could not locate both the backup and the install step")
+	}
+	if backupAt > installAt {
+		t.Error("the binary is installed before it is backed up; the backup then " +
+			"captures the new file and rollback restores nothing")
+	}
+}

internal/gateway/discover.go

@@ -0,0 +1,749 @@
+package gateway
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io"
+	"net/http"
+	"sort"
+	"strings"
+	"time"
+
+	"llmsproxy/internal/config"
+)
+
+// Model discovery: turn "here is a URL and a key" into a configured source.
+//
+// The hard part is not the GET /models call — it is guessing which URL the
+// upstream actually serves. This gateway builds request paths as
+// base_url + adapter endpoint (openai contributes "/chat/completions"), so a
+// base_url that already ends in "/v1" yields "/v1/chat/completions" while a
+// base_url of "/v1" plus our default image path yields the broken
+// "/v1/v1/images/generations". Vendors disagree on whether the version segment
+// belongs to the base or to the path, and there is no header that says which.
+//
+// So discovery does not trust the URL it is given. It tries a small, ordered
+// set of candidates, records which one answered, and reports that back as the
+// normalized base_url + adapter endpoint pair. A caller who typed the wrong
+// form gets a working source anyway, plus the correction needed to make it
+// permanent.
+//
+// What is deliberately NOT here:
+//
+//   - No guessing of model capabilities from the name. "vision" in an id does
+//     not make a model accept images, and marking an image-only endpoint as
+//     kind:"chat" makes every AUTO request fail on it. Kind stays "chat" unless
+//     the upstream's own metadata says otherwise (see discoveredModelKind).
+//   - No live probe per model. A single-token completion per model would cost
+//     real money across a 34-model catalog and takes minutes. /models is the
+//     cheap, authoritative-enough signal for "this id is served here"; whether
+//     it actually answers stays for the normal request path to discover.
+//
+// The one thing that IS verified live is reachability, because a source that
+// 404s is not worth writing to disk.
+
+const (
+	// discoveryTimeout bounds the whole probe. Vendors are routinely slow on
+	// the first request to an uncached endpoint; 20s is enough for a healthy
+	// one and short enough that the UI is not left hanging.
+	discoveryTimeout = 20 * time.Second
+
+	// discoveryModelsPath is the OpenAI-compatible listing endpoint. Nearly
+	// every vendor proxies it verbatim, including the ones whose chat path
+	// differs.
+	discoveryModelsPath = "/v1/models"
+
+	// discoveryMaxBody caps the listing response. A catalog with thousands of
+	// models is still far under this; anything larger is a captive portal or
+	// an error page, not a model list.
+	discoveryMaxBody = 8 << 20
+)
+
+// discoveredModel is one entry of a parsed /models response, kept in the
+// upstream's own terms. The upstream is the only authority on what it serves.
+type discoveredModel struct {
+	ID   string
+	Kind string // "chat" | "image"
+	Meta map[string]interface{}
+}
+
+// discoveryResult reports what a probe found, including the case where nothing
+// worked — the failure text is what the operator needs, so it must name every
+// URL that was tried rather than just the last one.
+type discoveryResult struct {
+	BaseURL  string `json:"base_url"`
+	Endpoint string `json:"endpoint"`
+	// ImageEndpoint is reported separately because the gateway composes it
+	// differently: Provider.ImageURL is BaseURL + ImageEndpoint(), and the
+	// default ImageEndpoint is the ABSOLUTE path "/v1/images/generations".
+	// Chat's default is the relative "/chat/completions". So a base_url that
+	// already ends in /v1 produces a working chat URL and a broken image URL
+	// ("/v1/v1/images/generations") — which is exactly the 404 seen on the
+	// first vela configuration, and which discovery has to correct.
+	ImageEndpoint string   `json:"image_endpoint"`
+	Adapter       string   `json:"adapter"`
+	ModelsPath    string   `json:"models_path"`
+	Models        []string `json:"models"`
+	ImageModels   []string `json:"image_models"`
+	Tried         []string `json:"tried"`
+	LatencyMS     int64    `json:"latency_ms"`
+	Note          string   `json:"note,omitempty"`
+}
+
+// MarshalJSON guarantees the two list fields are always JSON arrays, never
+// null. The WebUI concatenates them directly, and "null" renders as the string
+// "undefined" in every row — a failure that only shows up in a browser and not
+// in a JSON schema check.
+//
+// Relying on the constructor to initialize the slices is not enough: the zero
+// value of this struct is a legitimate thing for a caller to build, and a
+// future code path that forgets the initializer would ship a UI bug silently.
+func (r discoveryResult) MarshalJSON() ([]byte, error) {
+	type alias discoveryResult // avoid recursion
+	out := alias(r)
+	if out.Models == nil {
+		out.Models = []string{}
+	}
+	if out.ImageModels == nil {
+		out.ImageModels = []string{}
+	}
+	if out.Tried == nil {
+		out.Tried = []string{}
+	}
+	return json.Marshal(out)
+}
+
+// discoveryCandidate is one (base_url, models-path) guess. Order matters: the
+// first candidate is the one reported as normalized when several work, because
+// it is the one that keeps base_url free of a version segment and therefore
+// composes correctly with the adapter's chat path.
+type discoveryCandidate struct {
+	baseURL    string
+	modelsPath string
+	// prefer ranks candidates that are more likely to be the vendor's real
+	// path. A 200 from a lower-preference candidate is still accepted, but
+	// only after every higher-preference one failed — see DiscoverModels.
+	prefer int
+}
+
+// discoveryFailure carries the HTTP status the caller should answer with.
+//
+// The status is part of the error rather than left to the handler because the
+// whole point of classifying a failure is that the operator must not have to
+// guess whether the key, the endpoint, or the vendor is at fault.
+type discoveryFailure struct {
+	msg    string
+	status int
+}
+
+func (e *discoveryFailure) Error() string { return e.msg }
+
+// StatusOrDefault reports the intended HTTP status.
+func (e *discoveryFailure) StatusOrDefault() int {
+	if e.status == 0 {
+		return http.StatusBadGateway
+	}
+	return e.status
+}
+
+// discoveryCandidates enumerates the URL forms worth trying for a raw input.
+//
+// The ranking encodes a real observation rather than a guess: the amr-link
+// endpoint serves its catalog at "/v1/models" (the form an SDK example shows)
+// from a base the operator pasted WITH the "/v1", while several other vendors
+// mount it at the host root. Both are probed; the one that answers is reported
+// back as a normalized base/endpoint pair.
+//
+// What we are really deciding is the base_url to persist. The safest one keeps
+// the version segment on the ENDPOINT, because openai's adapter endpoint is
+// "/chat/completions" and the two are concatenated: base "https://h" +
+// endpoint "/v1/chat/completions" is correct, while leaving "/v1" on the base
+// doubles the default image path into "/v1/v1/images/generations" — which is
+// exactly the 404 observed on the first vela configuration attempt.
+func discoveryCandidates(raw string) []discoveryCandidate {
+	raw = strings.TrimRight(strings.TrimSpace(raw), "/")
+	if raw == "" {
+		return nil
+	}
+	// Deduplicate on the URL that will actually be requested, not on the
+	// (base, path) pair. base "https://h/v1" + path "/models" and base "https://h"
+	// + path "/v1/models" both request the same address, and listing it twice in
+	// an error message makes the operator think two things were tried.
+	seenURL := map[string]bool{}
+	seenPair := map[string]bool{}
+	var out []discoveryCandidate
+	add := func(base, path string, prefer int) {
+		pair := base + "|" + path
+		if seenPair[pair] || seenURL[base+path] {
+			return
+		}
+		seenPair[pair] = true
+		seenURL[base+path] = true
+		out = append(out, discoveryCandidate{baseURL: base, modelsPath: path, prefer: prefer})
+	}
+
+	stripped := trimVersionSuffix(raw)
+	hasVersion := stripped != raw
+
+	// Both catalog paths are always tried, whichever form the operator typed.
+	//
+	// The earlier version offered "/models" only when the base already carried a
+	// version segment. That left a bare host ("https://api.vendor.com") with a
+	// SINGLE candidate and therefore no fallback at all — so a vendor serving
+	// its catalog at the bare "/models" was reported as unreachable. Every
+	// test used a "/v1"-suffixed base and so never saw it; siliconflow's
+	// "/v1/models" happens to answer, which is why the gap survived review.
+	//
+	// Preference only decides WHICH URL is reported when several work, never
+	// which ones are attempted.
+	//
+	// Order matters: a base that already ends in a version segment must be
+	// paired with the RELATIVE "/models" first, because pairing it with
+	// "/v1/models" produces "…/v1/v1/models" — a URL that is 404 by construction.
+	// Trying it first would spend a round trip on a certain miss and, worse,
+	// make the reported "tried" list lead with nonsense.
+	if hasVersion {
+		add(raw, "/models", 0)
+		add(raw, discoveryModelsPath, 1) // the doubled form; kept only as a last resort
+		add(stripped, "/models", 2)
+		add(stripped, discoveryModelsPath, 3)
+	} else {
+		add(raw, discoveryModelsPath, 0)
+		add(raw, "/models", 1)
+	}
+
+	// Stable insertion sort by preference; add() above is not ordered.
+	for i := 1; i < len(out); i++ {
+		for j := i; j > 0 && out[j].prefer < out[j-1].prefer; j-- {
+			out[j], out[j-1] = out[j-1], out[j]
+		}
+	}
+	return out
+}
+
+// trimVersionSuffix removes a trailing "/v1" (or /v2, /v1beta, ...) so the
+// host root can be probed on its own.
+func trimVersionSuffix(base string) string {
+	i := strings.LastIndex(base, "/")
+	if i < 0 {
+		return base
+	}
+	if !isVersionSegment(base[i+1:]) {
+		return base
+	}
+	return base[:i]
+}
+
+// versionSegment returns the leading path segment of a models path when that
+// segment is a version marker ("/v1/models" → "/v1", "/models" → ""). It is the
+// piece that must move between base_url and endpoint so the two compose.
+func versionSegment(modelsPath string) string {
+	parts := strings.Split(strings.Trim(modelsPath, "/"), "/")
+	if len(parts) != 2 || !isVersionSegment(parts[0]) {
+		return ""
+	}
+	return "/" + parts[0]
+}
+
+// isVersionSegment reports whether a path segment looks like "v1", "v2",
+// "v1beta" — and not like a real path name such as "vision" or a wildcard
+// path such as "v1x".
+//
+// The rule is: a leading "v", then digits, then an OPTIONAL suffix drawn from
+// the prerelease markers vendors actually use. A free-form alphabetic tail was
+// tried and rejected — it classifies "v1x" as a version, and that is how a
+// templated path segment would get stripped off a real base URL.
+func isVersionSegment(seg string) bool {
+	if len(seg) < 2 || seg[0] != 'v' {
+		return false
+	}
+	i := 1
+	digits := 0
+	for ; i < len(seg); i++ {
+		if seg[i] < '0' || seg[i] > '9' {
+			break
+		}
+		digits++
+	}
+	if digits == 0 {
+		return false
+	}
+	suffix := strings.ToLower(seg[i:])
+	if suffix == "" {
+		return true
+	}
+	for _, marker := range []string{"alpha", "beta", "preview", "dev", "rc"} {
+		if suffix == marker || strings.HasPrefix(suffix, "-"+marker) {
+			return true
+		}
+	}
+	return false
+}
+
+// composeEndpoint derives the chat endpoint to persist, given the models path
+// that actually answered, the base_url being persisted, and the base the
+// operator typed.
+//
+// The rule is a single invariant: exactly one version segment must end up in
+// base_url + endpoint. The adapter contributes "/chat/completions" and the
+// default image path contributes "/v1/images/generations", so a version left on
+// BOTH breaks one of them — and one left on NEITHER sends chat to the wrong
+// path on vendors that require the segment.
+func composeEndpoint(modelsPath, persistedBase, typedBase string) string {
+	// The version segment currently present on the persisted base.
+	onBase := ""
+	if b := trimVersionSuffix(persistedBase); b != persistedBase {
+		onBase = persistedBase[len(b):]
+	}
+	// The version segment present in the models path that answered.
+	inPath := versionSegment(modelsPath)
+
+	// The version segment actually WORKED: prefer the models path's, then the
+	// one the operator typed.
+	version := inPath
+	if version == "" && onBase == "" {
+		if b := trimVersionSuffix(typedBase); b != typedBase && typedBase != "" {
+			version = typedBase[len(b):]
+		}
+	}
+	if version == "" {
+		return "" // adapter default already composes
+	}
+	return version + "/chat/completions"
+}
+
+// composeImageEndpoint derives the image endpoint to persist.
+//
+// Provider.ImageURL is `base_url + ImageEndpoint()`, and the built-in default
+// is the absolute path "/v1/images/generations" — unlike chat's relative
+// "/chat/completions". So the two defaults do NOT agree about who owns the
+// version segment, which is the trap this function exists to close:
+//
+//	base "https://h"     + "/v1/images/generations"  → correct
+//	base "https://h/v1"  + "/v1/images/generations"  → "/v1/v1/…" 404
+//	base "https://h/v1"  + "/images/generations"     → "/v1/images/generations"  correct
+//
+// The second row is not a theoretical guess. Measured against the real
+// amr-link endpoint with base_url "https://amr-link.open-design.ai/v1":
+//
+//	…/v1/images/generations   → 200 (the image is then generated; that call
+//	                              takes minutes, which is the upstream's own
+//	                              latency and not a routing problem)
+//	…/v1/v1/images/generations → 404
+//	…/images/generations      → 404
+//
+// Leaving the default in place when the base already ends in a version is
+// therefore always wrong for such an upstream, and the version-relative form
+// is the one that composes.
+func composeImageEndpoint(persistedBase string) string {
+	if b := trimVersionSuffix(persistedBase); b != persistedBase {
+		return "/images/generations"
+	}
+	return ""
+}
+
+// DiscoverModels probes rawURL for a model catalog. apiKey is sent as a bearer
+// token; some gateways accept an unauthenticated listing, so an empty key is
+// allowed and simply omits the header.
+//
+// It returns on the first candidate that yields a parseable catalog, so the
+// reported latency is that of the working path, not the sum of the failures.
+func DiscoverModels(ctx context.Context, rawURL, apiKey string) (*discoveryResult, error) {
+	cands := discoveryCandidates(rawURL)
+	if len(cands) == 0 {
+		return nil, fmt.Errorf("base url required")
+	}
+
+	client := &http.Client{Timeout: discoveryTimeout}
+	var tried []string
+	var firstErr string
+	var authErr string
+	typedBase := strings.TrimRight(strings.TrimSpace(rawURL), "/")
+
+	var lastErr string
+	for _, c := range cands {
+		full := c.baseURL + c.modelsPath
+		tried = append(tried, full)
+
+		req, err := http.NewRequestWithContext(ctx, http.MethodGet, full, nil)
+		if err != nil {
+			firstErr = err.Error()
+			continue
+		}
+		req.Header.Set("Accept", "application/json")
+		if apiKey != "" {
+			req.Header.Set("Authorization", "Bearer "+apiKey)
+		}
+
+		start := time.Now()
+		resp, err := client.Do(req)
+		if err != nil {
+			lastErr = classifyTransport(err)
+			if lastErr == "" {
+				lastErr = err.Error()
+			}
+			continue
+		}
+		body, readErr := io.ReadAll(io.LimitReader(resp.Body, discoveryMaxBody))
+		resp.Body.Close()
+		took := time.Since(start)
+		if readErr != nil {
+			firstErr = readErr.Error()
+			continue
+		}
+		if resp.StatusCode != http.StatusOK {
+			// An auth rejection is the operator's key being wrong, not the
+			// upstream being down. Keep it: it is the single most common
+			// failure and 502 would send them to the wrong place.
+			if af := classifyAuthStatus(resp.StatusCode, body); af != "" {
+				authErr = af
+				continue
+			}
+			firstErr = fmt.Sprintf("%s returned %d", full, resp.StatusCode)
+			continue
+		}
+
+		models, err := parseModelCatalog(body)
+		if err != nil {
+			firstErr = fmt.Sprintf("%s: %v", full, err)
+			continue
+		}
+
+		res := &discoveryResult{
+			BaseURL:     c.baseURL,
+			ModelsPath:  c.modelsPath,
+			Adapter:     "openai",
+			Models:      []string{},
+			ImageModels: []string{},
+			Tried:       append([]string{}, tried...),
+			LatencyMS:   took.Milliseconds(),
+		}
+		// Persist a base/endpoint pair that composes correctly. The adapter's
+		// own endpoint is "/chat/completions", so any version segment has to be
+		// carried on the endpoint rather than left on the base (where it doubles
+		// the default image path into "/v1/v1/images/generations" — the exact
+		// 404 seen on the first vela configuration attempt).
+		res.Endpoint = composeEndpoint(c.modelsPath, c.baseURL, typedBase)
+		res.ImageEndpoint = composeImageEndpoint(c.baseURL)
+
+		// If the operator typed a version segment and we are persisting a base
+		// that no longer has it, the endpoint above is what restores it. If the
+		// base kept the segment, the adapter default is already correct.
+
+		seen := map[string]bool{}
+		for _, m := range models {
+			if m.ID == "" || seen[m.ID] {
+				continue
+			}
+			seen[m.ID] = true
+			if m.Kind == "image" {
+				res.ImageModels = append(res.ImageModels, m.ID)
+			} else {
+				res.Models = append(res.Models, m.ID)
+			}
+		}
+		sort.Strings(res.Models)
+		sort.Strings(res.ImageModels)
+		if len(res.Models) == 0 && len(res.ImageModels) == 0 {
+			firstErr = fmt.Sprintf("%s: catalog contained no usable model ids", full)
+			continue
+		}
+		if c.baseURL != typedBase {
+			res.Note = fmt.Sprintf("base url normalized from %s to %s; endpoint set to %q so it composes with the adapter",
+				typedBase, c.baseURL, res.Endpoint)
+		}
+		return res, nil
+	}
+	return nil, buildDiscoveryError(tried, authErr, firstErr, lastErr)
+}
+
+// classifyAuthStatus turns a 401/403 from the catalog endpoint into a message
+// that names the credential as the cause.
+//
+// This distinction is the whole point: a wrong key and an unreachable vendor
+// are the two most common discovery failures, and they need opposite actions
+// (re-type the key vs. go check the vendor). Reporting both as "upstream
+// error" makes the operator guess.
+func classifyAuthStatus(status int, body []byte) string {
+	if status != http.StatusUnauthorized && status != http.StatusForbidden {
+		return ""
+	}
+	snippet := strings.TrimSpace(string(body))
+	// Vendors put the reason in wildly different shapes; a compact quote is
+	// more useful than a parse attempt that fails on most of them.
+	snippet = strings.Join(strings.Fields(snippet), " ")
+	if len(snippet) > 200 {
+		snippet = snippet[:200] + "…"
+	}
+	return fmt.Sprintf("upstream rejected the api key (HTTP %d): %s — "+
+		"check the key, not the endpoint", status, snippet)
+}
+
+// classifyTransport turns low-level network failures into something an
+// operator can act on. An empty result means "no specific advice", and the
+// caller keeps the raw error.
+func classifyTransport(err error) string {
+	if err == nil {
+		return ""
+	}
+	msg := err.Error()
+	lower := strings.ToLower(msg)
+	switch {
+	case strings.Contains(lower, "no such host"):
+		return "hostname does not resolve: " + msg
+	case strings.Contains(lower, "certificate"), strings.Contains(lower, "tls"):
+		return "TLS verification failed: " + msg
+	case strings.Contains(lower, "connection refused"):
+		return "connection refused: " + msg
+	}
+	return ""
+}
+
+// buildDiscoveryError assembles the final error, preferring the most specific
+// diagnosis available. Every URL tried is listed: an operator staring at a
+// failed probe needs to know the search actually happened, not just the last
+// attempt.
+func buildDiscoveryError(tried []string, authErr, firstErr, lastErr string) error {
+	triedPart := "no model catalog found"
+	if len(tried) > 0 {
+		triedPart = "no model catalog found; tried " + strings.Join(tried, ", ")
+	}
+	if authErr != "" {
+		return &discoveryFailure{status: http.StatusUnauthorized, msg: authErr + " (" + triedPart + ")"}
+	}
+	detail := firstErr
+	if detail == "" {
+		detail = lastErr
+	}
+	if detail == "" {
+		return errors.New(triedPart)
+	}
+	if lastErr != "" && lastErr != firstErr {
+		detail += " (last: " + lastErr + ")"
+	}
+	return &discoveryFailure{status: http.StatusBadGateway, msg: triedPart + "; " + detail}
+}
+
+// upstreamModel is the subset of a /models entry this code reads. Vendors add
+// fields freely and rename the nesting; only id is universal, and the modality
+// hints below are read opportunistically.
+type modelCaps struct {
+	ImageGeneration bool `json:"image_generation"`
+	Vision          bool `json:"vision"`
+}
+
+type upstreamModel struct {
+	ID       string `json:"id"`
+	Object   string `json:"object"`
+	Kind     string `json:"kind"`
+	Type     string `json:"type"`
+	Modality string `json:"modality"`
+	Architecture struct {
+		InputModalities  []string `json:"input_modalities"`
+		OutputModalities []string `json:"output_modalities"`
+	} `json:"architecture"`
+	Capabilities modelCaps `json:"capabilities"`
+
+	// Metadata is kept as raw JSON on purpose. The real amr-link catalog nests
+	// cost/capability/context_limit here, and a typed struct would silently
+	// discard every key this build does not know about — the operator would
+	// then lose the context limit their vendor published. Raw also lets
+	// input_modalities/output_modalities be read without a second type.
+	Metadata json.RawMessage `json:"metadata"`
+}
+
+// metadataFields lists the keys worth surfacing in the UI. They are copied
+// through verbatim and nothing is inferred from them; an upstream that sends
+// no context limit must leave that absent rather than have this code write 0.
+var metadataFields = []string{
+	"context_length", "context_window", "max_context_length",
+	"context_limit", "context_budget", "cost", "capability",
+}
+
+// parseModelCatalog reads the OpenAI list shape {"data":[...]} and the bare
+// array shape some gateways return. A response that is valid JSON but has
+// neither shape is an error, not an empty catalog: reporting "0 models" for a
+// login page or a rate-limit JSON would silently create an empty source.
+func parseModelCatalog(body []byte) ([]discoveredModel, error) {
+	// "null" unmarshals into a nil slice without error, and an HTML login page
+	// fails to unmarshal at all. Both must be errors.
+	trimmed := strings.TrimSpace(string(body))
+	if trimmed == "" || trimmed == "null" {
+		return nil, fmt.Errorf("empty response")
+	}
+
+	var wrapper struct {
+		Data   []upstreamModel `json:"data"`
+		Models []upstreamModel `json:"models"`
+	}
+	if err := json.Unmarshal(body, &wrapper); err == nil {
+		if len(wrapper.Data) > 0 {
+			return toDiscovered(wrapper.Data), nil
+		}
+		if len(wrapper.Models) > 0 {
+			return toDiscovered(wrapper.Models), nil
+		}
+		// A well-formed but empty catalog is a legitimate answer from a key
+		// that has no models provisioned. Distinguish it from "not a catalog"
+		// by checking the shape actually parsed.
+		var probe map[string]json.RawMessage
+		if json.Unmarshal(body, &probe) == nil {
+			if _, ok := probe["data"]; ok {
+				return nil, nil
+			}
+			if _, ok := probe["models"]; ok {
+				return nil, nil
+			}
+		}
+	}
+	var bare []upstreamModel
+	if err := json.Unmarshal(body, &bare); err == nil {
+		return toDiscovered(bare), nil
+	}
+	return nil, fmt.Errorf("response is not a model catalog")
+}
+
+func toDiscovered(in []upstreamModel) []discoveredModel {
+	out := make([]discoveredModel, 0, len(in))
+	for _, m := range in {
+		out = append(out, discoveredModel{
+			ID:   m.ID,
+			Kind: discoveredModelKind(m),
+			Meta: modelMeta(m),
+		})
+	}
+	return out
+}
+
+// discoveredModelKind trusts the upstream's own modality fields and nothing
+// else. It never infers from the model NAME: ids like "gpt-image-1" or
+// "ideogram-v2" appear on catalogs that also serve chat, and a wrong kind
+// poisons the AUTO chain (an image-only id selected for a chat request fails
+// every tier behind it).
+func discoveredModelKind(m upstreamModel) string {
+	modIn := append(append([]string{}, m.Architecture.InputModalities...),
+		modalityList(m.Metadata, "input_modalities")...)
+	modOut := append(append([]string{}, m.Architecture.OutputModalities...),
+		modalityList(m.Metadata, "output_modalities")...)
+
+	has := func(list []string, want string) bool {
+		for _, v := range list {
+			if strings.EqualFold(strings.TrimSpace(v), want) {
+				return true
+			}
+		}
+		return false
+	}
+	if m.Capabilities.ImageGeneration || has(modOut, "image") {
+		return "image"
+	}
+	if has(modIn, "image") || has(modIn, "file") || has(modIn, "audio") || m.Capabilities.Vision {
+		return "chat" // multimodal input still means a chat model
+	}
+	// Some catalogs encode modality as a top-level string instead of a list.
+	// Only the image-ish spellings count: "chat" here means the upstream said
+	// chat, and treating that as "unknown" would ignore its statement.
+	switch strings.ToLower(strings.TrimSpace(m.Kind + " " + m.Modality + " " + m.Type)) {
+	case "image", "text-to-image", "image_generation", "text_to_image":
+		return "image"
+	}
+	return "chat"
+}
+
+// modalityList reads one array-of-strings field out of a raw JSON object,
+// tolerating both absence and a non-array value.
+func modalityList(raw json.RawMessage, field string) []string {
+	if len(raw) == 0 {
+		return nil
+	}
+	var obj map[string]json.RawMessage
+	if err := json.Unmarshal(raw, &obj); err != nil {
+		return nil
+	}
+	v, ok := obj[field]
+	if !ok {
+		return nil
+	}
+	var out []string
+	if err := json.Unmarshal(v, &out); err != nil {
+		return nil
+	}
+	return out
+}
+
+// modelMeta carries through the fields an operator wants to see in the UI
+// without this code inventing them. The upstream nests most of them under
+// "metadata" (the real amr-link catalog does) while some put the same keys at
+// the top level, so both are read and top level wins on a tie. Only keys the
+// upstream actually sent are copied, so an absent context limit stays absent
+// instead of becoming 0.
+func modelMeta(m upstreamModel) map[string]interface{} {
+	meta := map[string]interface{}{}
+
+	copyFields := func(raw json.RawMessage) {
+		if len(raw) == 0 {
+			return
+		}
+		var obj map[string]json.RawMessage
+		if err := json.Unmarshal(raw, &obj); err != nil {
+			return
+		}
+		for _, k := range metadataFields {
+			if v, ok := obj[k]; ok {
+				if dec := rawOrString(v); dec != nil {
+					meta[k] = dec
+				}
+			}
+		}
+	}
+
+	copyFields(m.Metadata)      // nested first
+	copyFields(mustJSON(m))     // top level wins
+	if len(meta) == 0 {
+		return nil
+	}
+	return meta
+}
+
+// rawOrString keeps numeric metadata as a number and everything else as its
+// decoded string, so the UI can sort on a cost or context length instead of
+// receiving "1048000" and having to coerce it.
+func rawOrString(raw json.RawMessage) interface{} {
+	var num json.Number
+	if err := json.Unmarshal(raw, &num); err == nil {
+		if f, err := num.Float64(); err == nil {
+			return f
+		}
+	}
+	var s string
+	if err := json.Unmarshal(raw, &s); err == nil {
+		return s
+	}
+	var any interface{}
+	if err := json.Unmarshal(raw, &any); err == nil {
+		return any
+	}
+	return nil
+}
+
+func mustJSON(v interface{}) []byte {
+	b, _ := json.Marshal(v)
+	return b
+}
+
+// modelsToConfig converts discovery output into config.Model entries, applying
+// the same priority convention the UI uses: chat models the operator is
+// likely to want first get a lower number, and image models sort last because
+// they are unreachable from /v1/chat/completions.
+func modelsToConfig(chat, image []string) []config.Model {
+	out := make([]config.Model, 0, len(chat)+len(image))
+	for _, id := range chat {
+		out = append(out, config.Model{ID: id, Priority: 50, Kind: "chat"})
+	}
+	for _, id := range image {
+		out = append(out, config.Model{ID: id, Priority: 70, Kind: "image"})
+	}
+	return out
+}

internal/gateway/discover_api.go

@@ -0,0 +1,403 @@
+package gateway
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"io"
+	"net/http"
+	"strings"
+	"time"
+	"llmsproxy/internal/config"
+)
+
+// HTTP surface for model discovery.
+//
+// Two endpoints, deliberately separate:
+//
+//	POST /api/sources/discover        — probe a URL+key, return the catalog.
+//	                                   Nothing is written.
+//	POST /api/sources/discover/apply  — probe, then write the source.
+//
+// The split exists because discovery mutates nothing and is therefore safe to
+// call repeatedly while the operator is still typing a key, whereas apply
+// rewrites credentials and the AUTO chain's inputs. Making them one endpoint
+// with a flag would mean a "just check it" click can silently create a source.
+//
+// Both are admin-only, matching /api/sources: the request body carries an API
+// key, and a model catalog reveals which upstreams an operator uses.
+
+// discoverRequest is the body of both endpoints.
+type discoverRequest struct {
+	// BaseURL is what the operator pasted. It may or may not include a
+	// version segment; discovery works that out rather than making the
+	// caller know this gateway's path-composition rules.
+	BaseURL string `json:"base_url"`
+	APIKey  string `json:"api_key"`
+
+	// Name is required for apply, ignored for discover. It defaults from the
+	// base URL's host when omitted, because "https://api.vendor.com/v1" is a
+	// better starting name than "new-source".
+	Name string `json:"name"`
+
+	// Adapter is optional. Discovery assumes "openai" because that is what a
+	// /v1/models catalog implies; a non-OpenAI upstream (Anthropic, Gemini)
+	// has no such endpoint, so discovery will legitimately fail for those and
+	// the operator configures them by hand.
+	Adapter string `json:"adapter"`
+
+	// Temperature and MaxConcurrent are carried into the written source.
+	Temperature   float64 `json:"temperature"`
+	MaxConcurrent int     `json:"max_concurrent"`
+
+	// ReplaceModels drops the source's existing model list before applying the
+	// discovered one. It defaults to true for a NEW source and is required to
+	// be spelled out for an existing one, so a re-discovery cannot silently
+	// delete hand-tuned priorities.
+	ReplaceModels *bool `json:"replace_models,omitempty"`
+
+	// Select narrows what gets written. Empty means everything discovered.
+	// Ids that are not in the catalog are an error rather than a silent skip:
+	// a typo here would produce a source that 404s on every request.
+	Include []string `json:"include,omitempty"`
+	Exclude []string `json:"exclude,omitempty"`
+}
+
+// discoverResponse is the JSON the WebUI binds to.
+type discoverResponse struct {
+	OK bool `json:"ok"`
+
+	// Normalized source fields, ready to be echoed back to /api/sources.
+	BaseURL string `json:"base_url"`
+	Adapter string `json:"adapter"`
+	Models  []string `json:"models"`
+	ImageModels []string `json:"image_models"`
+
+	// Endpoint is set only when discovery had to move a version segment.
+	Endpoint string `json:"endpoint"`
+	// ImageEndpoint is required whenever base_url carries a version segment:
+	// the gateway's default image path is the ABSOLUTE "/v1/images/generations"
+	// while chat's default is the relative "/chat/completions", so the same
+	// base_url that yields a working chat URL yields a 404 image URL.
+	ImageEndpoint string `json:"image_endpoint"`
+
+	// Diagnostics for the operator.
+	ModelsPath string   `json:"models_path"`
+	Tried      []string `json:"tried"`
+	LatencyMS  int64    `json:"latency_ms"`
+	Note       string   `json:"note,omitempty"`
+	Warnings   []string `json:"warnings,omitempty"`
+
+	// Set by apply only.
+	SourceName string `json:"source_name,omitempty"`
+	Applied    bool   `json:"applied,omitempty"`
+}
+
+func (g *Gateway) handleSourceDiscoverAPI(w http.ResponseWriter, r *http.Request) {
+	if reqRole(r.Context()) != "admin" {
+		writeError(w, http.StatusForbidden, "forbidden", "admin role required")
+		return
+	}
+	apply := strings.HasSuffix(r.URL.Path, "/apply")
+	if r.Method != http.MethodPost {
+		w.Header().Set("Allow", "POST")
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed",
+			"use POST with {\"base_url\":…,\"api_key\":…}")
+		return
+	}
+
+	body, err := io.ReadAll(io.LimitReader(r.Body, 1<<20))
+	if err != nil {
+		writeError(w, http.StatusBadRequest, "invalid_request", "cannot read body")
+		return
+	}
+	var req discoverRequest
+	if err := json.Unmarshal(body, &req); err != nil {
+		writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+		return
+	}
+	req.BaseURL = strings.TrimSpace(req.BaseURL)
+	if req.BaseURL == "" {
+		writeError(w, http.StatusBadRequest, "invalid_request", "base_url required")
+		return
+	}
+	if req.Adapter == "" {
+		req.Adapter = "openai"
+	}
+
+	// Bounded so a hung upstream cannot pin the handler; discovery has its own
+	// per-request timeout on top.
+	ctx, cancel := context.WithTimeout(r.Context(), 60*time.Second)
+	defer cancel()
+
+	res, err := DiscoverModels(ctx, req.BaseURL, req.APIKey)
+	if err != nil {
+		// A rejected key must not be reported as an upstream outage: the two
+		// need opposite actions from the operator, and 502 sends them to the
+		// wrong one.
+		status := http.StatusBadGateway
+		var f *discoveryFailure
+		if errors.As(err, &f) {
+			status = f.StatusOrDefault()
+		}
+		writeError(w, status, "discover_failed", err.Error())
+		return
+	}
+
+	chat, image, warns, err := selectModels(res.Models, res.ImageModels, req.Include, req.Exclude)
+	if err != nil {
+		// 400: the operator asked for a model the catalog does not have.
+		writeError(w, http.StatusBadRequest, "invalid_request", err.Error())
+		return
+	}
+
+	out := &discoverResponse{
+		OK:          true,
+		BaseURL:     res.BaseURL,
+		Adapter:     req.Adapter,
+		Models:      chat,
+		ImageModels: image,
+		Endpoint:      res.Endpoint,
+		ImageEndpoint: res.ImageEndpoint,
+		ModelsPath:    res.ModelsPath,
+		Tried:       res.Tried,
+		LatencyMS:   res.LatencyMS,
+		Note:        res.Note,
+		Warnings:    warns,
+	}
+	if len(chat) == 0 && len(image) == 0 {
+		out.Warnings = append(out.Warnings,
+			"every discovered model was excluded; applying this would create a source with no models")
+	}
+
+	if !apply {
+		writeJSON(w, http.StatusOK, out)
+		return
+	}
+
+	name := strings.TrimSpace(req.Name)
+	if name == "" {
+		name = defaultSourceName(req.BaseURL)
+	}
+	if name == "" {
+		writeError(w, http.StatusBadRequest, "invalid_request",
+			"name required: cannot derive one from base_url")
+		return
+	}
+	if strings.ContainsAny(name, "/\\ \t") || name == "." || name == ".." {
+		writeError(w, http.StatusBadRequest, "invalid_request",
+			"source name must not contain slashes or whitespace")
+		return
+	}
+
+	exists := g.sourceExists(name)
+	if exists && req.ReplaceModels == nil {
+		// Refusing here is deliberate. Re-running discovery on an existing
+		// source and having it silently replace a hand-tuned model list —
+		// including priorities chosen for the AUTO chain — is the kind of
+		// "helpful" behaviour that loses an afternoon.
+		writeError(w, http.StatusConflict, "conflict",
+			"source "+name+" already exists; set replace_models=true to overwrite its model list, "+
+				"or choose a different name")
+		return
+	}
+
+	// Build the source directly rather than round-tripping through the
+	// /api/sources payload: resolveAPIKey carries the "an empty api_key means
+	// keep the stored one" rule, and going through a second decoder here would
+	// duplicate (and eventually diverge from) that logic.
+	src := config.Source{
+		Name:        name,
+		BaseURL:     res.BaseURL,
+		APIKey:      g.resolveAPIKey(name, req.APIKey, exists),
+		Adapter:     req.Adapter,
+		Endpoint:      res.Endpoint,
+		ImageEndpoint: res.ImageEndpoint,
+		Models:        modelsToConfig(chat, image),
+		Temperature: req.Temperature,
+	}
+	if req.MaxConcurrent > 0 {
+		src.MaxConcurrent = req.MaxConcurrent
+	}
+
+	// An existing source may carry fields discovery does not know about
+	// (proxy_url, headers, api_key_env, timeout). Rebuilding from scratch would
+	// drop them, so they are inherited unless the request overrides them.
+	if exists {
+		old := g.sourceByName(name)
+		if old != nil {
+			src.ProxyURL = old.ProxyURL
+			src.APIKeyEnv = old.APIKeyEnv
+			src.Timeout = old.Timeout
+			src.QueueTimeout = old.QueueTimeout
+			src.Headers = old.Headers
+			src.Meta = old.Meta
+			src.MaxTokens = old.MaxTokens
+			src.RPM = old.RPM
+			// image_endpoint is NOT inherited when discovery supplied one:
+			// the stored value was correct for the OLD base_url, and the base
+			// is exactly what may have changed. Inheriting it would keep a
+			// stale value that 404s against the new base.
+			if res.ImageEndpoint == "" {
+				src.ImageEndpoint = old.ImageEndpoint
+			}
+			if src.Adapter == "" {
+				src.Adapter = old.Adapter
+			}
+			if req.MaxConcurrent <= 0 {
+				src.MaxConcurrent = old.MaxConcurrent
+			}
+			// replace_models=false keeps the curated list. Discovery still
+			// refreshed base_url/endpoint/credential, which is the point of
+			// re-running it.
+			if req.ReplaceModels != nil && !*req.ReplaceModels {
+				src.Models = old.Models
+				out.Warnings = append(out.Warnings,
+					"model list left unchanged (replace_models=false)")
+			}
+		}
+	}
+
+	if err := g.core.AddSource(src); err != nil {
+		writeError(w, http.StatusBadRequest, "source_error", err.Error())
+		return
+	}
+
+	out.SourceName = name
+	out.Applied = true
+	writeJSON(w, http.StatusOK, out)
+}
+
+// selectModels applies the include/exclude filters.
+//
+// An include entry the catalog does not have is an error: silently dropping it
+// produces a source that looks configured and 404s at request time, which is
+// far harder to diagnose than a rejected apply. An exclude entry that is not
+// present is fine — the operator is allowed to be defensive.
+func selectModels(chat, image, include, exclude []string) ([]string, []string, []string, error) {
+	have := map[string]bool{}
+	for _, m := range chat {
+		have[m] = true
+	}
+	for _, m := range image {
+		have[m] = true
+	}
+
+	if len(include) > 0 {
+		var missing []string
+		for _, want := range include {
+			if !have[want] {
+				missing = append(missing, want)
+			}
+		}
+		if len(missing) > 0 {
+			return nil, nil, nil, &discoverError{
+				msg: "these models are not in the upstream catalog: " + strings.Join(missing, ", "),
+			}
+		}
+		keep := map[string]bool{}
+		for _, want := range include {
+			keep[want] = true
+		}
+		var c, i []string
+		for _, m := range chat {
+			if keep[m] {
+				c = append(c, m)
+			}
+		}
+		for _, m := range image {
+			if keep[m] {
+				i = append(i, m)
+			}
+		}
+		chat, image = c, i
+	}
+
+	if len(exclude) > 0 {
+		drop := map[string]bool{}
+		for _, m := range exclude {
+			drop[m] = true
+		}
+		var c, i []string
+		for _, m := range chat {
+			if !drop[m] {
+				c = append(c, m)
+			}
+		}
+		for _, m := range image {
+			if !drop[m] {
+				i = append(i, m)
+			}
+		}
+		chat, image = c, i
+	}
+
+	// chat and image must stay non-nil: the JSON contract forbids null here.
+	if chat == nil {
+		chat = []string{}
+	}
+	if image == nil {
+		image = []string{}
+	}
+
+	var warns []string
+	if len(chat) == 0 && len(image) == 0 {
+		// Caller can only get here by excluding everything, or from a catalog
+		// that was already empty. Either way "silence" would read as success.
+		warns = append(warns,
+			"no models selected; applying this would create a source that cannot serve any request")
+		return chat, image, warns, nil
+	}
+	if len(image) > 0 && len(chat) == 0 {
+		warns = append(warns,
+			"every model is an image model; AUTO chat requests have nothing to select")
+	}
+	if len(chat) > 0 && len(image) > 0 {
+		warns = append(warns,
+			"image models are only reachable through /v1/images/generations, not AUTO chat")
+	}
+	return chat, image, warns, nil
+}
+
+// discoverError is a client mistake rather than an upstream failure, which is
+// why it gets its own type instead of fmt.Errorf.
+type discoverError struct{ msg string }
+
+func (e *discoverError) Error() string { return e.msg }
+
+// defaultSourceName derives a source name from a base URL host, stripping the
+// version segment and any "api." prefix. It returns "" for input that is not a
+// URL at all — the caller must not fall back to echoing garbage into a source
+// name, since the name becomes a config key and an API path segment.
+func defaultSourceName(baseURL string) string {
+	s := strings.TrimSpace(baseURL)
+	if !strings.Contains(s, "://") {
+		return ""
+	}
+	s = s[strings.Index(s, "://")+3:]
+	if i := strings.IndexAny(s, "/?#"); i >= 0 {
+		s = s[:i]
+	}
+	if i := strings.LastIndex(s, ":"); i > 0 && !strings.Contains(s[i+1:], "]") {
+		s = s[:i] // strip port
+	}
+	// A host must still look like one after trimming the cosmetic prefixes.
+	s = strings.TrimPrefix(s, "api.")
+	s = strings.TrimPrefix(s, "www.")
+	if s == "" || strings.ContainsAny(s, " \t\"'<>") {
+		return ""
+	}
+	for _, r := range s {
+		switch {
+		case r >= 'a' && r <= 'z', r >= 'A' && r <= 'Z', r >= '0' && r <= '9':
+		case r == '.' || r == '-' || r == '_':
+		default:
+			return ""
+		}
+	}
+	if !strings.Contains(s, ".") {
+		// A bare label is far more likely to be a mistake than a hostname.
+		return ""
+	}
+	return s
+}

internal/gateway/discover_api_test.go

@@ -0,0 +1,314 @@
+package gateway
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+)
+
+// upstream that answers /v1/models with the real captured catalog
+func catalogUpstream(t *testing.T) *httptest.Server {
+	t.Helper()
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/v1/models" {
+			w.WriteHeader(http.StatusNotFound)
+			return
+		}
+		if got := r.Header.Get("Authorization"); got != "Bearer sk-secret" {
+			w.WriteHeader(http.StatusUnauthorized)
+			return
+		}
+		_, _ = w.Write([]byte(realCatalogBody))
+	}))
+	t.Cleanup(srv.Close)
+	return srv
+}
+
+func TestDefaultSourceName(t *testing.T) {
+	cases := map[string]string{
+		"https://api.vendor.com/v1":   "vendor.com",
+		"https://api.openai.com/v1":   "openai.com",
+		"https://www.vendor.com/v1":   "vendor.com",
+		"http://127.0.0.1:8080/v1":    "127.0.0.1",
+		"https://amr-link.open-design.ai/v1": "amr-link.open-design.ai",
+		"not a url at all":             "",
+	}
+	for in, want := range cases {
+		if got := defaultSourceName(in); got != want {
+			t.Errorf("defaultSourceName(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+func TestSelectModels_IncludeUnknownIsAnError(t *testing.T) {
+	chat := []string{"a", "b"}
+	image := []string{"img"}
+	// Silently dropping a typo'd id would create a source that looks
+	// configured and 404s on every request — much worse than a rejection.
+	_, _, _, err := selectModels(chat, image, []string{"a", "typo"}, nil)
+	if err == nil {
+		t.Fatal("expected an error for an include id that is not in the catalog")
+	}
+	if !strings.Contains(err.Error(), "typo") {
+		t.Errorf("error should name the offending id, got %q", err)
+	}
+}
+
+func TestSelectModels_IncludeNarrows(t *testing.T) {
+	chat, image, _, err := selectModels([]string{"a", "b", "c"}, []string{"img1", "img2"}, []string{"b", "img1"}, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(chat) != 1 || chat[0] != "b" {
+		t.Errorf("chat = %v, want [b]", chat)
+	}
+	if len(image) != 1 || image[0] != "img1" {
+		t.Errorf("image = %v, want [img1]", image)
+	}
+}
+
+func TestSelectModels_ExcludeUnknownIsFine(t *testing.T) {
+	chat, _, _, err := selectModels([]string{"a"}, nil, nil, []string{"never-existed"})
+	if err != nil {
+		t.Errorf("excluding a model that is not present should be allowed: %v", err)
+	}
+	if len(chat) != 1 {
+		t.Errorf("chat = %v, want [a] unchanged", chat)
+	}
+}
+
+func TestSelectModels_NeverReturnsNil(t *testing.T) {
+	chat, image, warns, err := selectModels(nil, nil, nil, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if chat == nil || image == nil {
+		t.Fatal("nil slices would serialize as JSON null and break the UI")
+	}
+	if len(warns) == 0 {
+		t.Error("an empty catalog deserves a warning, not silence")
+	}
+}
+
+func TestSelectModels_WarnsAboutImageOnlyCatalog(t *testing.T) {
+	_, _, warns, err := selectModels(nil, []string{"gpt-image-2"}, nil, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	joined := strings.Join(warns, " ")
+	if !strings.Contains(joined, "AUTO") {
+		t.Errorf("warnings %v should say AUTO has nothing to select", warns)
+	}
+}
+
+func TestSelectModels_WarnsThatImagesAreNotInAUTO(t *testing.T) {
+	_, _, warns, err := selectModels([]string{"a"}, []string{"img"}, nil, nil)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(strings.Join(warns, " "), "images/generations") {
+		t.Errorf("warnings %v should point at the image endpoint", warns)
+	}
+}
+
+// TestDiscoveryResultAndConfigRoundTrip checks the fields the WebUI will echo
+// back actually survive into a config.Source — a mismatch here means the
+// operator previews one set of values and saves another.
+func TestDiscoveryResultAndConfigRoundTrip(t *testing.T) {
+	res := &discoveryResult{
+		BaseURL:    "https://h",
+		Endpoint:   "/v1/chat/completions",
+		Models:     []string{"m1", "m2"},
+		ImageModels: []string{"img1"},
+	}
+	src := config.Source{
+		Name:      "x",
+		BaseURL:   res.BaseURL,
+		Endpoint:  res.Endpoint,
+		Models:    modelsToConfig(res.Models, res.ImageModels),
+	}
+	if len(src.Models) != 3 {
+		t.Fatalf("models = %d, want 3", len(src.Models))
+	}
+	if src.Endpoint != "/v1/chat/completions" {
+		t.Errorf("endpoint lost: %q", src.Endpoint)
+	}
+	var img int
+	for _, m := range src.Models {
+		if m.Kind == "image" {
+			img++
+		}
+	}
+	if img != 1 {
+		t.Errorf("image models = %d, want 1", img)
+	}
+}
+
+// TestDiscoverModels_SendsAuthorizationHeader pins the only way the feature
+// works against a real vendor: without the bearer token every paid /models
+// endpoint returns 401.
+func TestDiscoverModels_SendsAuthorizationHeader(t *testing.T) {
+	var seen string
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		seen = r.Header.Get("Authorization")
+		_, _ = w.Write([]byte(`{"data":[{"id":"m"}]}`))
+	}))
+	defer srv.Close()
+
+	if _, err := DiscoverModels(context.Background(), srv.URL+"/v1", "sk-secret"); err != nil {
+		t.Fatal(err)
+	}
+	if seen != "Bearer sk-secret" {
+		t.Errorf("Authorization = %q, want %q", seen, "Bearer sk-secret")
+	}
+}
+
+// TestDiscoverModels_WrongKeyIsUnauthorizedNotBadGateway pins the distinction
+// that decides whether the operator knows what to do next.
+//
+// A rejected key and an unreachable vendor are the two most common discovery
+// failures, and they need opposite actions: re-type the key, versus go check the
+// vendor's status page. Both used to surface as one generic error, so the
+// operator had to guess.
+func TestDiscoverModels_WrongKeyIsUnauthorizedNotBadGateway(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		w.WriteHeader(http.StatusUnauthorized)
+		_, _ = w.Write([]byte(`{"error":{"message":"invalid api key","code":"invalid_api_key"}}`))
+	}))
+	defer srv.Close()
+
+	_, err := DiscoverModels(context.Background(), srv.URL+"/v1", "sk-wrong")
+	if err == nil {
+		t.Fatal("expected a failure against a 401 endpoint")
+	}
+	var f *discoveryFailure
+	if !errors.As(err, &f) {
+		t.Fatalf("error should be classified, got %T: %v", err, err)
+	}
+	if got := f.StatusOrDefault(); got != http.StatusUnauthorized {
+		t.Errorf("status = %d, want %d — a wrong key must not read as an upstream outage",
+			got, http.StatusUnauthorized)
+	}
+	// The message has to name which credential is at fault, or the status code
+	// alone still sends the operator to the wrong place.
+	if !strings.Contains(strings.ToLower(f.Error()), "api key") {
+		t.Errorf("message %q should name the api key as the cause", f.Error())
+	}
+	if !strings.Contains(f.Error(), "/v1/models") {
+		t.Errorf("message %q should list what was tried", f.Error())
+	}
+}
+
+// TestDiscoverModels_UnreachableStaysBadGateway: the classification must not
+// over-correct — an unresolvable host is still the vendor's or the network's
+// problem, not the key's.
+func TestDiscoverModels_UnreachableStaysBadGateway(t *testing.T) {
+	_, err := DiscoverModels(context.Background(), "https://no-such-host-xyz.invalid/v1", "k")
+	if err == nil {
+		t.Fatal("expected a failure for an unresolvable host")
+	}
+	var f *discoveryFailure
+	if !errors.As(err, &f) {
+		t.Fatalf("error should be classified, got %T", err)
+	}
+	if got := f.StatusOrDefault(); got != http.StatusBadGateway {
+		t.Errorf("status = %d, want %d", got, http.StatusBadGateway)
+	}
+	if !strings.Contains(strings.ToLower(f.Error()), "resolve") {
+		t.Errorf("message %q should say the hostname does not resolve", f.Error())
+	}
+}
+
+// TestDiscoveryCandidates_NoDuplicateRequests: base "https://h/v1" with path
+// "/models" and base "https://h" with path "/v1/models" request the SAME URL.
+// Listing it twice in a failure message reads as "two things were tried" when
+// only one was.
+func TestDiscoveryCandidates_NoDuplicateRequests(t *testing.T) {
+	seen := map[string]bool{}
+	for _, c := range discoveryCandidates("https://h/v1") {
+		url := c.baseURL + c.modelsPath
+		if seen[url] {
+			t.Errorf("candidate %q would be requested twice", url)
+		}
+		seen[url] = true
+	}
+	if len(seen) < 2 {
+		t.Errorf("only %d distinct URLs; the search collapsed", len(seen))
+	}
+}
+
+// TestDiscoverModels_FailureListsEachURLOnce is the end-to-end form of the same
+// rule, on the error text an operator actually reads.
+func TestDiscoverModels_FailureListsEachURLOnce(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	}))
+	defer srv.Close()
+
+	_, err := DiscoverModels(context.Background(), srv.URL+"/v1", "k")
+	if err == nil {
+		t.Fatal("expected failure")
+	}
+	msg := err.Error()
+	needle := srv.URL + "/v1/models"
+	if n := strings.Count(msg, needle); n > 1 {
+		t.Errorf("URL %s appears %d times in %q", needle, n, msg)
+	}
+}
+
+func TestDiscoverModels_EmptyKeyOmitsHeader(t *testing.T) {
+	var present bool
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_, present = r.Header["Authorization"]
+		_, _ = w.Write([]byte(`{"data":[{"id":"m"}]}`))
+	}))
+	defer srv.Close()
+
+	if _, err := DiscoverModels(context.Background(), srv.URL+"/v1", ""); err != nil {
+		t.Fatal(err)
+	}
+	if present {
+		t.Error("an empty key should omit the header, not send an empty bearer")
+	}
+}
+
+// TestDiscoverModels_RealCatalogEndToEnd walks the actual captured response
+// through the whole path and asserts the numbers an operator would see.
+func TestDiscoverModels_RealCatalogEndToEnd(t *testing.T) {
+	srv := catalogUpstream(t)
+	res, err := DiscoverModels(context.Background(), srv.URL+"/v1", "sk-secret")
+	if err != nil {
+		t.Fatalf("discovery failed: %v", err)
+	}
+	if res.BaseURL != srv.URL+"/v1" {
+		t.Errorf("base = %q, want the typed URL unchanged", res.BaseURL)
+	}
+	if res.Endpoint != "" {
+		t.Errorf("endpoint = %q, want empty (base already carries /v1)", res.Endpoint)
+	}
+	if len(res.Models) != 2 || len(res.ImageModels) != 2 {
+		t.Errorf("chat/image = %d/%d, want 2/2", len(res.Models), len(res.ImageModels))
+	}
+	if res.Adapter != "openai" {
+		t.Errorf("adapter = %q, want openai", res.Adapter)
+	}
+
+	// The response must be JSON the UI can consume directly.
+	b, _ := json.Marshal(res)
+	var round map[string]interface{}
+	if err := json.Unmarshal(b, &round); err != nil {
+		t.Fatal(err)
+	}
+	if round["ok"] != nil {
+		// discoverResponse adds ok; discoveryResult does not. Both marshal
+		// fine, so this is only a shape note.
+		_ = round
+	}
+}

internal/gateway/discover_endpoint_test.go

@@ -0,0 +1,153 @@
+package gateway
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+)
+
+// TestComposeImageEndpoint_NoVersionBaseLeavesDefault is the counterpart of the
+// /v1 case: when the base carries no version segment, discovery must emit "" so
+// the gateway's own default "/v1/images/generations" composes.
+//
+// This is the shape every currently-working source uses — siliconflow, the only
+// source in production with kind:image, has base_url
+// "https://api.siliconflow.cn" — so writing a non-empty value here would break
+// exactly the configuration that is known to work today.
+func TestComposeImageEndpoint_NoVersionBaseLeavesDefault(t *testing.T) {
+	if got := composeImageEndpoint("https://api.siliconflow.cn"); got != "" {
+		t.Errorf("composeImageEndpoint = %q, want \"\" so the built-in default composes", got)
+	}
+	// And the composition it implies must be the vendor's real path.
+	base := "https://api.siliconflow.cn"
+	composed := base + "/v1/images/generations"
+	if !strings.HasSuffix(composed, "/v1/images/generations") {
+		t.Errorf("composed URL %q lost the image path", composed)
+	}
+}
+
+// TestDiscovery_BareHostNeedsVersionOnEndpointWhenCatalogIsVersioned
+// distinguishes two cases that look identical from the outside.
+//
+//   bare host + catalog at "/models"       → chat is "/chat/completions" (the
+//     adapter default, no endpoint written)
+//   bare host + catalog at "/v1/models"    → chat is "/v1/chat/completions",
+//     so an endpoint MUST be written
+//
+// The catalog path is the only evidence of whether the upstream wants the
+// version segment, and getting this backwards sends every chat request to a
+// path the vendor does not serve.
+func TestDiscovery_BareHostNeedsVersionOnEndpointWhenCatalogIsVersioned(t *testing.T) {
+	cases := []struct {
+		name         string
+		catalogPath  string
+		wantEndpoint string
+		wantImage    string
+	}{
+		{
+			name:         "catalog at the bare /models needs no endpoint",
+			catalogPath:  "/models",
+			wantEndpoint: "",
+			wantImage:    "",
+		},
+		{
+			name:         "catalog at /v1/models requires /v1 on the chat endpoint",
+			catalogPath:  "/v1/models",
+			wantEndpoint: "/v1/chat/completions",
+			wantImage:    "",
+		},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				if r.URL.Path != tc.catalogPath {
+					w.WriteHeader(http.StatusNotFound)
+					return
+				}
+				_, _ = w.Write([]byte(`{"data":[{"id":"m1"},{"id":"m2"}]}`))
+			}))
+			defer srv.Close()
+
+			res, err := DiscoverModels(context.Background(), srv.URL, "k")
+			if err != nil {
+				t.Fatalf("discovery failed: %v", err)
+			}
+			if res.BaseURL != srv.URL {
+				t.Errorf("base = %q, want %q", res.BaseURL, srv.URL)
+			}
+			if res.Endpoint != tc.wantEndpoint {
+				t.Errorf("endpoint = %q, want %q", res.Endpoint, tc.wantEndpoint)
+			}
+			if res.ImageEndpoint != tc.wantImage {
+				t.Errorf("image_endpoint = %q, want %q", res.ImageEndpoint, tc.wantImage)
+			}
+			// Whatever it wrote, the chat URL must reach the same version
+			// prefix the catalog was served under.
+			ep := res.Endpoint
+			if ep == "" {
+				ep = "/chat/completions"
+			}
+			if got := res.BaseURL + ep; strings.Contains(got, "/v1/v1") {
+				t.Errorf("chat URL %q has a doubled version", got)
+			}
+		})
+	}
+}
+
+// TestComposeEndpoints_BothFormsProduceWorkingURLs is the end-to-end
+// invariant across the two shapes, stated as the URLs the gateway will build.
+//
+// A base with a version and no image_endpoint produces "/v1/v1/…" — the exact
+// 404 observed before composeImageEndpoint existed. A base without one and a
+// non-empty image_endpoint produces a path missing the version. Both are
+// silent, expensive mistakes, so they are asserted literally.
+func TestComposeEndpoints_BothFormsProduceWorkingURLs(t *testing.T) {
+	cases := []struct {
+		name        string
+		base        string
+		wantChat    string
+		wantImage   string
+	}{
+		{
+			name:      "bare host: adapter and image defaults compose",
+			base:      "https://h",
+			wantChat:  "https://h/chat/completions",
+			// openai's adapter endpoint is /chat/completions (verified in
+			// internal/lua/adapters/openai.lua: adapter.endpoint).
+			wantImage: "https://h/v1/images/generations",
+		},
+		{
+			name:      "versioned base: the version moves onto both endpoints",
+			base:      "https://h/v1",
+			wantChat:  "https://h/v1/chat/completions",
+			wantImage: "https://h/v1/images/generations",
+		},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			ep := composeEndpoint("/models", tc.base, tc.base)
+			if ep == "" {
+				ep = "/chat/completions" // the adapter default
+			}
+			iep := composeImageEndpoint(tc.base)
+			if iep == "" {
+				iep = "/v1/images/generations" // the gateway default
+			}
+			gotChat := tc.base + ep
+			gotImage := tc.base + iep
+			if gotChat != tc.wantChat {
+				t.Errorf("chat URL = %q, want %q", gotChat, tc.wantChat)
+			}
+			if gotImage != tc.wantImage {
+				t.Errorf("image URL = %q, want %q", gotImage, tc.wantImage)
+			}
+			for _, u := range []string{gotChat, gotImage} {
+				if strings.Contains(u, "/v1/v1") {
+					t.Errorf("%q has a doubled version segment", u)
+				}
+			}
+		})
+	}
+}

internal/gateway/discover_mutate_test.go

@@ -0,0 +1,303 @@
+package gateway
+
+// Real mutation injection: rewrites the IMPLEMENTATION on disk, re-runs the
+// suite, and requires it to go red.
+//
+// Why this exists and why it is a Go file rather than a shell script:
+//
+// The first version was a shell script. Its -run filter contained a regex that
+// the shell mangled, so every mutant reported "SURVIVED" while go test printed
+// "[no tests to run]". Four mutants looked like judgement failures and were
+// actually never executed. A judgement that cannot tell "tests passed" from
+// "no tests ran" is not a judgement.
+//
+// This version therefore refuses to report anything unless it has first proved
+// the filter selects a non-trivial number of tests (see baseline below), and it
+// treats "[no tests to run]" as a hard error rather than a green run.
+//
+// Run: go test -tags luajit ./internal/gateway/ -run TestRealMutants -v
+
+import (
+	"os"
+	"os/exec"
+	"path/filepath"
+	"sort"
+	"strings"
+	"testing"
+)
+
+// mutationFilter must select every assertion that guards the code below. It is
+// deliberately broad: a narrow filter is how mutants get to survive for free.
+const mutationFilter = "Test(ParseModelCatalog|DiscoveredModelKind|Discover|TrimVersion|VersionSegment|ComposeEndpoint|ComposeImageEndpoint|ComposeEndpoints|IsVersionSegment|ModelsToConfig|DiscoveryResult|DiscoveryCandidates|Discovery_BareHost)"
+
+type mutation struct {
+	name string
+	from string
+	to   string
+}
+
+func TestRealMutants(t *testing.T) {
+	if testing.Short() {
+		t.Skip("mutation campaign compiles the package repeatedly")
+	}
+	wd, err := os.Getwd()
+	if err != nil {
+		t.Fatal(err)
+	}
+	target := filepath.Join(wd, "discover.go")
+	orig, err := os.ReadFile(target)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer func() {
+		if werr := os.WriteFile(target, orig, 0o644); werr != nil {
+			t.Fatalf("restore discover.go: %v", werr)
+		}
+	}()
+
+	// --- prove the harness is armed before trusting any verdict ---
+	// A filter that selects nothing exits 0, which is indistinguishable from a
+	// real pass. The only way to tell them apart is to compare against a filter
+	// KNOWN to be empty: if that empty filter produces the same result as the
+	// real one, the real one is probably empty too.
+	//
+	// Note go test does NOT print "[no tests to run]" for a wholly unmatched
+	// filter (it prints that only when the package has other matching tests), so
+	// the check has to compare outputs rather than look for a marker string.
+	// Getting this backwards made the harness refuse to run at all.
+	emptyOut := runSuite(wd, "TestZZZNoSuchTestExists")
+	realOut := runSuite(wd, mutationFilter)
+	if countTests(emptyOut) != 0 {
+		t.Fatalf("a filter matching nothing selected %d tests; the harness's own probe is broken",
+			countTests(emptyOut))
+	}
+	// Kept for the no-op check below: a mutant whose output matches the
+	// baseline never changed behaviour, so it is an invalid mutation rather
+	// than a gap in the tests.
+	baseOut := realOut
+	if strings.Contains(realOut, "[no tests to run]") {
+		t.Fatalf("baseline selected no tests:\n%s", realOut)
+	}
+	baseCount := countTests(realOut)
+	if baseCount < 25 {
+		t.Fatalf("baseline ran only %d tests; the filter is too narrow to judge anything", baseCount)
+	}
+	if strings.Contains(realOut, "FAIL") {
+		t.Logf("note: baseline is not green; kill verdicts remain valid")
+	}
+	t.Logf("baseline: %d tests selected, empty filter selects %d",
+		baseCount, countTests(emptyOut))
+
+	muts := []mutation{
+		{
+			// Written as a condition that still READS modOut, because a
+			// mutation that leaves a variable unused fails to compile — and a
+			// build failure is not a judgement (see the harness note above).
+			name: "image kind ignores upstream output modality",
+			from: `if m.Capabilities.ImageGeneration || has(modOut, "image") {`,
+			to:   "if m.Capabilities.ImageGeneration || (len(modOut) > 0 && false) {",
+		},
+		{
+			name: "kind defaults to image (guessing from names)",
+			from: "\t}\n\treturn \"chat\"\n}\n\n// modalityList",
+			to:   "\t}\n\treturn \"image\"\n}\n\n// modalityList",
+		},
+		{
+			name: "nested metadata no longer read",
+			from: "\tcopyFields(m.Metadata)      // nested first",
+			to:   "\tcopyFields(json.RawMessage(nil)) // mutant: nested metadata dropped",
+		},
+		{
+			name: "non-catalog bodies accepted",
+			from: "\tif trimmed == \"\" || trimmed == \"null\" {\n\t\treturn nil, fmt.Errorf(\"empty response\")\n\t}",
+			to:   "\tif false && (trimmed == \"\" || trimmed == \"null\") {\n\t\treturn nil, fmt.Errorf(\"empty response\")\n\t}",
+		},
+		{
+			// Removing the bare "/models" candidate must be observable: a vendor
+			// serving its catalog at the bare path would otherwise look
+			// unreachable. The `len(...) >= 0` is a tautology that only exists to
+			// keep the expression syntactically valid after the call is dropped.
+			name: "URL fallback search removed",
+			from: "\t\tadd(raw, \"/models\", 0)",
+			to:   "\t\tif len(\"/models\") >= 0 {\n\t\t\t_ = hasVersion\n\t\t}",
+		},
+		{
+			name: "version segment never moved onto the chat endpoint",
+			from: "\tif version == \"\" {\n\t\treturn \"\" // adapter default already composes\n\t}",
+			to:   "\tif version == \"\" || len(version) > 0 {\n\t\treturn \"\"\n\t}",
+		},
+		{
+			name: "image endpoint never corrected (/v1/v1 regression)",
+			from: "\tif b := trimVersionSuffix(persistedBase); b != persistedBase {\n\t\treturn \"/images/generations\"\n\t}",
+			to:   "\tif b := trimVersionSuffix(persistedBase); b == b && b != b {\n\t\treturn \"/images/generations\"\n\t}",
+		},
+		{
+			name: "version test loosened so 'vision' counts as a version",
+			from: "\tsuffix := strings.ToLower(seg[i:])",
+			to:   "\tsuffix := strings.ToLower(seg[i:])\n\tif len(suffix) >= 0 {\n\t\tsuffix = \"beta\"\n\t}",
+		},
+		{
+			// The guard must be removable WITHOUT the mutant becoming a
+			// semantic no-op. Two earlier attempts failed:
+			//
+			//   `&& out.BaseURL != ""`  — the guard this test exercises uses a
+			//     zero-value struct whose BaseURL is empty, so the added
+			//     condition made the mutant behave identically to the original.
+			//   `&& len(out.BaseURL) >= 0` — always true for any string, so the
+			//     mutant was STILL an identity transform.
+			//
+			// Reading the variable into a blank assignment keeps the compiler
+			// happy without changing behaviour, which is what makes the
+			// mutant observable.
+			name: "nil slice guard removed from JSON output",
+			from: "\tif out.ImageModels == nil {\n\t\tout.ImageModels = []string{}\n\t}",
+			to:   "\tif out.ImageModels == nil {\n\t\t_ = out.ImageModels\n\t}",
+		},
+		{
+			name: "auth rejection no longer classified",
+			from: "\tif status != http.StatusUnauthorized && status != http.StatusForbidden {\n\t\treturn \"\"\n\t}",
+			to:   "\tif status != http.StatusUnauthorized && status != http.StatusForbidden || len(body) >= 0 {\n\t\treturn \"\"\n\t}",
+		},
+		{
+			name: "duplicate candidate URLs no longer collapsed",
+			from: "\t\tif seenPair[pair] || seenURL[base+path] {",
+			to:   "\t\tif len(pair) > 0 && (seenPair[pair] || seenURL[base+path]) && false {",
+		},
+	}
+
+	survived := 0
+	for _, m := range muts {
+		t.Run(m.name, func(t *testing.T) {
+			if !strings.Contains(string(orig), m.from) {
+				t.Fatalf("anchor not found in discover.go — this mutation would be a no-op:\n%q", m.from)
+			}
+			mutated := strings.Replace(string(orig), m.from, m.to, 1)
+			if mutated == string(orig) {
+				t.Fatal("replacement changed nothing")
+			}
+			if werr := os.WriteFile(target, []byte(mutated), 0o644); werr != nil {
+				t.Fatal(werr)
+			}
+			defer os.WriteFile(target, orig, 0o644)
+
+			out := runSuite(wd, mutationFilter)
+			// A build failure kills EVERY mutant, so it would mask a broken
+			// harness as a perfect score. Distinguish it from a real test
+			// failure and fail loudly instead.
+			if strings.Contains(out, "[build failed]") {
+				t.Fatalf("mutant broke the build — that is not a judgement, it is a broken harness:\n%s", firstLines(out, 20))
+			}
+			if countTests(out) == 0 {
+				t.Fatalf("filter selected nothing — the verdict would be meaningless")
+			}
+			if !strings.Contains(out, "FAIL") {
+				// The mutant compiled and the suite stayed green. Before calling
+				// this a survivor, check the mutation was not a no-op: an earlier
+				// version reported four "survivors" that were really invalid
+				// mutations whose added guard made them behaviourally identical
+				// to the original. A silent no-op is a broken mutation, and
+				// conflating the two sends the next reader hunting for a missing
+				// test that does not exist.
+				if behaviourOnly(out) == behaviourOnly(baseOut) {
+					t.Fatalf("mutant is a no-op: the suite output is identical to the "+
+						"baseline, so this mutation never changed behaviour. Fix the "+
+						"mutation rather than looking for a missing test.\nmutant: %q", m.to)
+				}
+				survived++
+				t.Errorf("SURVIVED %s — no test noticed this wrong implementation", m.name)
+				return
+			}
+			t.Logf("KILL  %s", m.name)
+		})
+	}
+	if survived > 0 {
+		t.Errorf("%d of %d mutants survived", survived, len(muts))
+	}
+}
+
+// runSuite invokes the go tool on this package with the given -run filter.
+//
+// The luajit build tag is mandatory: without it the package falls back to plain
+// Lua 5.1, billing.lua's `goto` fails to compile, and the whole package fails
+// to build — which the harness would misread as "every mutant killed".
+func runSuite(wd, filter string) string {
+	tool := os.Getenv("GO_TEST_GO")
+	if tool == "" {
+		tool = "go"
+	}
+	cmd := exec.Command(tool, "test",
+		"-tags", "luajit",
+		".", "-run", filter, "-count=1", "-v")
+	cmd.Dir = wd
+	cmd.Env = os.Environ()
+	out, _ := cmd.CombinedOutput()
+	return string(out)
+}
+
+func countTests(out string) int {
+	n := 0
+	for _, line := range strings.Split(out, "\n") {
+		// Subtests are indented; only count top-level ones so a single test
+		// with a table of cases cannot inflate the count.
+		if strings.HasPrefix(line, "=== RUN   Test") && !strings.Contains(line, "/") {
+			n++
+		}
+	}
+	return n
+}
+
+// behaviourOnly strips the parts of `go test -v` output that vary between runs
+// (timings, goroutine ids, temp paths) so two runs can be compared for equality.
+// If a mutant produces output identical to the baseline under this projection,
+// the mutation did not change observable behaviour and is therefore invalid.
+func behaviourOnly(out string) string {
+	var kept []string
+	for _, line := range strings.Split(out, "\n") {
+		trimmed := strings.TrimSpace(line)
+		switch {
+		// Keep only the outcome lines. Test NAMES are deliberately dropped:
+		// two runs of the same suite can legitimately order or cache tests
+		// differently, and keeping the names made every mutant look like it had
+		// "changed behaviour" — which defeats the no-op detector entirely.
+		case strings.HasPrefix(trimmed, "--- PASS"), strings.HasPrefix(trimmed, "--- FAIL"),
+			strings.HasPrefix(trimmed, "--- SKIP"),
+			strings.HasPrefix(trimmed, "ok  "), trimmed == "FAIL",
+			strings.Contains(trimmed, ".go:"):
+			kept = append(kept, trimmed)
+		}
+	}
+	// Counting outcomes is stronger than comparing the sequence: a mutation
+	// that swaps which test fails still changes the multiset, and one that
+	// changes nothing cannot.
+	sort.Strings(kept)
+	return strings.Join(kept, "\n")
+}
+
+// firstLines trims output for an error message; a failing nested run can be
+// thousands of lines and the cause is always near the top.
+func firstLines(s string, n int) string {
+	lines := strings.Split(s, "\n")
+	if len(lines) > n {
+		lines = lines[:n]
+	}
+	return strings.Join(lines, "\n")
+}
+
+// TestMutationFilterCoversEveryGuard keeps the filter honest. Dropping a group
+// from it silently disarms that area's mutants.
+func TestMutationFilterCoversEveryGuard(t *testing.T) {
+	for _, group := range []string{
+		"ParseModelCatalog",   // nested metadata + non-catalog rejection
+		"DiscoveredModelKind", // kind follows upstream metadata
+		"Discover",            // URL fallback, auth classification
+		"ComposeEndpoint",     // version segment exactly once (chat)
+		"ComposeImageEndpoint",
+		"ComposeEndpoints", // the composed URLs, both base shapes
+		"IsVersionSegment", // "v1beta" vs "vision"
+		"DiscoveryResult",  // nil slices are arrays
+	} {
+		if !strings.Contains(mutationFilter, group) {
+			t.Errorf("mutation filter omits %s; mutants there survive silently", group)
+		}
+	}
+}

internal/gateway/discover_mutation_test.go

@@ -0,0 +1,171 @@
+package gateway
+
+// Mutation harness for the discovery code. Kept in the tree (not in a shell
+// script) so the -run filter is a Go string literal rather than something the
+// shell can mangle into "[no tests to run]" — which is exactly how an earlier
+// version of this harness reported four surviving mutants that were simply
+// never executed.
+//
+// Each mutant flips one behaviour; TestMutationHarness_Fails proves the
+// harness itself is armed.
+
+import (
+	"encoding/json"
+	"net/http"
+	"strings"
+	"testing"
+)
+
+// discoveryTestFilter must cover every assertion that guards the behaviour
+// below. Narrowing it is how mutants get to "survive" for free.
+const discoveryTestFilter = "Test(ParseModelCatalog|DiscoveredModelKind|Discover|TrimVersion|VersionSegment|ComposeEndpoint|IsVersionSegment|ModelsToConfig|DiscoveryResult|DiscoveryCandidates)"
+
+func TestDiscoveryFilterRunsRealTests(t *testing.T) {
+	if discoveryTestFilter == "" {
+		t.Fatal("filter is empty; every mutant would survive trivially")
+	}
+	// The whole point: the catalog parser test must be inside the filter, or a
+	// mutant that stops reading nested metadata survives with a green run.
+	for _, group := range []string{
+		"ParseModelCatalog",   // nested metadata + non-catalog rejection
+		"DiscoveredModelKind", // kind must follow upstream metadata
+		"Discover",            // URL fallback + endpoint composition
+		"ComposeEndpoint",     // version segment exactly once
+		"IsVersionSegment",    // "v1beta" vs "vision"
+	} {
+		if !strings.Contains(discoveryTestFilter, group) {
+			t.Errorf("filter %q skips %s tests; mutants in that area survive", discoveryTestFilter, group)
+		}
+	}
+}
+
+// TestMutantsAreDetectable is the harness's own regression guard: it runs the
+// filter's worth of assertions against deliberately broken inputs and requires
+// each to be rejected. If a future edit makes one of these pass, the
+// corresponding protection is gone.
+func TestMutantsAreDetectable(t *testing.T) {
+	t.Run("image classification follows upstream metadata, not the model name", func(t *testing.T) {
+		// The mutant: drop the output-modality check, so every model is chat.
+		m := upstreamModel{ID: "gpt-image-2"}
+		m.Architecture.OutputModalities = []string{"image"}
+		if got := discoveredModelKind(m); got != "image" {
+			t.Errorf("mutant survived: got %q, want \"image\"", got)
+		}
+	})
+
+	t.Run("nested metadata is read", func(t *testing.T) {
+		var ms []upstreamModel
+		if err := json.Unmarshal([]byte(realCatalogBody), &ms); err == nil {
+			t.Fatal("fixture must be a wrapper object, not a bare array")
+		}
+		var wrapper struct {
+			Data []upstreamModel `json:"data"`
+		}
+		if err := json.Unmarshal([]byte(realCatalogBody), &wrapper); err != nil {
+			t.Fatal(err)
+		}
+		if len(wrapper.Data) == 0 {
+			t.Fatal("fixture parsed to zero models")
+		}
+		// The mutant: skip copyFields(m.Metadata) entirely.
+		meta := modelMeta(wrapper.Data[0])
+		if meta == nil {
+			t.Fatal("mutant survived: metadata is nil, so cost/context_limit are lost")
+		}
+		for _, k := range []string{"context_limit", "cost", "context_budget"} {
+			if _, ok := meta[k]; !ok {
+				t.Errorf("mutant survived: %s dropped from nested metadata", k)
+			}
+		}
+	})
+
+	t.Run("non-catalog responses are rejected", func(t *testing.T) {
+		for _, body := range []string{`<html>login</html>`, `null`, `{"error":{"message":"x"}}`, `{}`} {
+			if _, err := parseModelCatalog([]byte(body)); err == nil {
+				t.Errorf("mutant survived: %q accepted as a catalog", body)
+			}
+		}
+	})
+
+	t.Run("version segments are distinguished from real path segments", func(t *testing.T) {
+		if !isVersionSegment("v1beta") {
+			t.Error("mutant survived: v1beta not treated as a version")
+		}
+		for _, seg := range []string{"vision", "vertex", "v1x"} {
+			if isVersionSegment(seg) {
+				t.Errorf("mutant survived: %q wrongly treated as a version", seg)
+			}
+		}
+	})
+
+	t.Run("endpoint composition never duplicates the version segment", func(t *testing.T) {
+		// The mutant: never move the segment onto the endpoint.
+		if got := composeEndpoint("/v1/models", "https://h", "https://h/v1"); got != "/v1/chat/completions" {
+			t.Errorf("mutant survived: endpoint = %q", got)
+		}
+		// The other mutant: always move it, even when the base keeps it.
+		if got := composeEndpoint("/models", "https://h/v1", "https://h/v1"); got != "" {
+			t.Errorf("mutant survived: endpoint = %q would double /v1", got)
+		}
+	})
+
+	t.Run("image_models is never nil", func(t *testing.T) {
+		res := &discoveryResult{Models: []string{"a"}, ImageModels: nil}
+		b, _ := json.Marshal(res)
+		if strings.Contains(string(b), `"image_models":null`) {
+			t.Errorf("mutant survived: %s", b)
+		}
+	})
+
+	t.Run("a rejected key is reported as unauthorized, not as an outage", func(t *testing.T) {
+		// The mutant: collapse every failure into one generic error, which is
+		// what made a typo'd key look like the vendor being down.
+		if classifyAuthStatus(http.StatusUnauthorized, []byte(`{"error":"invalid api key"}`)) == "" {
+			t.Error("mutant survived: 401 is not classified")
+		}
+		if classifyAuthStatus(http.StatusNotFound, []byte("nope")) != "" {
+			t.Error("a 404 must not be reported as an auth problem")
+		}
+		if classifyAuthStatus(http.StatusOK, nil) != "" {
+			t.Error("a 200 must not be reported as an auth problem")
+		}
+	})
+
+	t.Run("duplicate candidate URLs are collapsed", func(t *testing.T) {
+		seen := map[string]bool{}
+		for _, c := range discoveryCandidates("https://h/v1") {
+			u := c.baseURL + c.modelsPath
+			if seen[u] {
+				t.Errorf("mutant survived: %s requested twice", u)
+			}
+			seen[u] = true
+		}
+	})
+
+	t.Run("discovery falls back across URL forms", func(t *testing.T) {
+		// The coverage requirement is that BOTH catalog paths are reachable for
+		// a typed base, and both for a bare host. Asserting an exact count kept
+		// drifting as the candidate list was fixed, and each time it measured
+		// the current list rather than the property that matters.
+		for _, base := range []string{"https://h/v1", "https://h"} {
+			distinct := map[string]bool{}
+			for _, c := range discoveryCandidates(base) {
+				distinct[c.baseURL+c.modelsPath] = true
+			}
+			for _, want := range []string{base + "/v1/models", base + "/models"} {
+				if !distinct[want] {
+					t.Errorf("mutant survived: %s never requests %s", base, want)
+				}
+			}
+		}
+	})
+
+	t.Run("only the catalog is fetched, never each model", func(t *testing.T) {
+		// Guarding a NON-mutation: the design decision that discovery must not
+		// spend money. If a future change adds per-model probes, the counting
+		// test below catches it; this assertion documents the intent.
+		if discoveryModelsPath != "/v1/models" {
+			t.Errorf("catalog path changed to %q", discoveryModelsPath)
+		}
+	})
+}

internal/gateway/discover_test.go

@@ -0,0 +1,648 @@
+package gateway
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+)
+
+// realCatalogBody is the actual /models response captured from the amr-link
+// endpoint, not a hand-written fixture. Two properties of it are the reason
+// this test exists at all:
+//
+//   - it nests modalities under "architecture", which the naive parser misses;
+//   - it lists five image-generation ids among 34 chat ids, so a parser that
+//     returns everything as "chat" silently poisons the AUTO chain.
+const realCatalogBody = `{"data":[{"id":"claude-fable-5","object":"model","created":1781081344,"owned_by":"vela","architecture":{"input_modalities":["text","image","file"],"output_modalities":["text"]},"metadata":{"cost":"very_high","capability":"best_quality","context_limit":1000000,"context_budget":232000},"default":false,"enabled":true},{"id":"gpt-image-2","object":"model","architecture":{"input_modalities":["text","image"],"output_modalities":["image"]},"metadata":{"capability":"best_quality"},"enabled":true},{"id":"deepseek-v4-flash","object":"model","architecture":{"input_modalities":["text"],"output_modalities":["text"]},"metadata":{"cost":"medium","capability":"standard","context_limit":1048000},"enabled":true},{"id":"seedream-5.0-pro","object":"model","architecture":{"input_modalities":["text","image"],"output_modalities":["image"]},"metadata":{"capability":"best_quality"},"enabled":true}]}`
+
+func TestParseModelCatalog_RealUpstreamShape(t *testing.T) {
+	models, err := parseModelCatalog([]byte(realCatalogBody))
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	if len(models) != 4 {
+		t.Fatalf("want 4 models, got %d", len(models))
+	}
+
+	kind := map[string]string{}
+	for _, m := range models {
+		kind[m.ID] = m.Kind
+	}
+	// The two image generators must be classified by the upstream's declared
+	// output modality, not by their names.
+	for _, id := range []string{"gpt-image-2", "seedream-5.0-pro"} {
+		if kind[id] != "image" {
+			t.Errorf("%s: got kind %q, want \"image\"", id, kind[id])
+		}
+	}
+	for _, id := range []string{"claude-fable-5", "deepseek-v4-flash"} {
+		if kind[id] != "chat" {
+			t.Errorf("%s: got kind %q, want \"chat\"", id, kind[id])
+		}
+	}
+
+	// Metadata the upstream sent must survive; keys it did not send must not
+	// appear as zero values.
+	if got := models[0].Meta["context_limit"]; got == nil {
+		t.Error("claude-fable-5: context_limit dropped (it is nested under metadata in the real response)")
+	} else if f, ok := got.(float64); !ok || f != 1000000 {
+		t.Errorf("context_limit = %#v, want numeric 1000000 so the UI can sort on it", got)
+	}
+	if _, present := models[0].Meta["cost"]; !present {
+		t.Error("claude-fable-5: cost dropped")
+	}
+	// context_budget is present on this entry and must not be dropped either.
+	if _, present := models[0].Meta["context_budget"]; !present {
+		t.Error("claude-fable-5: context_budget dropped")
+	}
+	// deepseek-v4-flash DOES send metadata; gpt-image-2 sends only capability.
+	// Neither may lose it.
+	if models[2].Meta["context_limit"] == nil {
+		t.Error("deepseek-v4-flash: context_limit dropped")
+	}
+	if models[2].Meta["cost"] == nil {
+		t.Error("deepseek-v4-flash: cost dropped")
+	}
+	if _, bogus := models[0].Meta["made_up_field"]; bogus {
+		t.Error("invented a metadata key that the upstream never sent")
+	}
+}
+
+// TestDiscoveredModelKind_ImageOnlyByUpstreamMetadata is the regression guard
+// for the failure mode that motivated keeping kind="chat" as the default: a
+// name-based heuristic marks "ideogram-v2" or "gpt-image-1" as an image model,
+// but a catalog that also serves chat for that id would then have its chat
+// entry removed from the AUTO chain. Only the upstream's declared OUTPUT
+// modality may classify a model as image.
+func TestDiscoveredModelKind_ImageOnlyByUpstreamMetadata(t *testing.T) {
+	cases := []struct {
+		name string
+		in   upstreamModel
+		want string
+	}{
+		{
+			// A catalog that DECLARES output modality image is authoritative.
+			// The "kind" string saying chat must not override it.
+			name: "declared output modality image wins over a chat kind string",
+			in: func() upstreamModel {
+				var m upstreamModel
+				m.Kind = "chat"
+				m.Architecture.OutputModalities = []string{"image"}
+				return m
+			}(),
+			want: "image",
+		},
+		{
+			name: "bare array modality string",
+			in:   upstreamModel{ID: "x", Modality: "text-to-image"},
+			want: "image",
+		},
+		{
+			name: "image INPUT is still a chat model",
+			in: func() upstreamModel {
+				var m upstreamModel
+				m.Architecture.InputModalities = []string{"text", "image"}
+				return m
+			}(),
+			want: "chat",
+		},
+		{
+			name: "file input is still a chat model",
+			in: func() upstreamModel {
+				var m upstreamModel
+				m.Architecture.InputModalities = []string{"text", "file"}
+				return m
+			}(),
+			want: "chat",
+		},
+		{
+			name: "vision capability flag is chat",
+			in:   upstreamModel{ID: "y", Capabilities: modelCaps{Vision: true}},
+			want: "chat",
+		},
+		{
+			name: "capabilities.image_generation is image",
+			in:   upstreamModel{ID: "z", Capabilities: modelCaps{ImageGeneration: true}},
+			want: "image",
+		},
+		{
+			name: "no modality info at all defaults to chat",
+			in:   upstreamModel{ID: "plain"},
+			want: "chat",
+		},
+		{
+			// The default must be chat even when the name LOOKS like an image
+			// model. Guessing from the name is what puts gpt-image-2 into the
+			// chat AUTO chain, where every request to it fails.
+			name: "image-looking name without upstream metadata stays chat",
+			in:   upstreamModel{ID: "gpt-image-2"},
+			want: "chat",
+		},
+		{
+			name: "vision-looking name without upstream metadata stays chat",
+			in:   upstreamModel{ID: "ideogram-v2"},
+			want: "chat",
+		},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			if got := discoveredModelKind(tc.in); got != tc.want {
+				t.Errorf("got %q, want %q", got, tc.want)
+			}
+		})
+	}
+}
+
+func TestParseModelCatalog_RejectsNonCatalog(t *testing.T) {
+	// A 200 with valid JSON that is not a catalog must NOT be reported as an
+	// empty result. Reporting zero models would let a login page or a
+	// rate-limit body create a source with no models in it.
+	for name, body := range map[string]string{
+		"login page html": `<html><body>Sign in</body></html>`,
+		"rate limit json": `{"error":{"message":"slow down","type":"rate_limit"}}`,
+		"empty object":    `{}`,
+		"null":            `null`,
+		"wrong shape":     `{"objects":[{"id":"m"}]}`,
+	} {
+		t.Run(name, func(t *testing.T) {
+			if _, err := parseModelCatalog([]byte(body)); err == nil {
+				t.Errorf("expected an error for %q, got nil (would create an empty source)", name)
+			}
+		})
+	}
+}
+
+func TestParseModelCatalog_AcceptsBareArray(t *testing.T) {
+	models, err := parseModelCatalog([]byte(`[{"id":"a"},{"id":"b"}]`))
+	if err != nil {
+		t.Fatalf("bare array should parse: %v", err)
+	}
+	if len(models) != 2 {
+		t.Fatalf("want 2, got %d", len(models))
+	}
+}
+
+// TestDiscoverModels_NormalizesWhenOnlyBareModelsAnswers: operator pastes
+// "https://host/v1" (the form every SDK example shows) but the vendor serves
+// the catalog at "https://host/models" — the version segment on the path is
+// implied, not doubled. Discovery must strip the segment off the base and
+// report it on the endpoint, because openai's adapter endpoint is
+// "/chat/completions" and the two are concatenated.
+func TestDiscoverModels_NormalizesWhenOnlyBareModelsAnswers(t *testing.T) {
+	var paths []string
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		paths = append(paths, r.URL.Path)
+		if r.URL.Path != "/models" {
+			w.WriteHeader(http.StatusNotFound)
+			return
+		}
+		_, _ = w.Write([]byte(realCatalogBody))
+	}))
+	defer srv.Close()
+
+	res, err := DiscoverModels(context.Background(), srv.URL+"/v1", "sk-test")
+	if err != nil {
+		t.Fatalf("discovery failed: %v", err)
+	}
+	if res.BaseURL != srv.URL {
+		t.Errorf("base_url = %q, want %q (version segment moved off the base)", res.BaseURL, srv.URL)
+	}
+	// "/models" carries no version segment, but the operator typed one — the
+	// endpoint has to restore it or the chat call would miss /v1.
+	if res.Endpoint != "/v1/chat/completions" {
+		t.Errorf("endpoint = %q, want %q", res.Endpoint, "/v1/chat/completions")
+	}
+	if res.Note == "" {
+		t.Error("silently rewriting the operator's base_url is worse than saying so")
+	}
+	if len(res.Models) != 2 || len(res.ImageModels) != 2 {
+		t.Errorf("got %d chat / %d image, want 2 / 2", len(res.Models), len(res.ImageModels))
+	}
+	if len(paths) < 2 {
+		t.Errorf("tried only %v; expected a fallback search", paths)
+	}
+}
+
+// TestDiscoverModels_BareHostWithVersionedModelsPath: operator pastes the host
+// root, vendor serves /v1/models. The base needs no rewriting; the endpoint
+// does.
+func TestDiscoverModels_BareHostWithVersionedModelsPath(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/v1/models" {
+			w.WriteHeader(http.StatusNotFound)
+			return
+		}
+		_, _ = w.Write([]byte(`{"data":[{"id":"m1"}]}`))
+	}))
+	defer srv.Close()
+
+	res, err := DiscoverModels(context.Background(), srv.URL, "")
+	if err != nil {
+		t.Fatalf("discovery failed: %v", err)
+	}
+	if res.BaseURL != srv.URL {
+		t.Errorf("base_url = %q, want %q", res.BaseURL, srv.URL)
+	}
+	if res.Endpoint != "/v1/chat/completions" {
+		t.Errorf("endpoint = %q, want %q", res.Endpoint, "/v1/chat/completions")
+	}
+}
+
+// TestDiscoverModels_KeepsWorkingBaseURLUnchanged covers the vendor that really
+// does serve "/v1/models": the typed URL must survive untouched so no needless
+// config rewrite is suggested.
+func TestDiscoverModels_KeepsWorkingBaseURLUnchanged(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/v1/models" {
+			w.WriteHeader(http.StatusNotFound)
+			return
+		}
+		_, _ = w.Write([]byte(`{"data":[{"id":"m1"},{"id":"m2"}]}`))
+	}))
+	defer srv.Close()
+
+	// Base typed WITH "/v1", catalog served at "…/v1/models". That is exactly
+	// the amr-link shape: first candidate succeeds, nothing is rewritten, and
+	// the adapter's default endpoint already composes.
+	res, err := DiscoverModels(context.Background(), srv.URL+"/v1", "")
+	if err != nil {
+		t.Fatalf("discovery failed: %v", err)
+	}
+	if res.BaseURL != srv.URL+"/v1" {
+		t.Errorf("base_url = %q, want it unchanged", res.BaseURL)
+	}
+	if res.Endpoint != "" {
+		t.Errorf("endpoint = %q, want empty (the typed base already carries /v1)", res.Endpoint)
+	}
+	if res.Note != "" {
+		t.Errorf("note = %q, want empty when nothing was rewritten", res.Note)
+	}
+	if len(res.Models) != 2 {
+		t.Errorf("got %d chat models, want 2", len(res.Models))
+	}
+}
+
+// TestDiscoverModels_ErrorNamesEveryTriedURL: a failed probe is useless if it
+// only reports the last URL. The operator needs to see the full search.
+func TestDiscoverModels_ErrorNamesEveryTriedURL(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusNotFound)
+	}))
+	defer srv.Close()
+
+	_, err := DiscoverModels(context.Background(), srv.URL+"/v1", "k")
+	if err == nil {
+		t.Fatal("expected failure against an all-404 server")
+	}
+	msg := err.Error()
+	for _, want := range []string{"/models", "/v1/models", "404"} {
+		if !strings.Contains(msg, want) {
+			t.Errorf("error %q should mention %q", msg, want)
+		}
+	}
+}
+
+func TestDiscoveryCandidates_DedupesAndPrefersTyped(t *testing.T) {
+	cands := discoveryCandidates("https://h/v1/")
+	if len(cands) == 0 {
+		t.Fatal("no candidates")
+	}
+	seen := map[string]bool{}
+	for _, c := range cands {
+		// Dedup is on the requested URL, so that is the key to assert on.
+		k := c.baseURL + c.modelsPath
+		if seen[k] {
+			t.Errorf("duplicate request for %s", k)
+		}
+		seen[k] = true
+	}
+	// The versioned spelling of the typed URL must be reachable: on the real
+	// amr-link endpoint that is what answers, and it is the preferred candidate.
+	if !seen["https://h/v1/models"] {
+		t.Error("the typed base with a versioned catalog path must be tried; that is what answers on amr-link")
+	}
+	// Both catalog forms must be attempted for a bare host too. Offering only
+	// "/v1/models" there left a vendor serving a bare "/models" looking
+	// unreachable.
+	if !seen["https://h/models"] {
+		t.Error("host root with a bare /models path was never tried")
+	}
+	// The first candidate is preferred, so its URL is what gets reported when
+	// several answer.
+	if cands[0].baseURL+cands[0].modelsPath != "https://h/v1/models" {
+		t.Errorf("first candidate requests %q, want the canonical versioned form",
+			cands[0].baseURL+cands[0].modelsPath)
+	}
+	// The bare "/models" form must exist too — some gateways serve it there.
+	hasBare := false
+	for _, c := range cands {
+		if c.modelsPath == "/models" {
+			hasBare = true
+		}
+	}
+	if !hasBare {
+		t.Error("no /models candidate; vendors serving the catalog at /models would be missed")
+	}
+	// Preference must be monotonically non-decreasing so the sort is stable.
+	for i := 1; i < len(cands); i++ {
+		if cands[i].prefer < cands[i-1].prefer {
+			t.Errorf("candidates out of preference order at %d: %d < %d", i, cands[i].prefer, cands[i-1].prefer)
+		}
+	}
+}
+
+func TestVersionSegment(t *testing.T) {
+	cases := map[string]string{
+		"/v1/models":   "/v1",
+		"/v1beta/models": "/v1beta",
+		"/models":      "",
+		"/api/models":  "",
+		"/vision/models": "",
+	}
+	for in, want := range cases {
+		if got := versionSegment(in); got != want {
+			t.Errorf("versionSegment(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+func TestTrimVersionSuffix(t *testing.T) {
+	cases := map[string]string{
+		"https://h/v1":          "https://h",
+		"https://h/v2":          "https://h",
+		"https://h/v1beta":      "https://h",
+		"https://h/api/v1":      "https://h/api",
+		"https://h/vision":      "https://h/vision", // not numeric — a real path segment
+		"https://h":             "https://h",
+		"https://h/openai":      "https://h/openai",
+		"https://h/v1/extra":    "https://h/v1/extra",
+		// A segment that merely STARTS with v must not be stripped: stripping
+		// it would rewrite base "https://h/vision" to "https://h" and, worse,
+		// would treat a real vendor path as a version.
+		"https://h/vertex":      "https://h/vertex",
+		"https://h/volcengine":  "https://h/volcengine",
+		"https://h/vertex-ai/v1": "https://h/vertex-ai",
+	}
+	for in, want := range cases {
+		if got := trimVersionSuffix(in); got != want {
+			t.Errorf("trimVersionSuffix(%q) = %q, want %q", in, got, want)
+		}
+	}
+}
+
+// TestIsVersionSegment_RequiresDigits pins the distinction the whole endpoint
+// derivation rests on: "v1"/"v1beta" are versions, "vision"/"vertex" are path
+// segments. Without the digit requirement, stripping turns a vendor's real
+// base URL into a different one.
+func TestIsVersionSegment_RequiresDigits(t *testing.T) {
+	yes := []string{"v1", "v2", "v10", "v1beta", "v1alpha", "v1preview"}
+	no := []string{"vision", "vertex", "volcengine", "openai", "v", "vv", "api", "beta", "v1x"}
+	for _, s := range yes {
+		if !isVersionSegment(s) {
+			t.Errorf("isVersionSegment(%q) = false, want true", s)
+		}
+	}
+	for _, s := range no {
+		if isVersionSegment(s) {
+			t.Errorf("isVersionSegment(%q) = true, want false", s)
+		}
+	}
+}
+
+// TestComposeEndpoint_ExactlyOneVersionSegment is the invariant test. Every row
+// below is a URL form that was observed to 404 or to route wrongly when the
+// version segment ended up in the wrong place.
+func TestComposeEndpoint_ExactlyOneVersionSegment(t *testing.T) {
+	cases := []struct {
+		name          string
+		modelsPath    string
+		persistedBase string
+		typedBase     string
+		want          string
+	}{
+		{
+			name:       "typed base has /v1 and catalog served there: base keeps it, endpoint empty",
+			modelsPath: "/models", persistedBase: "https://h/v1", typedBase: "https://h/v1",
+			want: "",
+		},
+		{
+			name:       "catalog served at versioned path, base stripped: endpoint carries it",
+			modelsPath: "/v1/models", persistedBase: "https://h", typedBase: "https://h/v1",
+			want: "/v1/chat/completions",
+		},
+		{
+			name:       "bare host, bare catalog, nothing typed: adapter default composes",
+			modelsPath: "/models", persistedBase: "https://h", typedBase: "https://h",
+			want: "",
+		},
+		{
+			name:       "bare host, versioned catalog: endpoint owns the version",
+			modelsPath: "/v1/models", persistedBase: "https://h", typedBase: "https://h",
+			want: "/v1/chat/completions",
+		},
+		{
+			name:       "typed /v1 but catalog only at bare /models: strip base, restore on endpoint",
+			modelsPath: "/models", persistedBase: "https://h", typedBase: "https://h/v1",
+			want: "/v1/chat/completions",
+		},
+		{
+			name:       "non-version segment must never be promoted onto the endpoint",
+			modelsPath: "/models", persistedBase: "https://h/vision", typedBase: "https://h/vision",
+			want: "",
+		},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			got := composeEndpoint(tc.modelsPath, tc.persistedBase, tc.typedBase)
+			if got != tc.want {
+				t.Errorf("composeEndpoint = %q, want %q", got, tc.want)
+			}
+			// Whichever way it resolved, base + endpoint must never carry the
+			// same version segment twice.
+			onBase := trimVersionSuffix(tc.persistedBase) != tc.persistedBase
+			onEndpoint := versionSegment(strings.TrimSuffix(got, "/chat/completions") + "/models") != ""
+			if onBase && onEndpoint {
+				t.Errorf("version segment duplicated: base %q + endpoint %q", tc.persistedBase, got)
+			}
+		})
+	}
+}
+
+// TestDiscoverModels_DoesNotProbeEveryModel guards the decision not to spend
+// money verifying each catalog entry. Only the catalog request is made.
+func TestDiscoverModels_DoesNotProbeEveryModel(t *testing.T) {
+	var calls int
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		calls++
+		_, _ = w.Write([]byte(`{"data":[{"id":"a"},{"id":"b"},{"id":"c"}]}`))
+	}))
+	defer srv.Close()
+
+	if _, err := DiscoverModels(context.Background(), srv.URL+"/v1", "k"); err != nil {
+		t.Fatalf("discovery failed: %v", err)
+	}
+	if calls != 1 {
+		t.Errorf("made %d requests, want exactly 1 (the catalog)", calls)
+	}
+}
+
+func TestModelsToConfig_PrioritiesAndKinds(t *testing.T) {
+	got := modelsToConfig([]string{"a", "b"}, []string{"img1"})
+	if len(got) != 3 {
+		t.Fatalf("want 3 models, got %d", len(got))
+	}
+	if got[2].ID != "img1" || got[2].Kind != "image" {
+		t.Errorf("image model mis-sorted: %+v", got[2])
+	}
+	if got[2].Priority <= got[0].Priority {
+		t.Errorf("image model priority %d should sort after chat %d", got[2].Priority, got[0].Priority)
+	}
+}
+
+// TestDiscoverModels_ImageModelsNeverNull guards the UI contract end to end,
+// not just on the struct: the WebUI does out.concat(image_models), and a JSON
+// null there renders "undefined" next to every image model row. This asserts on
+// the served bytes rather than the Go field, because that is what the bug was.
+func TestDiscoverModels_ImageModelsNeverNull(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		// Only chat models — the case where the slice is never appended to.
+		_, _ = w.Write([]byte(`{"data":[{"id":"a"},{"id":"b"}]}`))
+	}))
+	defer srv.Close()
+
+	res, err := DiscoverModels(context.Background(), srv.URL+"/v1", "k")
+	if err != nil {
+		t.Fatalf("discovery failed: %v", err)
+	}
+	if len(res.ImageModels) != 0 {
+		t.Fatalf("precondition: want zero image models, got %d", len(res.ImageModels))
+	}
+	b, err := json.Marshal(res)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(b), `"image_models":[]`) {
+		t.Errorf("served %s\nwant \"image_models\":[] — null breaks concat in the UI", b)
+	}
+}
+
+// TestComposeImageEndpoint pins the rule that fixed a real 404.
+//
+// Provider.ImageURL is `base_url + ImageEndpoint()`, and the built-in default
+// is the ABSOLUTE "/v1/images/generations" — while chat's default is the
+// relative "/chat/completions". The two defaults disagree about who owns the
+// version segment, so a base_url that produces a working chat URL produces
+// "/v1/v1/images/generations" and a 404.
+func TestComposeImageEndpoint(t *testing.T) {
+	cases := []struct {
+		base string
+		want string
+	}{
+		// base without a version: the absolute default composes correctly.
+		{"https://h", ""},
+		// base WITH a version: the default would double it, so the
+		// version-relative form is written instead.
+		{"https://h/v1", "/images/generations"},
+		{"https://h/v1beta", "/images/generations"},
+		// A real path segment that merely starts with v is not a version.
+		{"https://h/vision", ""},
+	}
+	for _, tc := range cases {
+		got := composeImageEndpoint(tc.base)
+		if got != tc.want {
+			t.Errorf("composeImageEndpoint(%q) = %q, want %q", tc.base, got, tc.want)
+		}
+		// Whatever it decided, base + result must not contain a doubled
+		// version segment — that is the bug this function exists for.
+		joined := tc.base + got
+		if strings.Contains(joined, "/v1/v1") {
+			t.Errorf("composed %q contains a doubled version segment", joined)
+		}
+	}
+}
+
+// TestDiscoveryReportsImageEndpointWhenBaseCarriesVersion is the end-to-end
+// version: a pasted base_url with "/v1" must come back with an image endpoint,
+// or the image models it reported will all 404 on use.
+func TestDiscoveryReportsImageEndpointWhenBaseCarriesVersion(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path != "/v1/models" {
+			w.WriteHeader(http.StatusNotFound)
+			return
+		}
+		_, _ = w.Write([]byte(realCatalogBody))
+	}))
+	defer srv.Close()
+
+	res, err := DiscoverModels(context.Background(), srv.URL+"/v1", "k")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(res.ImageModels) == 0 {
+		t.Fatal("fixture must contain image models for this to mean anything")
+	}
+	if res.BaseURL != srv.URL+"/v1" {
+		t.Fatalf("base = %q, want the typed URL with /v1", res.BaseURL)
+	}
+	if res.ImageEndpoint != "/images/generations" {
+		t.Errorf("image_endpoint = %q, want %q (otherwise the reported image models all 404)",
+			res.ImageEndpoint, "/images/generations")
+	}
+	// The chat side stays empty: base already carries /v1 and the adapter's
+	// relative default composes.
+	if res.Endpoint != "" {
+		t.Errorf("endpoint = %q, want empty", res.Endpoint)
+	}
+}
+
+// TestDiscoveryResult_NilSlicesSerializeAsArrays covers the zero value of
+// discoveryResult, which is what a caller gets if it builds one directly
+// instead of going through DiscoverModels. DiscoverModels happens to
+// initialize the slices, so testing only that path would let the MarshalJSON
+// nil guard look redundant — and removing it would ship a latent UI bug the
+// first time some other caller builds the struct.
+func TestDiscoveryResult_NilSlicesSerializeAsArrays(t *testing.T) {
+	b, err := json.Marshal(&discoveryResult{BaseURL: "https://h"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	got := string(b)
+	for _, want := range []string{`"models":[]`, `"image_models":[]`, `"tried":[]`} {
+		if !strings.Contains(got, want) {
+			t.Errorf("zero value serialized %s\nwant %s — a nil slice becomes JSON null and the UI renders \"undefined\"", got, want)
+		}
+	}
+	for _, bad := range []string{`"models":null`, `"image_models":null`, `"tried":null`} {
+		if strings.Contains(got, bad) {
+			t.Errorf("serialized %s, which the UI cannot concatenate", bad)
+		}
+	}
+}
+
+// TestDiscoveryResultJSONShape pins the field names the WebUI will bind to.
+func TestDiscoveryResultJSONShape(t *testing.T) {
+	b, err := json.Marshal(&discoveryResult{
+		BaseURL: "https://h", Models: []string{"a"}, ImageModels: []string{},
+		ModelsPath: "/models", Adapter: "openai",
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	var got map[string]interface{}
+	if err := json.Unmarshal(b, &got); err != nil {
+		t.Fatal(err)
+	}
+	// image_models must serialize as [] and never null: the UI does
+	// out.concat(image_models) and null would render "undefined".
+	if v, ok := got["image_models"]; !ok || v == nil {
+		t.Errorf("image_models = %v, want an empty array (null breaks concat in the UI)", v)
+	}
+	for _, k := range []string{"base_url", "endpoint", "models", "image_models", "models_path", "tried"} {
+		if _, ok := got[k]; !ok {
+			t.Errorf("missing field %q in discovery response", k)
+		}
+	}
+}

internal/gateway/gateway_test.go

@@ -58,7 +58,7 @@ func newTestGateway(t *testing.T, srcs ...config.Source) *Gateway {
 		t.Fatalf("core: %v", err)
 	}
 	t.Cleanup(c.Close)
-	g, err := New(c, []string{"sk-test"})
+	g, err := New(c)
 	if err != nil {
 		t.Fatalf("gateway: %v", err)
 	}
@@ -670,8 +670,11 @@ func TestHasScopeModelWithSourcePrefix(t *testing.T) {
 	if g.hasScopeModel(scope, "deepseek-v4-flash-free-extra") {
 		t.Error("hasScopeModel returned true for unrelated model")
 	}
-	if !g.hasScopeModel([]config.ModelScope{{Model: "AUTO"}}, "zen:anything") {
-		t.Error("AUTO scope should allow any prefixed model")
+	if g.hasScopeModel([]config.ModelScope{{Model: "AUTO"}}, "zen:deepseek-v4-flash-free") {
+		t.Error("AUTO scope must NOT grant specific model ids even with prefix")
+	}
+	if g.hasScopeModel([]config.ModelScope{{Model: "AUTO"}}, "deepseek-v4-pro") {
+		t.Error("AUTO scope must NOT grant specific model ids")
 	}
 }
 

internal/gateway/gui_contract_test.go

@@ -0,0 +1,192 @@
+package gateway
+
+import (
+	"os"
+	"path/filepath"
+	"regexp"
+	"strings"
+	"testing"
+)
+
+// The Electron shell (cmd/gui) had NO tests at all, so the plugin panel went in
+// with references to CSS classes that do not exist (.tag, .sm) and to helper
+// functions that were never defined in that document (esc / escAttr). All of
+// it rendered as unstyled text and would have thrown a ReferenceError at click
+// time — and none of that is visible without opening the app.
+//
+// These tests are deliberately static. They do not launch Electron: what they
+// guard is the class of mistake that "looks fine until someone themes it",
+// which is exactly what a missing CSS class or a missing helper is.
+
+func guiFile(t *testing.T, rel string) string {
+	t.Helper()
+	// The tests live in internal/gateway, so walk up to the repo root.
+	p := filepath.Join("..", "..", rel)
+	b, err := os.ReadFile(p)
+	if err != nil {
+		t.Skipf("%s not readable: %v", rel, err)
+	}
+	return string(b)
+}
+
+// classUseRe finds class="..." occurrences in a document.
+var classUseRe = regexp.MustCompile(`class="([^"]+)"`)
+
+// classTokenRe matches ANY ".name" inside the stylesheet. Deliberately loose:
+// it also matches inside compound selectors (".tb-btn.tb-close:hover" must count
+// as defining .tb-close, which a "must be at the start of a selector" rule
+// misses) and inside comments, which only ever makes the check MORE permissive.
+// A false pass here would be bad, so the strictness lives elsewhere: the
+// variable below is what actually guards the new code.
+var classTokenRe = regexp.MustCompile(`\.([A-Za-z_][A-Za-z0-9_-]*)`)
+
+// guiKnownUnstyled lists classes the shell markup has always used with no
+// matching rule. They are pre-existing cosmetic gaps, not regressions, and
+// failing on them would make this test useless as a guard for NEW work.
+var guiKnownUnstyled = map[string]bool{
+	"blob": true, // decorative blur blobs, styled per-instance via .b1/.b2/.b3
+	"tgl":  true, // rail toggle affordance that leaned on .rail-btn
+	"rail": true, // the rail container itself has no rule; .rail-btn children carry the look
+}
+
+// TestGUICSSClassesExist is the guard that would have caught .tag and .sm: every
+// class used in the shell's markup must be defined in its stylesheet.
+//
+// The comparison is on the LAST segment of a selector, because the stylesheet
+// scopes things (`.form .actions .primary`, `#bgfx .b1`): a rule for
+// `.pl-acts button` defines no class at all, and `.form .row .toggle` defines
+// `.toggle`. Requiring a top-level class would be too strict; requiring that
+// some selector's last identifier matches is the right level.
+func TestGUICSSClassesExist(t *testing.T) {
+	html := guiFile(t, "cmd/gui/renderer/index.html")
+	css := guiFile(t, "cmd/gui/renderer/style.css")
+
+	defined := map[string]bool{}
+
+	// Collect every class token that appears at the START of a selector
+	// position. A full CSS parser is overkill and was the source of two wrong
+	// turns here; what the check needs is simply "does the name .foo appear
+	// anywhere in the stylesheet as a selector component".
+	//
+	// Scoping is respected loosely: `.form .actions .primary` counts as
+	// defining `.primary`, and `.pl-acts button` defines no class — which is
+	// exactly why the plugin panel needed its own rules.
+	for _, m := range classTokenRe.FindAllStringSubmatch(css, -1) {
+		defined[m[1]] = true
+	}
+	if len(defined) == 0 {
+		t.Fatal("no classes parsed from the stylesheet; the check is broken")
+	}
+
+	// Classes the JS builds as strings must exist too.
+	js := guiFile(t, "cmd/gui/renderer/app.js")
+	var missing []string
+	seen := map[string]bool{}
+	note := func(cls, where string) {
+		// A "${...}" token is a template literal being spliced at runtime, not
+		// a class name; the classes it can expand to are checked at their
+		// definition sites instead.
+		if cls == "" || strings.ContainsAny(cls, "${}") || seen[cls] {
+			return
+		}
+		seen[cls] = true
+		if guiKnownUnstyled[cls] {
+			return
+		}
+		if !defined[cls] {
+			missing = append(missing, cls+" ("+where+")")
+		}
+	}
+	for _, m := range classUseRe.FindAllStringSubmatch(html, -1) {
+		for _, c := range strings.Fields(m[1]) {
+			note(c, "index.html")
+		}
+	}
+	for _, m := range classUseRe.FindAllStringSubmatch(js, -1) {
+		for _, c := range strings.Fields(m[1]) {
+			note(c, "app.js")
+		}
+	}
+	if len(missing) > 0 {
+		t.Errorf("classes used but not defined in style.css (they render unstyled):\n  %s",
+			strings.Join(missing, "\n  "))
+	}
+}
+
+// TestGUIHelperFunctionsAreDefined catches the other half: renderer/app.js is a
+// separate document from the WebUI, so it does NOT have the WebUI's esc/escAttr.
+// Referencing them gives a ReferenceError only when the line runs.
+func TestGUIHelperFunctionsAreDefined(t *testing.T) {
+	js := guiFile(t, "cmd/gui/renderer/app.js")
+	for _, fn := range []string{"esc", "escAttr", "toast", "loadPlugins", "togglePlugin", "enableAllPlugins"} {
+		defined := regexp.MustCompile(`function ` + fn + `\b`).MatchString(js)
+		called := regexp.MustCompile(`\b` + fn + `\s*\(`).MatchString(js)
+		if called && !defined {
+			t.Errorf("%s() is called but never defined in app.js", fn)
+		}
+		if !called && !defined {
+			// A defined-but-unused helper is dead code, not an error.
+			continue
+		}
+	}
+}
+
+// TestGUIPluginPanelIsReachable: the panel must be inside the settings overlay
+// AND the settings overlay must actually open it. A panel wired to a button
+// that was never bound is invisible-but-present, which passes a grep review.
+func TestGUIPluginPanelIsReachable(t *testing.T) {
+	html := guiFile(t, "cmd/gui/renderer/index.html")
+	js := guiFile(t, "cmd/gui/renderer/app.js")
+	if !strings.Contains(html, `id="pl-list"`) {
+		t.Error("no #pl-list in the settings overlay")
+	}
+	if !strings.Contains(html, `id="settings-overlay"`) {
+		t.Fatal("the settings overlay is gone")
+	}
+	// inside the overlay: the element index must come after the overlay's
+	if strings.Index(html, `id="settings-overlay"`) > strings.Index(html, `id="pl-list"`) {
+		t.Error("#pl-list appears before the settings overlay, so it renders outside the panel")
+	}
+	// The buttons must be bound.
+	for _, id := range []string{"pl-reload", "pl-toggle-all"} {
+		if !strings.Contains(html, `id="`+id+`"`) {
+			t.Errorf("#%s is missing from the markup", id)
+		}
+		if !strings.Contains(js, `"`+id+`"`) {
+			t.Errorf("#%s exists but app.js never binds it", id)
+		}
+	}
+	// And openSettings must trigger the load, or the panel shows a stale empty
+	// list on every open.
+	if !strings.Contains(js, "loadPlugins()") {
+		t.Error("app.js never calls loadPlugins()")
+	}
+}
+
+// TestGUIIPCPathIsConstrained: plugins:proxy is a raw pass-through, which is
+// convenient but would be a hole if it accepted arbitrary paths. The main
+// process must reject anything outside /api/plugins and any traversal.
+func TestGUIIPCPathIsConstrained(t *testing.T) {
+	main := guiFile(t, "cmd/gui/main.js")
+	for _, needle := range []string{
+		`path.startsWith("/api/plugins")`,
+		`path.includes("..")`,
+		"plugins:proxy",
+		"unsealViaCore()",
+	} {
+		if !strings.Contains(main, needle) {
+			t.Errorf("cmd/gui/main.js is missing the guard %q", needle)
+		}
+	}
+	// The plugins channel must be a proxy, not a key passthrough: the renderer
+	// sends (method, path) and the main process attaches the key.
+	//
+	// NOTE: preload does expose a pre-existing `core.key` getter — the shell
+	// needs the admin key to load the embedded WebUI without a login. That is
+	// existing, deliberate design and out of scope here; asserting on "key:" in
+	// preload would flag a pre-existing feature as a new hole.
+	pre := guiFile(t, "cmd/gui/preload.js")
+	if !strings.Contains(pre, "request: (method, path, body)") {
+		t.Error("preload does not expose the plugins request proxy")
+	}
+}

internal/gateway/key_auto_api_test.go

@@ -0,0 +1,257 @@
+package gateway
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"path/filepath"
+	"os"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+	"llmsproxy/internal/core"
+)
+
+// Per-key AUTO chain admin API. The behaviours pinned here are the ones that
+// are easy to get wrong and invisible when they are:
+//
+//   - /api/keys/me/auto must be reachable by a NON-admin (it is their own
+//     chain); routing it through the generic {key}/auto handler would 403 it.
+//   - /api/keys/{key}/auto must 403 a non-admin.
+//   - an empty PUT and a DELETE mean the same thing ("inherit global"), so the
+//     WebUI cannot persist a chain that would then 503 with no slots.
+//   - a PUT that resolves to zero slots is reported back, so an admin who
+//     mistyped a model learns it now instead of from the user's next 503.
+
+// keyAutoGateway builds a gateway with one admin key and one user key.
+func keyAutoGateway(t *testing.T) *Gateway {
+	t.Helper()
+	td := t.TempDir()
+	cfgPath := filepath.Join(td, "config.yaml")
+	if err := os.WriteFile(cfgPath, []byte("listen: :0"), 0644); err != nil {
+		t.Fatal(err)
+	}
+	cfg := &config.Config{
+		Path:        cfgPath,
+		AdapterDir:  filepath.Join(td, "adapters"),
+		RuntimeFile: filepath.Join(td, "runtime.json"),
+		DefaultModel: "AUTO",
+		GatewayKeys: []string{"sk-admin"},
+		Keys: []config.GWKey{
+			{Key: "sk-admin", Role: "admin", Name: "admin"},
+			{Key: "sk-user", Role: "user", Name: "user"},
+		},
+		Sources: []config.Source{
+			{Name: "s1", BaseURL: "http://127.0.0.1:1/v1", Adapter: "openai",
+				Models: []config.Model{{ID: "gpt-4o", Priority: 10}, {ID: "gpt-4o-mini", Priority: 5}}},
+		},
+	}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(c.Close)
+	g, err := New(c)
+	if err != nil {
+		t.Fatalf("gateway: %v", err)
+	}
+	return g
+}
+
+// doReqAs issues an authenticated request as a specific key.
+func doReqAs(t *testing.T, g *Gateway, key, method, path, body string) *httptest.ResponseRecorder {
+	t.Helper()
+	req, _ := http.NewRequest(method, path, strings.NewReader(body))
+	req.Header.Set("Authorization", "Bearer "+key)
+	if body != "" {
+		req.Header.Set("Content-Type", "application/json")
+	}
+	rr := httptest.NewRecorder()
+	g.Handler().ServeHTTP(rr, req)
+	return rr
+}
+
+func TestKeyAutoAPIPermissions(t *testing.T) {
+	g := keyAutoGateway(t)
+
+	// A user may read their own chain...
+	if rr := doReqAs(t, g, "sk-user", "GET", "/api/keys/me/auto", ""); rr.Code != http.StatusOK {
+		t.Fatalf("GET /api/keys/me/auto as user = %d, want 200 (body %s)", rr.Code, rr.Body.String())
+	}
+	// ...but not another key's.
+	if rr := doReqAs(t, g, "sk-user", "GET", "/api/keys/sk-admin/auto", ""); rr.Code != http.StatusForbidden {
+		t.Fatalf("GET other key's auto as user = %d, want 403", rr.Code)
+	}
+	// ...and not their own via the admin route either.
+	if rr := doReqAs(t, g, "sk-user", "PUT", "/api/keys/sk-user/auto", `{"auto":[{"model":"gpt-4o"}]}`); rr.Code != http.StatusForbidden {
+		t.Fatalf("user PUT own auto = %d, want 403 (only admin configures)", rr.Code)
+	}
+}
+
+func TestKeyAutoAPICrudAndInherit(t *testing.T) {
+	g := keyAutoGateway(t)
+
+	// Initially inherits.
+	var got struct {
+		Inherits bool               `json:"inherits"`
+		Auto     []config.ModelScope `json:"auto"`
+	}
+	rr := doReqAs(t, g, "sk-admin", "GET", "/api/keys/sk-user/auto", "")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("GET = %d: %s", rr.Code, rr.Body.String())
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if !got.Inherits || len(got.Auto) != 0 {
+		t.Fatalf("fresh key must inherit global, got inherits=%v auto=%v", got.Inherits, got.Auto)
+	}
+
+	// Admin sets a chain; the response reports the resolved slot count.
+	rr = doReqAs(t, g, "sk-admin", "PUT", "/api/keys/sk-user/auto",
+		`{"auto":[{"model":"gpt-4o-mini","source":"s1","tier":1}]}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("PUT = %d: %s", rr.Code, rr.Body.String())
+	}
+	var put struct {
+		Inherits bool `json:"inherits"`
+		Slots    int  `json:"slots"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &put); err != nil {
+		t.Fatal(err)
+	}
+	if put.Inherits {
+		t.Fatal("after a non-empty PUT the key must no longer inherit")
+	}
+	if put.Slots != 1 {
+		t.Fatalf("resolved slots = %d, want 1", put.Slots)
+	}
+
+	// And it reads back as an own chain.
+	rr = doReqAs(t, g, "sk-admin", "GET", "/api/keys/sk-user/auto", "")
+	if err := json.Unmarshal(rr.Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if got.Inherits || len(got.Auto) != 1 || got.Auto[0].Model != "gpt-4o-mini" {
+		t.Fatalf("own chain must read back, got inherits=%v auto=%v", got.Inherits, got.Auto)
+	}
+
+	// DELETE restores inheritance.
+	if rr := doReqAs(t, g, "sk-admin", "DELETE", "/api/keys/sk-user/auto", ""); rr.Code != http.StatusOK {
+		t.Fatalf("DELETE = %d: %s", rr.Code, rr.Body.String())
+	}
+	rr = doReqAs(t, g, "sk-admin", "GET", "/api/keys/sk-user/auto", "")
+	if err := json.Unmarshal(rr.Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if !got.Inherits {
+		t.Fatal("after DELETE the key must inherit the global chain again")
+	}
+}
+
+// An empty PUT must behave like DELETE. Otherwise a WebUI that submits an
+// empty list (every slot deleted in the editor) would persist an empty chain
+// and the next AUTO request would fail with no_provider.
+func TestKeyAutoAPIEmptyPutClearsOverride(t *testing.T) {
+	g := keyAutoGateway(t)
+
+	if rr := doReqAs(t, g, "sk-admin", "PUT", "/api/keys/sk-user/auto",
+		`{"auto":[{"model":"gpt-4o","source":"s1","tier":1}]}`); rr.Code != http.StatusOK {
+		t.Fatalf("seed PUT = %d: %s", rr.Code, rr.Body.String())
+	}
+	rr := doReqAs(t, g, "sk-admin", "PUT", "/api/keys/sk-user/auto", `{"auto":[]}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("empty PUT = %d: %s", rr.Code, rr.Body.String())
+	}
+	var got struct {
+		Inherits bool `json:"inherits"`
+	}
+	if rr := doReqAs(t, g, "sk-admin", "GET", "/api/keys/sk-user/auto", ""); rr.Code != http.StatusOK {
+		t.Fatal(rr.Body.String())
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if !got.Inherits {
+		t.Fatal("an empty PUT must clear the override, not persist an empty chain")
+	}
+}
+
+// A chain naming a model that does not exist compiles to zero slots. The API
+// must say so at write time instead of letting the user's next request 503.
+func TestKeyAutoAPIReportsUnresolvableSlots(t *testing.T) {
+	g := keyAutoGateway(t)
+
+	rr := doReqAs(t, g, "sk-admin", "PUT", "/api/keys/sk-user/auto",
+		`{"auto":[{"model":"does-not-exist","source":"s1","tier":1}]}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("PUT = %d: %s", rr.Code, rr.Body.String())
+	}
+	var put struct {
+		Inherits bool `json:"inherits"`
+		Slots    int  `json:"slots"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &put); err != nil {
+		t.Fatal(err)
+	}
+	if put.Slots != 0 {
+		t.Fatalf("unknown model must resolve to 0 slots, got %d", put.Slots)
+	}
+}
+
+func TestKeyAutoAPIUnknownKey404(t *testing.T) {
+	g := keyAutoGateway(t)
+
+	if rr := doReqAs(t, g, "sk-admin", "GET", "/api/keys/sk-nope/auto", ""); rr.Code != http.StatusNotFound {
+		t.Fatalf("GET unknown key auto = %d, want 404", rr.Code)
+	}
+	if rr := doReqAs(t, g, "sk-admin", "PUT", "/api/keys/sk-nope/auto", `{"auto":[{"model":"gpt-4o"}]}`); rr.Code != http.StatusNotFound {
+		t.Fatalf("PUT unknown key auto = %d, want 404", rr.Code)
+	}
+}
+
+// A bad quota must be rejected with 400, not persisted and not silently
+// clamped — same rule as the model scope path.
+func TestKeyAutoAPIRejectsBadQuota(t *testing.T) {
+	g := keyAutoGateway(t)
+
+	rr := doReqAs(t, g, "sk-admin", "PUT", "/api/keys/sk-user/auto",
+		`{"auto":[{"model":"gpt-4o","source":"s1","tier":1,"token_quota":-1}]}`)
+	if rr.Code != http.StatusBadRequest {
+		t.Fatalf("negative quota PUT = %d, want 400 (body %s)", rr.Code, rr.Body.String())
+	}
+	// Nothing may have been persisted.
+	var got struct {
+		Inherits bool `json:"inherits"`
+	}
+	get := doReqAs(t, g, "sk-admin", "GET", "/api/keys/sk-user/auto", "")
+	if get.Code != http.StatusOK {
+		t.Fatal(get.Body.String())
+	}
+	if err := json.Unmarshal(get.Body.Bytes(), &got); err != nil {
+		t.Fatal(err)
+	}
+	if !got.Inherits {
+		t.Fatal("a rejected PUT must not persist a chain")
+	}
+}
+
+// The other key endpoints must keep working: the {key}/auto route must not
+// shadow /api/keys/{key} itself.
+func TestKeyAutoAPIDoesNotShadowKeyRoutes(t *testing.T) {
+	g := keyAutoGateway(t)
+
+	if rr := doReqAs(t, g, "sk-admin", "GET", "/api/keys", ""); rr.Code != http.StatusOK {
+		t.Fatalf("GET /api/keys = %d, want 200", rr.Code)
+	}
+	if rr := doReqAs(t, g, "sk-admin", "GET", "/api/keys/me", ""); rr.Code != http.StatusOK {
+		t.Fatalf("GET /api/keys/me = %d, want 200", rr.Code)
+	}
+	if rr := doReqAs(t, g, "sk-user", "GET", "/api/keys/me", ""); rr.Code != http.StatusOK {
+		t.Fatalf("GET /api/keys/me as user = %d, want 200", rr.Code)
+	}
+}

internal/gateway/key_quota_api_test.go

@@ -0,0 +1,253 @@
+package gateway
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+	"llmsproxy/internal/core"
+)
+
+// adminGateway builds a gateway whose admin key can call /api/keys.
+func adminGateway(t *testing.T, keys ...config.GWKey) *Gateway {
+	t.Helper()
+	td := t.TempDir()
+	cfgPath := td + "/config.yaml"
+	if err := os.WriteFile(cfgPath, []byte("listen: :0"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	cfg := &config.Config{
+		Path:        cfgPath,
+		AdapterDir:  td + "/adapters",
+		RuntimeFile: td + "/runtime.json",
+		Keys:        keys,
+	}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(c.Close)
+	g, err := New(c)
+	if err != nil {
+		t.Fatalf("gateway: %v", err)
+	}
+	return g
+}
+
+func adminReq(t *testing.T, g *Gateway, method, path, body string) *httptest.ResponseRecorder {
+	t.Helper()
+	req, _ := http.NewRequest(method, path, strings.NewReader(body))
+	req.Header.Set("Authorization", "Bearer sk-admin")
+	req.Header.Set("Content-Type", "application/json")
+	rr := httptest.NewRecorder()
+	g.Handler().ServeHTTP(rr, req)
+	return rr
+}
+
+// keyRecord pulls one key's stored record out of the admin list.
+func keyRecord(t *testing.T, g *Gateway, secret string) config.GWKey {
+	t.Helper()
+	rr := adminReq(t, g, "GET", "/api/keys", "")
+	if rr.Code != 200 {
+		t.Fatalf("GET /api/keys: %d %s", rr.Code, rr.Body.String())
+	}
+	var out struct {
+		Keys []config.GWKey `json:"keys"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &out); err != nil {
+		t.Fatalf("decode: %v (%s)", err, rr.Body.String())
+	}
+	for _, k := range out.Keys {
+		if k.Key == secret {
+			return k
+		}
+	}
+	t.Fatalf("key %q not found in %s", secret, rr.Body.String())
+	return config.GWKey{}
+}
+
+func scopeOf(t *testing.T, k config.GWKey, model string) config.ModelScope {
+	t.Helper()
+	for _, m := range k.Models {
+		if m.Model == model {
+			return m
+		}
+	}
+	t.Fatalf("scope %q not found in %+v", model, k.Models)
+	return config.ModelScope{}
+}
+
+// Quotas live on the scope entries, not on the key: creating a key with a
+// budget means creating scopes that carry it, and they must survive a
+// read-back (persisted, not just echoed).
+func TestKeyAPICreatesPerModelQuota(t *testing.T) {
+	g := adminGateway(t, config.GWKey{Key: "sk-admin", Role: "admin"})
+	rr := adminReq(t, g, "POST", "/api/keys",
+		`{"name":"agent-x","role":"user","models":[{"model":"m1","token_quota":50000,"req_quota":200,"period":"nhour","hours":6}]}`)
+	if rr.Code != 200 {
+		t.Fatalf("create: %d %s", rr.Code, rr.Body.String())
+	}
+	var created struct {
+		Key config.GWKey `json:"key"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &created); err != nil {
+		t.Fatalf("decode: %v", err)
+	}
+	sc := scopeOf(t, created.Key, "m1")
+	if sc.TokenQuota != 50000 || sc.ReqQuota != 200 || sc.Period != "nhour" || sc.Hours != 6 {
+		t.Fatalf("created scope did not carry the caps: %+v", sc)
+	}
+	back := scopeOf(t, keyRecord(t, g, created.Key.Key), "m1")
+	if back.TokenQuota != 50000 || back.ReqQuota != 200 || back.Period != "nhour" || back.Hours != 6 {
+		t.Errorf("read-back lost the caps: %+v", back)
+	}
+}
+
+// Two models on one key carry independent budgets.
+func TestKeyAPIKeepsPerModelQuotaIndependent(t *testing.T) {
+	g := adminGateway(t, config.GWKey{Key: "sk-admin", Role: "admin"})
+	rr := adminReq(t, g, "POST", "/api/keys",
+		`{"name":"agent-y","role":"user","models":[{"model":"m1","token_quota":1000,"period":"hour"},{"model":"m2","token_quota":9999,"req_quota":7,"period":"week"}]}`)
+	if rr.Code != 200 {
+		t.Fatalf("create: %d %s", rr.Code, rr.Body.String())
+	}
+	var created struct {
+		Key config.GWKey `json:"key"`
+	}
+	_ = json.Unmarshal(rr.Body.Bytes(), &created)
+	back := keyRecord(t, g, created.Key.Key)
+	if a := scopeOf(t, back, "m1"); a.TokenQuota != 1000 || a.ReqQuota != 0 || a.Period != "hour" {
+		t.Errorf("m1 caps wrong: %+v", a)
+	}
+	if b := scopeOf(t, back, "m2"); b.TokenQuota != 9999 || b.ReqQuota != 7 || b.Period != "week" {
+		t.Errorf("m2 caps wrong: %+v", b)
+	}
+}
+
+// Sending 0 explicitly lifts that model's cap.
+func TestKeyAPIUpdateZeroLiftsCap(t *testing.T) {
+	g := adminGateway(t, config.GWKey{Key: "sk-admin", Role: "admin"})
+	rr := adminReq(t, g, "POST", "/api/keys",
+		`{"name":"z","role":"user","models":[{"model":"m1","token_quota":1000,"period":"hour"}]}`)
+	if rr.Code != 200 {
+		t.Fatalf("create: %d %s", rr.Code, rr.Body.String())
+	}
+	var created struct {
+		Key config.GWKey `json:"key"`
+	}
+	_ = json.Unmarshal(rr.Body.Bytes(), &created)
+	secret := created.Key.Key
+
+	rr = adminReq(t, g, "PUT", "/api/keys/"+secret,
+		`{"models":[{"model":"m1","token_quota":0,"req_quota":0,"period":""}]}`)
+	if rr.Code != 200 {
+		t.Fatalf("lift: %d %s", rr.Code, rr.Body.String())
+	}
+	if back := scopeOf(t, keyRecord(t, g, secret), "m1"); back.TokenQuota != 0 || back.ReqQuota != 0 || back.Period != "" {
+		t.Errorf("caps not lifted: %+v", back)
+	}
+}
+
+// A misspelled period must be refused, not quietly turned into an all-time
+// quota — which is the exact opposite of what the operator typed.
+func TestKeyAPIRejectsBadPeriod(t *testing.T) {
+	g := adminGateway(t, config.GWKey{Key: "sk-admin", Role: "admin"})
+	rr := adminReq(t, g, "POST", "/api/keys",
+		`{"name":"bad","role":"user","models":[{"model":"m1","token_quota":1000,"period":"houre"}]}`)
+	if rr.Code != http.StatusBadRequest {
+		t.Fatalf("want 400 for a bad period, got %d %s", rr.Code, rr.Body.String())
+	}
+	if !strings.Contains(rr.Body.String(), "period") {
+		t.Errorf("error should name the period field: %s", rr.Body.String())
+	}
+}
+
+func TestKeyAPIRejectsNegativeQuota(t *testing.T) {
+	g := adminGateway(t, config.GWKey{Key: "sk-admin", Role: "admin"})
+	rr := adminReq(t, g, "POST", "/api/keys",
+		`{"name":"bad","role":"user","models":[{"model":"m1","token_quota":-5}]}`)
+	if rr.Code != http.StatusBadRequest {
+		t.Fatalf("want 400 for a negative quota, got %d %s", rr.Code, rr.Body.String())
+	}
+}
+
+// The rejection must say WHICH model is over budget, so an operator looking at
+// a key with a dozen scopes can tell which one to raise.
+func TestKeyAPIRejectionNamesTheModel(t *testing.T) {
+	g := adminGateway(t, config.GWKey{Key: "sk-admin", Role: "admin"})
+	rr := adminReq(t, g, "POST", "/api/keys",
+		`{"name":"bad","role":"user","models":[{"model":"m1","req_quota":-1}]}`)
+	if rr.Code != http.StatusBadRequest {
+		t.Fatalf("want 400, got %d", rr.Code)
+	}
+	if !strings.Contains(rr.Body.String(), "m1") {
+		t.Errorf("error should name the offending model: %s", rr.Body.String())
+	}
+}
+
+// A non-admin key must not be able to mint keys.
+func TestKeyAPIQuotaIsAdminOnly(t *testing.T) {
+	g := adminGateway(t,
+		config.GWKey{Key: "sk-admin", Role: "admin"},
+		config.GWKey{Key: "sk-u", Role: "user", Models: []config.ModelScope{{Model: "m1", TokenQuota: 10, Period: "hour"}}},
+	)
+	req, _ := http.NewRequest("POST", "/api/keys",
+		strings.NewReader(`{"name":"x","role":"admin","models":[{"model":"m1"}]}`))
+	req.Header.Set("Authorization", "Bearer sk-u")
+	req.Header.Set("Content-Type", "application/json")
+	rr := httptest.NewRecorder()
+	g.Handler().ServeHTTP(rr, req)
+	if rr.Code != http.StatusForbidden {
+		t.Fatalf("non-admin create: want 403, got %d %s", rr.Code, rr.Body.String())
+	}
+	// /api/v1/keys exposes the per-model caps but never a secret
+	rr = adminReq(t, g, "GET", "/api/v1/keys", "")
+	if rr.Code != 200 {
+		t.Fatalf("GET /api/v1/keys: %d", rr.Code)
+	}
+	if strings.Contains(rr.Body.String(), "sk-u") {
+		t.Error("/api/v1/keys leaked a key secret")
+	}
+	if !strings.Contains(rr.Body.String(), `"token_quota":10`) {
+		t.Errorf("/api/v1/keys should expose the per-model caps: %s", rr.Body.String())
+	}
+}
+
+// A user must be able to see their own budget: /api/keys/me is the only key
+// view a non-admin gets, so a cap missing from it is invisible to the very
+// client it constrains.
+func TestKeyMeExposesOwnQuota(t *testing.T) {
+	g := adminGateway(t,
+		config.GWKey{Key: "sk-admin", Role: "admin"},
+		config.GWKey{Key: "sk-u", Role: "user", Name: "agent",
+			Models: []config.ModelScope{{Model: "m1", TokenQuota: 123456, ReqQuota: 42, Period: "week"}}},
+	)
+	req, _ := http.NewRequest("GET", "/api/keys/me", nil)
+	req.Header.Set("Authorization", "Bearer sk-u")
+	rr := httptest.NewRecorder()
+	g.Handler().ServeHTTP(rr, req)
+	if rr.Code != 200 {
+		t.Fatalf("GET /api/keys/me: %d %s", rr.Code, rr.Body.String())
+	}
+	// the endpoint wraps the record: {"key": {...}}
+	var wrap struct {
+		Key config.GWKey `json:"key"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &wrap); err != nil {
+		t.Fatalf("decode: %v (%s)", err, rr.Body.String())
+	}
+	sc := scopeOf(t, wrap.Key, "m1")
+	if sc.TokenQuota != 123456 || sc.ReqQuota != 42 || sc.Period != "week" {
+		t.Errorf("own quota not visible to the key's owner: %+v", sc)
+	}
+}
+
+var _ = fmt.Sprintf

internal/gateway/key_quota_concurrency_test.go

@@ -0,0 +1,60 @@
+package gateway
+
+import (
+	"sync"
+	"testing"
+	"time"
+)
+
+// The per-key quota buckets are shared maps mutated on every recorded request
+// and read on every quota check. Single-threaded unit tests cannot see a race
+// there at all, so this drives 64 goroutines through record + read together
+// under `go test -race`. It also opts one model into the lazily-created pinned
+// bucket mid-flight, which is the path that mutates two bucket maps at once.
+func TestConcurrentQuotaAccounting(t *testing.T) {
+	s := NewStats(0)
+	keys := []string{"k0", "k1", "k2", "k3", "k4", "k5", "k6"}
+	models := []string{"m0", "m1", "m2", "AUTO"}
+	var wg sync.WaitGroup
+	stop := time.Now().Add(2 * time.Second)
+
+	for w := 0; w < 64; w++ {
+		wg.Add(1)
+		go func(w int) {
+			defer wg.Done()
+			k := keys[w%len(keys)]
+			m := models[w%len(models)]
+			now := time.Now()
+			for time.Now().Before(stop) {
+				s.Record(Req{Time: now.UnixMilli(), Key: k, Model: m, Source: "src",
+					Prompt: 100, Compl: 50, OK: true, Status: 200, Type: "chat"})
+				_ = s.KeyWindowTokens(k, 3600)
+				_ = s.KeyWindowReqs(k, 3600)
+				_ = s.KeyWindowModelTokens(k, m, "", 3600)
+				_ = s.KeyWindowModelTokens(k, m, "src", 3600) // opts into the pinned bucket
+			}
+		}(w)
+	}
+	wg.Wait()
+
+	// Every recorded request must be accounted for exactly once.
+	s.mu.Lock()
+	var total int64
+	for _, hm := range s.keyHour {
+		for _, v := range hm {
+			total += v
+		}
+	}
+	recs := int64(0)
+	for _, r := range s.recs {
+		recs += r.Prompt + r.Compl
+	}
+	s.mu.Unlock()
+	if total == 0 {
+		t.Fatal("no tokens accounted")
+	}
+	t.Logf("accounted %d tokens across %d keys", total, len(keys))
+	if recs == 0 {
+		t.Error("no records retained; the concurrent writers lost work")
+	}
+}

internal/gateway/key_quota_perf_test.go

@@ -0,0 +1,125 @@
+package gateway
+
+import (
+	"math/rand"
+	"testing"
+	"time"
+)
+
+// Perf: the per-key quota bookkeeping runs on every recorded request and on
+// every quota check, so its cost lands directly on the request path. These
+// benchmarks exist to catch a regression that would make the feature
+// expensive; the numbers that matter are relative to each other and to the
+// pre-change baseline (Record was 275 ns/op with 3 allocs before this work).
+
+func BenchmarkRecordQuotaBucket(b *testing.B) {
+	s := NewStats(0)
+	r := Req{Time: time.Now().UnixMilli(), Key: "keyABC", Model: "deepseek-v4-flash",
+		Source: "deepseek", Prompt: 800, Compl: 200, OK: true, Status: 200, Type: "chat"}
+	b.ReportAllocs()
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		s.Record(r)
+	}
+}
+
+// A key that has been busy for the whole 40-day retention.
+func BenchmarkRecordQuotaBucketFullRetention(b *testing.B) {
+	s := NewStats(0)
+	nowH := time.Now().Unix() / 3600
+	for h := int64(0); h < quotaRetentionHours; h++ {
+		s.Record(Req{Time: (nowH - h) * 3600 * 1000, Key: "keyABC", Model: "m1",
+			Source: "deepseek", Prompt: 100, Compl: 100, OK: true, Status: 200})
+	}
+	r := Req{Time: time.Now().UnixMilli(), Key: "keyABC", Model: "m1", Source: "deepseek",
+		Prompt: 100, Compl: 100, OK: true, Status: 200, Type: "chat"}
+	b.ReportAllocs()
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		s.Record(r)
+	}
+}
+
+// The whole per-request quota check: total tokens + request count.
+func BenchmarkQuotaCheckHourly(b *testing.B) {
+	s := NewStats(0)
+	for h := 0; h < 24; h++ {
+		s.Record(Req{Time: time.Now().Add(-time.Duration(h) * time.Hour).UnixMilli(),
+			Key: "keyABC", Model: "m1", Source: "deepseek", Prompt: 100, Compl: 100, OK: true, Status: 200})
+	}
+	b.ReportAllocs()
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		s.KeyWindowTokens("keyABC", 3600)
+		s.KeyWindowReqs("keyABC", 3600)
+	}
+}
+
+// A 30-day window has to visit 720 hour buckets. This is the worst case the
+// quota feature puts on the request path.
+func BenchmarkQuotaCheckMonthly(b *testing.B) {
+	s := NewStats(0)
+	nowH := time.Now().Unix() / 3600
+	for h := int64(0); h < 720; h++ {
+		s.Record(Req{Time: (nowH - h) * 3600 * 1000, Key: "keyABC", Model: "m1",
+			Prompt: 100, Compl: 100, OK: true, Status: 200})
+	}
+	b.ReportAllocs()
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		s.KeyWindowTokens("keyABC", 30*24*3600)
+	}
+}
+
+// TestSumBucketsMatchesFullScan is the correctness guard for the optimization
+// that made the quota check proportional to the window instead of to the whole
+// retention: the windowed sum must agree with a full scan for every bucket
+// placement, or a quota would silently start letting traffic through. It
+// caught a real off-by-one (floor instead of ceil on the first hour).
+func TestSumBucketsMatchesFullScan(t *testing.T) {
+	const hour = 3600
+	base := int64(1_700_000_000) / hour * hour
+	rng := rand.New(rand.NewSource(7))
+	for trial := 0; trial < 400; trial++ {
+		hm := map[int64]int64{}
+		for i, n := 0, rng.Intn(40)+1; i < n; i++ {
+			h := base/hour - int64(rng.Intn(1200))
+			hm[h] += int64(rng.Intn(1000)) + 1
+		}
+		now := base + int64(rng.Intn(hour))
+		for _, sec := range []int64{3600, 2 * 3600, 6 * 3600, 24 * 3600, 7 * 24 * 3600, 30 * 24 * 3600} {
+			var want int64
+			cut := now - sec
+			for h, v := range hm {
+				if h*hour >= cut {
+					want += v
+				}
+			}
+			if got := sumBuckets(hm, now, sec); got != want {
+				t.Fatalf("trial %d sec=%d: sumBuckets=%d, full scan=%d", trial, sec, got, want)
+			}
+		}
+	}
+}
+
+// The pinned bucket is maintained only for (key, model) pairs whose quota pins
+// a source. Doing it unconditionally doubled the bucket count — measured at
+// 26.7 MB for 20 keys x 8 models x 3 sources at full retention, against a
+// documented ~32 MB total memory budget — to serve a lookup nobody performs.
+func TestPinnedBucketNotMaintainedUnlessWanted(t *testing.T) {
+	s := NewStats(100)
+	now := time.Now().UnixMilli()
+	for i := 0; i < 5; i++ {
+		s.Record(Req{Time: now, Key: "keyA", Model: "m1", Source: "srcX", Prompt: 100, Compl: 100, OK: true, Status: 200})
+	}
+	s.mu.Lock()
+	_, pinned := s.keyModelHour["keyA"]["srcX::m1"]
+	_, bare := s.keyModelHour["keyA"]["m1"]
+	s.mu.Unlock()
+	if pinned {
+		t.Error("a pinned bucket exists before any quota read the pin — memory regression")
+	}
+	if !bare {
+		t.Error("the bare-model bucket must always exist; an unpinned quota depends on it")
+	}
+}

internal/gateway/key_quota_test.go

@@ -0,0 +1,225 @@
+package gateway
+
+import (
+	"testing"
+	"time"
+
+	"llmsproxy/internal/config"
+)
+
+// ---- data layer: per-key isolation ----
+
+func TestKeyWindowTokensIsolatesKeys(t *testing.T) {
+	s := NewStats(100)
+	now := time.Now().UnixMilli()
+	s.Record(Req{Time: now, Key: "keyA", Model: "m1", Prompt: 500, Compl: 500, OK: true, Status: 200})
+	s.Record(Req{Time: now, Key: "keyB", Model: "m1", Prompt: 500, Compl: 500, OK: true, Status: 200})
+
+	if got := s.KeyWindowModelTokens("keyA", "m1", "", 3600); got != 1000 {
+		t.Errorf("keyA model tokens = %d, want 1000 (its own usage only)", got)
+	}
+	if got := s.KeyWindowModelTokens("keyB", "m1", "", 3600); got != 1000 {
+		t.Errorf("keyB model tokens = %d, want 1000", got)
+	}
+	// the key-blind bucket stays global on purpose: it backs the AUTO slot
+	// quota, which limits the whole gateway, not one key.
+	if got := s.WindowTokens("m1", "", 3600); got != 2000 {
+		t.Errorf("WindowTokens = %d, want 2000 (global per-model total must be unchanged)", got)
+	}
+}
+
+// A source pin scopes the cap to that one upstream. The pinned bucket is
+// maintained lazily (it is only read by quotas that actually pin a source), so
+// usage recorded before anything queried the pin is not attributed to it.
+func TestKeyWindowTokensSeparatesSourcePin(t *testing.T) {
+	s := NewStats(100)
+	now := time.Now().UnixMilli()
+	s.Record(Req{Time: now, Key: "keyA", Model: "m1", Source: "srcX", Prompt: 100, Compl: 100, OK: true, Status: 200}) // 200 tok
+	s.Record(Req{Time: now, Key: "keyA", Model: "m1", Source: "srcY", Prompt: 200, Compl: 200, OK: true, Status: 200}) // 400 tok
+
+	// The unpinned bucket always counts — it is what a scope entry without a
+	// source pin reads. 200 + 400.
+	if got := s.KeyWindowModelTokens("keyA", "m1", "", 3600); got != 600 {
+		t.Errorf("unpinned model tokens = %d, want 600 (both sources)", got)
+	}
+
+	// Reading a pin opts this (key, model) into pinned accounting.
+	if got := s.KeyWindowModelTokens("keyA", "m1", "srcX", 3600); got != 0 {
+		t.Errorf("srcX pin before opt-in = %d, want 0 (the lazy bucket has not accrued yet)", got)
+	}
+
+	// From now on both pins accrue.
+	more := now + 1
+	s.Record(Req{Time: more, Key: "keyA", Model: "m1", Source: "srcX", Prompt: 10, Compl: 10, OK: true, Status: 200}) // 20 tok
+	s.Record(Req{Time: more, Key: "keyA", Model: "m1", Source: "srcY", Prompt: 30, Compl: 30, OK: true, Status: 200}) // 60 tok
+	if got := s.KeyWindowModelTokens("keyA", "m1", "srcX", 3600); got != 20 {
+		t.Errorf("srcX-pinned tokens after opt-in = %d, want 20", got)
+	}
+	if got := s.KeyWindowModelTokens("keyA", "m1", "srcY", 3600); got != 60 {
+		t.Errorf("srcY-pinned tokens = %d, want 60 (a pin read for one source must not blind the other)", got)
+	}
+}
+
+// This is the shape of every real chat record: Source is always populated.
+// The unpinned bucket must still see the tokens, or a per-model quota without
+// a source pin reads an empty bucket and never trips.
+func TestKeyWindowModelTokensUnpinnedSeesSourcedTraffic(t *testing.T) {
+	s := NewStats(100)
+	s.Record(Req{Time: time.Now().UnixMilli(), Key: "keyA", Model: "m1", Source: "deepseek",
+		Prompt: 1000, Compl: 1000, OK: true, Status: 200})
+
+	if got := s.KeyWindowModelTokens("keyA", "m1", "", 3600); got != 2000 {
+		t.Errorf("unpinned tokens = %d, want 2000 — an unpinned per-model quota would never trip otherwise", got)
+	}
+}
+
+func TestKeyWindowRespectsResetWindow(t *testing.T) {
+	s := NewStats(100)
+	old := time.Now().Add(-30 * 24 * time.Hour).UnixMilli()
+	s.Record(Req{Time: old, Key: "keyA", Model: "m1", Prompt: 100, Compl: 100, OK: true, Status: 200})
+	s.Record(Req{Time: time.Now().UnixMilli(), Key: "keyA", Model: "m1", Prompt: 5, Compl: 5, OK: true, Status: 200})
+
+	if got := s.KeyWindowTokens("keyA", 3600); got != 10 {
+		t.Errorf("1h window = %d, want 10 (30-day-old usage must not count)", got)
+	}
+	if got := s.KeyWindowTokens("keyA", 0); got != 210 {
+		t.Errorf("all-time total = %d, want 210", got)
+	}
+}
+
+func TestKeyWindowReqsCountsEveryRequest(t *testing.T) {
+	s := NewStats(100)
+	now := time.Now().UnixMilli()
+	// a failed request and an image request both count: a client that loops on
+	// failures must still burn its request quota
+	s.Record(Req{Time: now, Key: "keyA", Model: "m1", Type: "chat", OK: true, Status: 200})
+	s.Record(Req{Time: now, Key: "keyA", Model: "m1", Type: "chat", OK: false, Status: 500})
+	s.Record(Req{Time: now, Key: "keyA", Model: "img", Type: "image", OK: true, Status: 200})
+	s.Record(Req{Time: now, Key: "keyB", Model: "m1", Type: "chat", OK: true, Status: 200})
+
+	if got := s.KeyWindowReqs("keyA", 3600); got != 3 {
+		t.Errorf("keyA reqs = %d, want 3 (chat ok + chat fail + image)", got)
+	}
+	if got := s.KeyWindowReqs("keyB", 3600); got != 1 {
+		t.Errorf("keyB reqs = %d, want 1", got)
+	}
+}
+
+func TestKeyWindowTokensZeroTokensNotCounted(t *testing.T) {
+	s := NewStats(100)
+	now := time.Now().UnixMilli()
+	// a request that reported no usage must not create a bucket entry
+	s.Record(Req{Time: now, Key: "keyA", Model: "", Type: "chat", OK: true, Status: 200})
+	if got := s.KeyWindowTokens("keyA", 3600); got != 0 {
+		t.Errorf("tokens = %d, want 0", got)
+	}
+	if got := s.KeyWindowReqs("keyA", 3600); got != 1 {
+		t.Errorf("reqs = %d, want 1 (request still happened)", got)
+	}
+}
+
+// ---- reset window arithmetic ----
+
+func TestAutoSecondsToReset(t *testing.T) {
+	if got := AutoSecondsToReset("", 0); got != 0 {
+		t.Errorf("no period = %d, want 0 (never resets -> no retry hint)", got)
+	}
+	now := time.Now().Unix()
+	for _, tc := range []struct {
+		period string
+		hours  int64
+		want   int64
+	}{
+		{"hour", 0, 3600},
+		{"week", 0, 7 * 24 * 3600},
+		{"month", 0, 30 * 24 * 3600},
+		{"nhour", 6, 6 * 3600},
+	} {
+		got := AutoSecondsToReset(tc.period, tc.hours)
+		if got <= 0 || got > tc.want {
+			t.Errorf("AutoSecondsToReset(%q,%d) = %d, want in (0,%d]", tc.period, tc.hours, got, tc.want)
+		}
+		// must never exceed the window itself
+		if got < tc.want-now%3600 {
+			t.Logf("note: %q hint %ds < remaining-window %ds (rounds to hour boundary)", tc.period, got, tc.want-now%3600)
+		}
+		if got > tc.want {
+			t.Errorf("hint %d exceeds window %d", got, tc.want)
+		}
+	}
+	_ = now
+}
+
+func TestAutoSecondsToResetNeverExceedsWindow(t *testing.T) {
+	// "nhour" with a tiny window must not hand out a longer wait than the
+	// window itself (which would stall a client past its own reset)
+	for hours := int64(1); hours <= 48; hours++ {
+		got := AutoSecondsToReset("nhour", hours)
+		if got <= 0 || got > hours*3600 {
+			t.Errorf("nhour/%d = %d, want in (0,%d]", hours, got, hours*3600)
+		}
+	}
+}
+
+// ---- config validation ----
+
+func TestKeyQuotaValidate(t *testing.T) {
+	cases := []struct {
+		name    string
+		q       config.KeyQuota
+		wantErr bool
+	}{
+		{"unlimited", config.KeyQuota{}, false},
+		{"tokens hourly", config.KeyQuota{TokenQuota: 1000, Period: "hour"}, false},
+		{"tokens n-hour", config.KeyQuota{TokenQuota: 1000, Period: "nhour", Hours: 6}, false},
+		{"reqs weekly", config.KeyQuota{ReqQuota: 100, Period: "week"}, false},
+		{"no period never resets", config.KeyQuota{TokenQuota: 1000}, false},
+		{"negative tokens", config.KeyQuota{TokenQuota: -1}, true},
+		{"negative reqs", config.KeyQuota{ReqQuota: -1}, true},
+		{"negative hours", config.KeyQuota{TokenQuota: 5, Hours: -1}, true},
+		{"typo period becomes all-time if accepted", config.KeyQuota{TokenQuota: 5, Period: "houre"}, true},
+		{"nhour without hours", config.KeyQuota{TokenQuota: 5, Period: "nhour"}, true},
+		{"nhour with 0 hours", config.KeyQuota{TokenQuota: 5, Period: "nhour", Hours: 0}, true},
+		{"junk period with only req quota", config.KeyQuota{ReqQuota: 5, Period: "daily"}, true},
+		{"junk period with no caps is irrelevant", config.KeyQuota{Period: "daily"}, false},
+	}
+	for _, tc := range cases {
+		err := tc.q.Validate()
+		if tc.wantErr && err == nil {
+			t.Errorf("%s: expected error, got nil", tc.name)
+		}
+		if !tc.wantErr && err != nil {
+			t.Errorf("%s: unexpected error: %v", tc.name, err)
+		}
+	}
+}
+
+func TestNormalizeRole(t *testing.T) {
+	if got := config.NormalizeRole(""); got != "user" {
+		t.Errorf("empty role = %q, want user", got)
+	}
+	if got := config.NormalizeRole("admin"); got != "admin" {
+		t.Errorf("admin role = %q, want admin", got)
+	}
+	if got := config.NormalizeRole("root"); got != "user" {
+		t.Errorf("unknown role = %q, want user (never escalate)", got)
+	}
+}
+
+// Retention must hold for the lazy pinned buckets too, or a long-lived key
+// would grow without bound.
+func TestPinnedBucketRespectsRetention(t *testing.T) {
+	s := NewStats(100)
+	_ = s.KeyWindowModelTokens("keyA", "m1", "srcX", 3600) // opt in
+	nowH := time.Now().Unix() / 3600
+	for h := int64(0); h < quotaRetentionHours+50; h++ {
+		s.Record(Req{Time: (nowH - h) * 3600 * 1000, Key: "keyA", Model: "m1", Source: "srcX",
+			Prompt: 10, Compl: 10, OK: true, Status: 200})
+	}
+	s.mu.Lock()
+	n := len(s.keyModelHour["keyA"]["srcX::m1"])
+	s.mu.Unlock()
+	if n > quotaRetentionHours {
+		t.Errorf("pinned bucket holds %d hours, want <= %d", n, quotaRetentionHours)
+	}
+}

internal/gateway/key_quota_wiring_test.go

@@ -0,0 +1,444 @@
+package gateway
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"llmsproxy/internal/config"
+	"llmsproxy/internal/core"
+)
+
+// quotaGateway builds a gateway with two user keys that differ only in their
+// configured caps, against one mock upstream that reports 4 tokens per call.
+func quotaGateway(t *testing.T, keyA, keyB config.GWKey) (*Gateway, *upstreamCtrl) {
+	t.Helper()
+	ctrl := &upstreamCtrl{}
+	up := upstream(t, ctrl)
+	t.Cleanup(up.Close)
+
+	td := t.TempDir()
+	cfgPath := td + "/config.yaml"
+	if err := os.WriteFile(cfgPath, []byte("listen: :0"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	cfg := &config.Config{
+		Path:        cfgPath,
+		AdapterDir:  td + "/adapters",
+		RuntimeFile: td + "/runtime.json",
+		Keys:        []config.GWKey{keyA, keyB},
+		Sources: []config.Source{{
+			Name: "up", BaseURL: up.URL, Adapter: "openai",
+			Models: []config.Model{{ID: "m1", Priority: 100}},
+		}},
+	}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(c.Close)
+	g, err := New(c)
+	if err != nil {
+		t.Fatalf("gateway: %v", err)
+	}
+	return g, ctrl
+}
+
+// chatAs issues a chat completion with a specific gateway key and returns the
+// recorder plus the decoded error code.
+func chatAs(t *testing.T, g *Gateway, key, model string) (*httptest.ResponseRecorder, string) {
+	t.Helper()
+	req, _ := http.NewRequest("POST", "/v1/chat/completions",
+		strings.NewReader(fmt.Sprintf(`{"model":%q,"messages":[{"role":"user","content":"hi"}]}`, model)))
+	req.Header.Set("Authorization", "Bearer "+key)
+	req.Header.Set("Content-Type", "application/json")
+	rr := httptest.NewRecorder()
+	g.Handler().ServeHTTP(rr, req)
+	code := ""
+	if rr.Code != 200 {
+		var e struct {
+			Error struct {
+				Type string `json:"type"`
+			} `json:"error"`
+		}
+		_ = json.Unmarshal(rr.Body.Bytes(), &e)
+		code = e.Error.Type
+	}
+	return rr, code
+}
+
+// A key whose token budget is spent must be refused, and the refusal must be a
+// 429 the client can retry after the window rolls over — not a 403 that reads
+// as "this key may never use this model".
+func TestKeyTokenQuotaBlocksWithRetryAfter(t *testing.T) {
+	g, _ := quotaGateway(t,
+		config.GWKey{Key: "sk-a", Role: "user",
+			Models: []config.ModelScope{{Model: "m1", TokenQuota: 8, Period: "hour"}}},
+		config.GWKey{Key: "sk-b", Role: "user", Name: "b"},
+	)
+	// 4 tokens per call, budget 8 -> the third call crosses it
+	for i := 1; i <= 2; i++ {
+		rr, _ := chatAs(t, g, "sk-a", "m1")
+		if rr.Code != 200 {
+			t.Fatalf("call %d: want 200, got %d (%s)", i, rr.Code, rr.Body.String())
+		}
+	}
+	rr, code := chatAs(t, g, "sk-a", "m1")
+	if rr.Code != http.StatusTooManyRequests {
+		t.Fatalf("third call: want 429, got %d (%s)", rr.Code, rr.Body.String())
+	}
+	if code != "rate_limit_exceeded" {
+		t.Errorf("error code = %q, want rate_limit_exceeded", code)
+	}
+	if ra := rr.Header().Get("Retry-After"); ra == "" {
+		t.Error("429 must carry Retry-After so a client knows when to come back")
+	} else if n := mustAtoi(t, ra); n <= 0 || n > 3600 {
+		t.Errorf("Retry-After = %q, want a positive value within the 1h window", ra)
+	}
+	if !strings.Contains(rr.Body.String(), "quota") {
+		t.Errorf("body should name the quota: %s", rr.Body.String())
+	}
+}
+
+// The quota is per key: exhausting one key's budget must not affect another
+// key that is allowed the same model.
+func TestKeyTokenQuotaIsIsolatedPerKey(t *testing.T) {
+	g, _ := quotaGateway(t,
+		config.GWKey{Key: "sk-a", Role: "user",
+			Models: []config.ModelScope{{Model: "m1", TokenQuota: 4, Period: "hour"}}},
+		config.GWKey{Key: "sk-b", Role: "user",
+			Models: []config.ModelScope{{Model: "m1", TokenQuota: 1000, Period: "hour"}}},
+	)
+	rr, _ := chatAs(t, g, "sk-a", "m1")
+	if rr.Code != 200 {
+		t.Fatalf("keyA first call: want 200, got %d", rr.Code)
+	}
+	if rr, code := chatAs(t, g, "sk-a", "m1"); rr.Code != http.StatusTooManyRequests {
+		t.Fatalf("keyA second call: want 429, got %d (%s)", rr.Code, code)
+	}
+	// keyB has its own budget and must still be served
+	if rr, _ := chatAs(t, g, "sk-b", "m1"); rr.Code != 200 {
+		t.Fatalf("keyB must be unaffected by keyA's exhausted quota, got %d", rr.Code)
+	}
+}
+
+// A per-model quota inside the scope must also be per key: one key burning a
+// shared model budget must not lock the other key out.
+func TestScopeModelTokenQuotaIsolatedPerKey(t *testing.T) {
+	g, _ := quotaGateway(t,
+		config.GWKey{Key: "sk-a", Role: "user", Models: []config.ModelScope{
+			{Model: "m1", TokenQuota: 4, Period: "hour"},
+		}},
+		config.GWKey{Key: "sk-b", Role: "user", Models: []config.ModelScope{
+			{Model: "m1", TokenQuota: 4, Period: "hour"},
+		}},
+	)
+	if rr, _ := chatAs(t, g, "sk-a", "m1"); rr.Code != 200 {
+		t.Fatalf("keyA first call: want 200, got %d", rr.Code)
+	}
+	if rr, _ := chatAs(t, g, "sk-a", "m1"); rr.Code != http.StatusTooManyRequests {
+		t.Fatalf("keyA second call: want 429 (its own model quota), got %d", rr.Code)
+	}
+	// keyB shares the model but not the counter — it must be served twice too
+	if rr, _ := chatAs(t, g, "sk-b", "m1"); rr.Code != 200 {
+		t.Fatalf("keyB first call: want 200, got %d", rr.Code)
+	}
+	if rr, _ := chatAs(t, g, "sk-b", "m1"); rr.Code != http.StatusTooManyRequests {
+		t.Fatalf("keyB second call: want 429, got %d", rr.Code)
+	}
+}
+
+// An admin key is never capped: a budget set on an admin key must not lock the
+// operator out of the gateway they administer.
+func TestAdminKeyIsNeverQuotaCapped(t *testing.T) {
+	g, _ := quotaGateway(t,
+		config.GWKey{Key: "sk-admin", Role: "admin",
+			Models: []config.ModelScope{{Model: "m1", TokenQuota: 1, ReqQuota: 1, Period: "hour"}}},
+		config.GWKey{Key: "sk-b", Role: "user"},
+	)
+	for i := 1; i <= 3; i++ {
+		if rr, _ := chatAs(t, g, "sk-admin", "m1"); rr.Code != 200 {
+			t.Fatalf("admin call %d: want 200 (admin keys are uncapped), got %d", i, rr.Code)
+		}
+	}
+}
+
+// A request quota caps sustained volume; RPM-style bursts are not its job but
+// the count must still stop the key.
+func TestKeyRequestQuotaBlocks(t *testing.T) {
+	g, ctrl := quotaGateway(t,
+		config.GWKey{Key: "sk-a", Role: "user",
+			Models: []config.ModelScope{{Model: "m1", ReqQuota: 2, Period: "hour"}}},
+		config.GWKey{Key: "sk-b", Role: "user"},
+	)
+	for i := 1; i <= 2; i++ {
+		if rr, _ := chatAs(t, g, "sk-a", "m1"); rr.Code != 200 {
+			t.Fatalf("call %d: want 200, got %d", i, rr.Code)
+		}
+	}
+	rr, code := chatAs(t, g, "sk-a", "m1")
+	if rr.Code != http.StatusTooManyRequests || code != "rate_limit_exceeded" {
+		t.Fatalf("third call: want 429/rate_limit_exceeded, got %d/%s", rr.Code, code)
+	}
+	if ctrl.hits != 2 {
+		t.Errorf("upstream saw %d calls, want 2 (a refused request must not reach upstream)", ctrl.hits)
+	}
+}
+
+// A model outside the scope is still 403, not 429: retrying cannot help.
+func TestModelOutsideScopeStaysForbidden(t *testing.T) {
+	g, _ := quotaGateway(t,
+		config.GWKey{Key: "sk-a", Role: "user",
+			Models: []config.ModelScope{{Model: "other-model", TokenQuota: 1000, Period: "hour"}}},
+		config.GWKey{Key: "sk-b", Role: "user"},
+	)
+	rr, code := chatAs(t, g, "sk-a", "m1")
+	if rr.Code != http.StatusForbidden {
+		t.Fatalf("want 403, got %d (%s)", rr.Code, rr.Body.String())
+	}
+	if code != "model_not_allowed" {
+		t.Errorf("error code = %q, want model_not_allowed", code)
+	}
+	if rr.Header().Get("Retry-After") != "" {
+		t.Error("a 403 must not advertise a retry time")
+	}
+}
+
+// An unlimited key (no caps at all) is never rejected — the default config
+// must keep working exactly as before.
+func TestKeyWithoutCapsIsUnlimited(t *testing.T) {
+	g, _ := quotaGateway(t,
+		config.GWKey{Key: "sk-a", Role: "user", Models: []config.ModelScope{{Model: "m1"}}},
+		config.GWKey{Key: "sk-b", Role: "user"},
+	)
+	for i := 1; i <= 5; i++ {
+		if rr, _ := chatAs(t, g, "sk-a", "m1"); rr.Code != 200 {
+			t.Fatalf("call %d: want 200 (no caps = unlimited), got %d", i, rr.Code)
+		}
+	}
+}
+
+func mustAtoi(t *testing.T, s string) int64 {
+	t.Helper()
+	var n int64
+	if _, err := fmt.Sscanf(s, "%d", &n); err != nil {
+		t.Fatalf("Retry-After %q is not an integer: %v", s, err)
+	}
+	return n
+}
+
+// quotaCtx builds a request context carrying a gateway key, as the auth
+// middleware would.
+func quotaCtx(t *testing.T, g *Gateway, key string) context.Context {
+	t.Helper()
+	return withAuth(context.Background(), key, "user")
+}
+
+func nowMSOffset(sec int64) int64 {
+	return time.Now().Add(time.Duration(sec) * time.Second).UnixMilli()
+}
+
+// The key-wide quota and the AUTO slot quota are different limits with
+// different scopes: the slot quota is gateway-wide (a shared upstream budget),
+// the key quota belongs to one caller. When both are exhausted the caller must
+// see the KEY quota, because that is the one it can act on — the slot quota
+// would otherwise be reported as "no capacity", which reads like an outage.
+func TestKeyQuotaWinsOverSlotQuota(t *testing.T) {
+	up := upstream(t, &upstreamCtrl{})
+	defer up.Close()
+	g := newQuotaGW(t, up.URL,
+		config.GWKey{Key: "sk-a", Role: "user",
+			Models: []config.ModelScope{{Model: "AUTO", TokenQuota: 4, Period: "hour"}}})
+	ctx := quotaCtx(t, g, "sk-a")
+
+	// exhaust the key first
+	g.stats.Record(Req{Time: time.Now().UnixMilli(), Key: keyID("sk-a"), Model: "m1",
+		Source: "up", Prompt: 100, Compl: 100, OK: true, Status: 200})
+	rr, code := chatAs(t, g, "sk-a", "AUTO")
+	if rr.Code != http.StatusTooManyRequests {
+		t.Fatalf("want 429 from the key quota, got %d (%s)", rr.Code, rr.Body.String())
+	}
+	if code != "rate_limit_exceeded" {
+		t.Errorf("code = %q, want rate_limit_exceeded", code)
+	}
+	_ = ctx
+}
+
+// A key with NO caps must never be blocked by the slot quota check reaching it
+// through the scope path: scopeTokens on an uncapped key returns 0 usage and
+// the limit check must treat that as "no cap", not "exhausted".
+func TestUncappedKeyNeverBlockedByEmptyScope(t *testing.T) {
+	up := upstream(t, &upstreamCtrl{})
+	defer up.Close()
+	g := newQuotaGW(t, up.URL,
+		config.GWKey{Key: "sk-a", Role: "user", Models: []config.ModelScope{{Model: "AUTO"}}})
+	for i := 0; i < 5; i++ {
+		if rr, _ := chatAs(t, g, "sk-a", "AUTO"); rr.Code != 200 {
+			t.Fatalf("call %d: want 200 (AUTO scope with no quota), got %d (%s)", i, rr.Code, rr.Body.String())
+		}
+	}
+}
+
+// newQuotaGW builds a gateway with one mock upstream serving model m1 and the
+// given keys.
+func newQuotaGW(t *testing.T, upURL string, keys ...config.GWKey) *Gateway {
+	t.Helper()
+	td := t.TempDir()
+	cfgPath := filepath.Join(td, "config.yaml")
+	if err := os.WriteFile(cfgPath, []byte("listen: :0"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	cfg := &config.Config{
+		Path:        cfgPath,
+		AdapterDir:  filepath.Join(td, "adapters"),
+		RuntimeFile: filepath.Join(td, "runtime.json"),
+		Keys:        keys,
+		Sources: []config.Source{{
+			Name: "up", BaseURL: upURL, Adapter: "openai",
+			Models: []config.Model{
+				{ID: "m1", Priority: 100},
+				{ID: "m2", Priority: 90},
+			},
+		}},
+	}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(c.Close)
+	secrets := make([]string, 0, len(keys))
+	for _, k := range keys {
+		secrets = append(secrets, k.Key)
+	}
+	g, err := New(c)
+	if err != nil {
+		t.Fatalf("gateway: %v", err)
+	}
+	return g
+}
+
+// The core of per-model quotas: a model that runs out of budget must stop
+// being served on its own, while every other model on the SAME key keeps
+// working. A key-wide total would fail this — it would block m2 because m1 was
+// capped, which is exactly the coupling this design removes.
+func TestOneModelsQuotaDoesNotBlockAnother(t *testing.T) {
+	up := upstream(t, &upstreamCtrl{})
+	defer up.Close()
+	td := t.TempDir()
+	cfgPath := td + "/config.yaml"
+	if err := os.WriteFile(cfgPath, []byte("listen: :0"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	cfg := &config.Config{
+		Path: cfgPath, AdapterDir: filepath.Join(td, "adapters"),
+		RuntimeFile: filepath.Join(td, "runtime.json"),
+		Keys: []config.GWKey{{
+			Key: "sk-a", Role: "user", Name: "two-models",
+			Models: []config.ModelScope{
+				{Model: "m1", TokenQuota: 4, Period: "hour"},
+				{Model: "m2", TokenQuota: 1000, Period: "hour"},
+			},
+		}},
+		Sources: []config.Source{{
+			Name: "up", BaseURL: up.URL, Adapter: "openai",
+			Models: []config.Model{
+				{ID: "m1", Priority: 100},
+				{ID: "m2", Priority: 90},
+			},
+		}},
+	}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(c.Close)
+	g, err := New(c)
+	if err != nil {
+		t.Fatalf("gateway: %v", err)
+	}
+
+	// Spend on m2 FIRST. Without this the two designs are
+	// indistinguishable: a key-wide counter and m1's own counter would both
+	// read 0 before m1 is used, so the test would pass either way (it did —
+	// see the commit that rewrote it).
+	for i := 1; i <= 3; i++ {
+		if rr, _ := chatAs(t, g, "sk-a", "m2"); rr.Code != 200 {
+			t.Fatalf("m2 priming call %d: want 200, got %d", i, rr.Code)
+		}
+	}
+	// m1's budget is 4 and each call costs 4, so the key-wide total (m2+m1) is
+	// already 12 when m1 starts: a key-wide cap would refuse m1 immediately.
+	if rr, _ := chatAs(t, g, "sk-a", "m1"); rr.Code != 200 {
+		t.Fatalf("m1 must be served on its own budget (m2's spend is not its problem), got %d (%s)",
+			rr.Code, rr.Body.String())
+	}
+	if rr, code := chatAs(t, g, "sk-a", "m1"); rr.Code != http.StatusTooManyRequests {
+		t.Fatalf("m1 second call: want 429, got %d (%s)", rr.Code, code)
+	}
+	// m2 must keep working
+	for i := 1; i <= 3; i++ {
+		if rr, _ := chatAs(t, g, "sk-a", "m2"); rr.Code != 200 {
+			t.Fatalf("m2 call %d must be served while m1 is capped, got %d (%s)", i, rr.Code, rr.Body.String())
+		}
+	}
+	// and the message must name m1, not the key
+	rr, _ := chatAs(t, g, "sk-a", "m1")
+	if !strings.Contains(rr.Body.String(), "m1") {
+		t.Errorf("rejection should name the capped model: %s", rr.Body.String())
+	}
+}
+
+// A model with no quota on it is never blocked by a sibling's cap, and an
+// uncapped key is never blocked at all.
+func TestUncappedModelNeverBlocked(t *testing.T) {
+	up := upstream(t, &upstreamCtrl{})
+	defer up.Close()
+	g := newQuotaGW(t, up.URL,
+		config.GWKey{Key: "sk-a", Role: "user", Models: []config.ModelScope{
+			{Model: "m1", TokenQuota: 1, Period: "hour"},
+			{Model: "m2"},
+		}})
+	if rr, _ := chatAs(t, g, "sk-a", "m1"); rr.Code != 200 {
+		t.Fatalf("m1 first: want 200, got %d", rr.Code)
+	}
+	if rr, _ := chatAs(t, g, "sk-a", "m1"); rr.Code != http.StatusTooManyRequests {
+		t.Fatalf("m1 second: want 429, got %d", rr.Code)
+	}
+	for i := 1; i <= 4; i++ {
+		if rr, _ := chatAs(t, g, "sk-a", "m2"); rr.Code != 200 {
+			t.Fatalf("m2 (uncapped) call %d: want 200, got %d", i, rr.Code)
+		}
+	}
+}
+
+// An admin key is never capped even when its scopes carry budgets: a cap that
+// locked the operator out would be unrecoverable through the UI.
+func TestAdminKeyScopesAreNotEnforced(t *testing.T) {
+	up := upstream(t, &upstreamCtrl{})
+	defer up.Close()
+	g := newQuotaGW(t, up.URL,
+		config.GWKey{Key: "sk-admin", Role: "admin", Models: []config.ModelScope{
+			{Model: "m1", TokenQuota: 1, ReqQuota: 1, Period: "hour"},
+		}})
+	for i := 1; i <= 4; i++ {
+		if rr, _ := chatAs(t, g, "sk-admin", "m1"); rr.Code != 200 {
+			t.Fatalf("admin call %d: want 200 (admin scopes are not enforced), got %d", i, rr.Code)
+		}
+	}
+}

internal/gateway/keys.go

@@ -7,6 +7,7 @@ import (
 	"strings"
 
 	"llmsproxy/internal/config"
+	"llmsproxy/internal/scheduler"
 )
 
 // handleKeysAPI manages gateway keys: GET /api/keys (admin: all keys),
@@ -17,10 +18,22 @@ func (g *Gateway) handleKeysAPI(w http.ResponseWriter, r *http.Request) {
 	path = strings.Trim(path, "/")
 	role := reqRole(r.Context())
 
+	// /api/keys/me/auto — a user's own chain, readable without admin rights.
+	// Must be matched before the generic {key}/auto case below so it does not
+	// ask admin rights of a plain user asking about their own chain.
+	if path == "me/auto" {
+		g.handleKeyMeAuto(w, r)
+		return
+	}
 	if path == "me" {
 		g.handleKeyMe(w, r)
 		return
 	}
+	// /api/keys/{key}/auto — the per-key AUTO chain editor.
+	if i := strings.LastIndex(path, "/auto"); i > 0 && path[i+len("/auto"):] == "" {
+		g.handleKeyAutoAPI(w, r, path[:i])
+		return
+	}
 	if role != "admin" {
 		writeError(w, http.StatusForbidden, "forbidden", "admin role required")
 		return
@@ -48,13 +61,7 @@ func (g *Gateway) handleKeysAPI(w http.ResponseWriter, r *http.Request) {
 			writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
 			return
 		}
-		if body.Role == "" {
-			body.Role = "user"
-		}
-		if body.Role != "admin" && body.Role != "user" {
-			writeError(w, http.StatusBadRequest, "invalid_request", "role must be admin or user")
-			return
-		}
+		body.Role = config.NormalizeRole(body.Role)
 		rec, err := g.core.CreateKey(body.Name, body.Role, body.Models, body.Note)
 		if err != nil {
 			writeError(w, http.StatusBadRequest, "key_error", err.Error())
@@ -76,6 +83,9 @@ func (g *Gateway) handleKeysAPI(w http.ResponseWriter, r *http.Request) {
 			writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
 			return
 		}
+		// Each scope entry carries its own token/request caps, so replacing the
+		// scope replaces the budgets with it — there is no separate key-wide
+		// quota that could drift out of sync with the models.
 		rec, err := g.core.UpdateKey(path, body.Name, body.Role, body.Models, body.Note)
 		if err != nil {
 			writeError(w, http.StatusBadRequest, "key_error", err.Error())
@@ -142,6 +152,101 @@ func (g *Gateway) allowedModels(ctx context.Context) []config.ModelScope {
 	return rec.Models
 }
 
+// handleKeyAutoAPI manages one key's own AUTO chain:
+// GET /api/keys/{key}/auto returns it, PUT replaces it, DELETE clears the
+// override so the key falls back to the global chain. Admin only — a user can
+// inspect their own chain through GET /api/keys/me/auto.
+//
+// An empty chain and a cleared chain are deliberately the same state: "inherit
+// the global chain". That keeps the WebUI's "use global" toggle a plain DELETE
+// with no sentinel value to carry through config.yaml.
+func (g *Gateway) handleKeyAutoAPI(w http.ResponseWriter, r *http.Request, key string) {
+	if reqRole(r.Context()) != "admin" {
+		writeError(w, http.StatusForbidden, "forbidden", "admin role required")
+		return
+	}
+	rec, ok := g.core.FindKey(key)
+	if !ok {
+		writeError(w, http.StatusNotFound, "not_found", "key not found")
+		return
+	}
+	switch r.Method {
+	case http.MethodGet:
+		writeJSON(w, http.StatusOK, map[string]interface{}{
+			"key":      key,
+			"auto":     rec.Auto,
+			"inherits": !rec.HasOwnAuto(),
+		})
+	case http.MethodPut, http.MethodPatch:
+		var body struct {
+			Auto []config.ModelScope `json:"auto"`
+		}
+		if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+			writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+			return
+		}
+		if len(body.Auto) == 0 {
+			// Treat an empty PUT as "give up the override" so the endpoint
+			// cannot persist a chain that would 503 with no slots.
+			if err := g.core.SaveKeyAuto(key, nil); err != nil {
+				writeError(w, http.StatusBadRequest, "key_error", err.Error())
+				return
+			}
+			writeJSON(w, http.StatusOK, map[string]interface{}{"ok": true, "inherits": true})
+			return
+		}
+		if err := g.core.SaveKeyAuto(key, body.Auto); err != nil {
+			writeError(w, http.StatusBadRequest, "key_error", err.Error())
+			return
+		}
+		// Report whether the chain actually resolved to usable slots. An
+		// admin who typed a model that no longer exists should learn it here,
+		// not from the user's next 503.
+		chain, _ := g.core.AutoChainFor(key)
+		writeJSON(w, http.StatusOK, map[string]interface{}{
+			"ok": true, "inherits": false, "slots": chainSlots(chain),
+		})
+	case http.MethodDelete:
+		if err := g.core.SaveKeyAuto(key, nil); err != nil {
+			writeError(w, http.StatusBadRequest, "key_error", err.Error())
+			return
+		}
+		writeJSON(w, http.StatusOK, map[string]interface{}{"ok": true, "inherits": true})
+	default:
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "")
+	}
+}
+
+// chainSlots counts the usable slots in a chain, for API feedback after an
+// edit. A chain that compiled to zero slots is reported so the caller can warn
+// instead of letting the next request fail with no_provider.
+func chainSlots(ch *scheduler.Chain) int {
+	if ch == nil {
+		return 0
+	}
+	n := 0
+	for _, tn := range ch.Tiers {
+		n += len(tn.Slots)
+	}
+	return n
+}
+
+// handleKeyMeAuto lets a user read their own AUTO chain without admin rights.
+func (g *Gateway) handleKeyMeAuto(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodGet {
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "use GET")
+		return
+	}
+	rec, ok := g.core.FindKey(reqKey(r.Context()))
+	if !ok {
+		writeError(w, http.StatusUnauthorized, "invalid_api_key", "key not found")
+		return
+	}
+	writeJSON(w, http.StatusOK, map[string]interface{}{
+		"key": rec.Key, "auto": rec.Auto, "inherits": !rec.HasOwnAuto(),
+	})
+}
+
 // handleAutoAPI manages the AUTO scheduling slots: GET /api/auto returns the
 // current rules; PUT /api/auto replaces them (admin only).
 func (g *Gateway) handleAutoAPI(w http.ResponseWriter, r *http.Request) {

internal/gateway/overflow_err_test.go

@@ -0,0 +1,167 @@
+package gateway
+
+import (
+	"errors"
+	"regexp"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/scheduler"
+	"llmsproxy/internal/types"
+)
+
+// piOverflowRe 是 @earendil-works/pi-ai 的 OVERFLOW_PATTERNS 中对应的一条。
+// 归一化标记必须命中它，否则改了等于没改：客户端仍不会压缩重试。
+var piOverflowRe = regexp.MustCompile(`context[_ ]length[_ ]exceeded`)
+
+func TestOverflowCanonicalMatchesPiPattern(t *testing.T) {
+	if !piOverflowRe.MatchString(overflowCanonical) {
+		t.Fatalf("归一化标记 %q 不匹配 pi 的超窗正则，客户端仍不会压缩重试", overflowCanonical)
+	}
+}
+
+// justworker 的真实措辞：与 pi 现有 25 条正则一条都不匹配，
+// 所以必须在网关侧归一化，否则超窗对客户端完全不可见。
+func TestJustworkerOverflowIsNormalized(t *testing.T) {
+	raw := "请精简对话历史…(Context window is full, please reduce the conversation history)…"
+
+	// 前提校验：原始措辞确实不被 pi 认出（否则本用例失去意义）。
+	if piOverflowRe.MatchString(raw) {
+		t.Fatalf("前提不成立：原始措辞本就匹配 pi 正则: %q", raw)
+	}
+	if !looksLikeOverflow(raw) {
+		t.Fatalf("网关应识别出这是超窗: %q", raw)
+	}
+
+	err := &scheduler.ChainErr{Tiers: []scheduler.TierError{{
+		Tier: 1, Source: "justworker", Model: "claude-opus-5", Err: errors.New(raw),
+	}}}
+
+	got := clientUpstreamErr(err)
+	if !piOverflowRe.MatchString(got) {
+		t.Fatalf("归一化后仍不匹配 pi 正则，客户端不会压缩重试: %q", got)
+	}
+}
+
+// 直连路径（非 AUTO 链）也要归一化——它走的是同一个入口。
+func TestDirectPathOverflowIsNormalized(t *testing.T) {
+	got := clientUpstreamErr(errors.New("upstream 400: Context window is full"))
+	if !piOverflowRe.MatchString(got) {
+		t.Fatalf("直连超窗未被归一化: %q", got)
+	}
+}
+
+// 非超窗错误不能被加上超窗标记，否则会把普通失败误报成需要压缩。
+func TestNonOverflowNotNormalized(t *testing.T) {
+	err := &scheduler.ChainErr{Tiers: []scheduler.TierError{{
+		Tier: 1, Source: "x", Model: "y", Err: errors.New("502 Bad Gateway: upstream connect error"),
+	}}}
+	got := clientUpstreamErr(err)
+	if strings.Contains(got, overflowCanonical) {
+		t.Fatalf("普通上游失败被误标为超窗: %q", got)
+	}
+	if !strings.Contains(got, "502 Bad Gateway") {
+		t.Fatalf("普通失败的诊断信息应保留: %q", got)
+	}
+}
+
+// 回归：短诊断词常落在错误尾部，80 字节截断会正好把它切掉，
+// 超窗就永远认不出来。宽度放宽后必须保住。
+func TestOverflowPhraseSurvivesTruncationWidth(t *testing.T) {
+	prefix := strings.Repeat("x", 100)
+	// OneLine 的长度限制作用于整条 "source/model: err"，不是只算 err 本身。
+	full := "justworker/claude-opus-5: " + prefix + " Context window is full"
+
+	// 前提：80 字节下短语确实会被切掉（这正是修复前的行为）。
+	if got := types.OneLine(full, 80); strings.Contains(got, "Context window is full") {
+		t.Fatalf("前提不成立：80 字节下短语居然还在: %q", got)
+	}
+	// 160 字节下必须保住，否则超窗识别会被截断悄悄破坏。
+	if got := types.OneLine(full, 160); !strings.Contains(got, "Context window is full") {
+		t.Fatalf("160 字节应保住尾部短诊断词: %q", got)
+	}
+
+	err := &scheduler.ChainErr{Tiers: []scheduler.TierError{{
+		Tier: 1, Source: "justworker", Model: "claude-opus-5",
+		Err: errors.New(prefix + " Context window is full"),
+	}}}
+
+	got := clientUpstreamErr(err)
+	if !strings.Contains(strings.ToLower(got), "context window is full") {
+		t.Fatalf("宽度 160 应保住尾部短诊断词，实际: %q", got)
+	}
+	if !piOverflowRe.MatchString(got) {
+		t.Fatalf("归一化标记丢失: %q", got)
+	}
+}
+
+// piNonOverflowRe 是 @earendil-works/pi-ai 的 NON_OVERFLOW_PATTERNS。
+// pi 的 isContextOverflow 先查这一组，命中则直接判为「非超窗」——
+// 即使 OVERFLOW_PATTERNS 也命中，也不会压缩重试。
+var piNonOverflowRe = []*regexp.Regexp{
+	regexp.MustCompile(`^(Throttling error|Service unavailable):`),
+	regexp.MustCompile(`(?i)rate limit`),
+	regexp.MustCompile(`(?i)too many requests`),
+}
+
+// piWouldCompact 复刻 isContextOverflow 的 Case 1 判定顺序：
+// 先排除 NON_OVERFLOW，再匹配 OVERFLOW。
+func piWouldCompact(errorMessage string) bool {
+	for _, p := range piNonOverflowRe {
+		if p.MatchString(errorMessage) {
+			return false
+		}
+	}
+	return piOverflowRe.MatchString(errorMessage)
+}
+
+// issue 回归：AUTO 链失败消息天然把「超窗 tier」和「配额/限流 tier」拼在一起，
+// 超窗标记后面只要出现 rate limit / too many requests 的字样，
+// pi 就会把整条消息判为「非超窗」而拒绝压缩重试。
+func TestOverflowMessageSurvivesRateLimitedSiblingTier(t *testing.T) {
+	cases := []struct {
+		name    string
+		sibling string
+	}{
+		{"rate limit 措辞", "api error 429: rate limit exceeded"},
+		{"too many requests 措辞", "429 Too Many Requests"},
+	}
+
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			err := &scheduler.ChainErr{Tiers: []scheduler.TierError{
+				{Tier: 1, Source: "sensenova", Model: "glm-5.2", Err: errors.New(tc.sibling)},
+				{Tier: 3, Source: "gozen", Model: "deepseek-v4.1-flash",
+					Err: errors.New("api error 400: This model's maximum context length is 1048576 tokens")},
+			}}
+
+			got := clientUpstreamErr(err)
+			if !piWouldCompact(got) {
+				t.Fatalf("同链限流 tier 封杀了超窗识别，pi 不会压缩重试: %q", got)
+			}
+			// 干净：不得把限流 tier 的措辞带出去。
+			for _, p := range piNonOverflowRe {
+				if p.MatchString(got) {
+					t.Fatalf("超窗消息仍带限流措辞 %v: %q", p, got)
+				}
+			}
+			// 但仍要能看出是哪个源超的窗。
+			if !strings.Contains(got, "gozen/deepseek-v4.1-flash") {
+				t.Errorf("超窗源信息丢失: %q", got)
+			}
+		})
+	}
+}
+
+// 直连超窗同样要干净可用。
+func TestDirectOverflowMessageIsClean(t *testing.T) {
+	raw := "api error 400: Error from provider (Console Go): This model's maximum context length is 1048576 tokens. " +
+		"However, you requested 1440050 tokens. Please reduce the length of the messages or completion."
+	got := clientUpstreamErr(errors.New(raw))
+	if !piWouldCompact(got) {
+		t.Fatalf("直连超窗未被 pi 识别: %q", got)
+	}
+	if !strings.Contains(got, overflowCanonical) {
+		t.Fatalf("缺少归一化标记: %q", got)
+	}
+}

internal/gateway/plugin_call_api_test.go

@@ -0,0 +1,150 @@
+package gateway
+
+// The /api/plugins/{name}/call/{fn} endpoint splits its path with string
+// surgery, and the split point is easy to get wrong in a way that still
+// "works": keeping the "call/" segment in the function name produced
+// `does not export "call/collect_account"` — the plugin was correct, the router
+// was not, and every call 502'd while the exports list looked perfectly fine.
+//
+// These tests drive the real router, so a refactor that reintroduces the
+// off-by-one is caught here rather than in production.
+
+import (
+	"encoding/json"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+	"llmsproxy/internal/core"
+)
+
+// callPlugin exposes exactly one export plus a function that is NOT exported.
+// If the exports gate is not honoured, secretFn is reachable and with it the
+// plugin's private state — for a plugin holding an API key that means handing
+// the key to anyone who can POST to the gateway.
+const callPlugin = `
+local P = {}
+P.name = "calltest"
+P.version = "1.0.0"
+P.exports = { "add" }
+P.secret = "do-not-leak"
+function P.add(a)
+  return { result = (tonumber(a and a.x) or 0) + 1 }
+end
+function P.secretFn()
+  return { leaked = P.secret }
+end
+return P
+`
+
+func gatewayWithCallPlugin(t *testing.T) *Gateway {
+	t.Helper()
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	body := "listen: :0\n" +
+		"adapter_dir: " + filepath.Join(dir, "adapters") + "\n" +
+		"plugin_dir: " + filepath.Join(dir, "plugins") + "\n" +
+		"runtime_file: " + filepath.Join(dir, "runtime.json") + "\n" +
+		"gateway_keys:\n  - sk-test\n"
+	if err := os.WriteFile(cfgPath, []byte(body), 0600); err != nil {
+		t.Fatal(err)
+	}
+	cfg, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(c.Close)
+	if err := c.Plugins().LoadSource("calltest", callPlugin); err != nil {
+		t.Fatalf("load calltest: %v", err)
+	}
+	g, err := New(c)
+	if err != nil {
+		t.Fatalf("gateway: %v", err)
+	}
+	return g
+}
+
+func TestPluginCallSplitsFunctionNameCorrectly(t *testing.T) {
+	g := gatewayWithCallPlugin(t)
+	rr := doReq(t, g, http.MethodPost,
+		"/api/plugins/calltest/call/add", `{"x":41}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("call failed: HTTP %d %s", rr.Code, rr.Body.String())
+	}
+	var out struct {
+		Result struct {
+			Result int `json:"result"`
+		} `json:"result"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &out); err != nil {
+		t.Fatalf("decode: %v (body=%s)", err, rr.Body.String())
+	}
+	// 41 + 1 == 42. Reaching 42 proves BOTH halves of the contract: the "/call/"
+	// segment did not leak into the function name, and the payload arrived.
+	if out.Result.Result != 42 {
+		t.Fatalf("got %d, want 42 — either the /call/ segment leaked into the "+
+			"function name or the argument table never reached Lua", out.Result.Result)
+	}
+}
+
+func TestPluginCallRefusesUndeclaredFunction(t *testing.T) {
+	g := gatewayWithCallPlugin(t)
+	rr := doReq(t, g, http.MethodPost,
+		"/api/plugins/calltest/call/secretFn", `{}`)
+	if rr.Code == http.StatusOK {
+		t.Fatalf("an unexported function must not be callable, got 200: %s",
+			rr.Body.String())
+	}
+	if strings.Contains(rr.Body.String(), "do-not-leak") {
+		t.Fatal("the refusal leaked the plugin's private state")
+	}
+}
+
+// A plugin that declares no exports at all has no callable surface. Reading the
+// missing list as "allow everything" would turn every plugin into a remote
+// evaluator the moment this endpoint ships.
+func TestPluginCallWithoutExportsHasNoCallableSurface(t *testing.T) {
+	g := gatewayWithCallPlugin(t)
+	const bare = `
+local P = {}
+P.name = "bare"
+P.version = "1.0.0"
+function P.anything()
+  return { ok = true }
+end
+return P
+`
+	if err := g.core.Plugins().LoadSource("bare", bare); err != nil {
+		t.Fatalf("load bare: %v", err)
+	}
+	rr := doReq(t, g, http.MethodPost,
+		"/api/plugins/bare/call/anything", `{}`)
+	if rr.Code == http.StatusOK {
+		t.Fatalf("a plugin with no plugin.exports must expose nothing, got 200: %s",
+			rr.Body.String())
+	}
+}
+
+func TestPluginCallIsAdminOnly(t *testing.T) {
+	g := gatewayWithCallPlugin(t)
+	// A real user-role key, not a bare one from config: sk-test/sk-user in the
+	// config both default to admin, so a test using them would pass the gate for
+	// the wrong reason and prove nothing.
+	rec, err := g.core.CreateKey("viewer", "user", nil, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	rr := doReqAs(t, g, rec.Key, http.MethodPost,
+		"/api/plugins/calltest/call/add", `{"x":1}`)
+	if rr.Code != http.StatusForbidden {
+		t.Fatalf("a non-admin key reached the call endpoint: HTTP %d %s",
+			rr.Code, rr.Body.String())
+	}
+}

internal/gateway/plugin_http_routes.go

@@ -0,0 +1,207 @@
+package gateway
+
+// Mount for plugin-registered HTTP endpoints.
+//
+// The gateway owns the listener, the authentication, the role check and the
+// response framing; the plugin owns the logic. That split is the point: a plugin
+// cannot open a socket, cannot bypass auth, and cannot choose a path outside
+// /api/plugins/http/.
+//
+// Why this exists rather than "plugins can call each other": some things are
+// worth querying from OUTSIDE the browser — a CI job asking whether the upstream
+// balance is above zero, a monitoring probe, another tool on the box. billing
+// never wanted this (it only receives hooks); a quota plugin does. Hard-coding a
+// route per plugin would mean a gateway change for every such plugin, and an
+// arbitrary plugin's code in the main mux is not something the routing table
+// should ever grow.
+
+import (
+	"net/http"
+	"strconv"
+	"strings"
+
+	"llmsproxy/internal/lua"
+)
+
+// pluginHTTPPrefix is the mount point. It is deliberately narrow: a plugin's
+// routes live under one prefix, which makes them greppable, makes them easy to
+// revoke wholesale, and means a plugin cannot shadow /api/keys or /api/stats.
+const pluginHTTPPrefix = "/api/plugins/http/"
+
+// handlePluginHTTP dispatches a request to whichever plugin registered the
+// longest matching path.
+//
+// Longest-match rather than first-match: a plugin may register /health and
+// another /health/deep, and both must work. With registration order deciding it,
+// the outcome would depend on load order and change when a plugin is added.
+func (g *Gateway) handlePluginHTTP(w http.ResponseWriter, r *http.Request) {
+	rest := strings.TrimPrefix(r.URL.Path, pluginHTTPPrefix)
+	ps := g.core.Plugins()
+	if ps == nil {
+		writeError(w, http.StatusNotFound, "not_found", "no plugins loaded")
+		return
+	}
+
+	matches := ps.HTTPHandlers()
+	var chosen *lua.HTTPHandler
+	var params map[string]string
+	for i := range matches {
+		h := matches[i]
+		if p, ok := matchHandlerPath(h.Path, rest); ok {
+			if chosen == nil || segmentCount(h.Path) > segmentCount(chosen.Path) {
+				chosen, params = &h, p
+			}
+		}
+	}
+	if chosen == nil {
+		writeError(w, http.StatusNotFound, "not_found",
+			"no plugin serves this path")
+		return
+	}
+
+	if !strings.EqualFold(r.Method, chosen.Method) {
+		w.Header().Set("Allow", chosen.Method)
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed",
+			"this endpoint accepts "+chosen.Method+" only")
+		return
+	}
+
+	role := reqRole(r.Context())
+	if !chosen.Allows(role, reqKeyName(r.Context())) {
+		writeError(w, http.StatusForbidden, "forbidden",
+			handlerAccessHint(*chosen))
+		return
+	}
+
+	body, err := readBody(r)
+	if err != nil {
+		writeError(w, http.StatusBadRequest, "invalid_request", err.Error())
+		return
+	}
+	req := &lua.HandlerRequest{
+		Method:  r.Method,
+		Path:    rest,
+		Query:   flattenQuery(r.URL.Query()),
+		Params:  params,
+		Headers: flattenHeader(r.Header),
+		Body:    body,
+		Role:    role,
+		Key:     reqKey(r.Context()),
+	}
+	resp, err := ps.InvokeHandler(handlerOwner(chosen), chosen.Handler, req)
+	if err != nil {
+		// A plugin error is a 502, not a 500: the gateway is fine, the plugin is
+		// not, and the operator needs to tell those apart in their logs.
+		writeError(w, http.StatusBadGateway, "plugin_error", err.Error())
+		return
+	}
+	lua.WriteHandlerResponse(w, resp)
+}
+
+// handlerAccessHint explains, in the refusal, exactly which declaration would
+// open this endpoint. A 403 with no reason is the kind of thing an operator
+// opens the source to work out, and then the fix is a guess.
+func handlerAccessHint(h lua.HTTPHandler) string {
+	switch {
+	case len(h.Keys) > 0:
+		return "this endpoint allows only the named keys (" +
+			strings.Join(h.Keys, ", ") + "); authenticate with one of them"
+	case len(h.Roles) > 0:
+		return "this endpoint requires role " + strings.Join(h.Roles, " or ") +
+			" (widen it in the plugin's http declaration, or set public=true)"
+	default:
+		return "this endpoint declares no roles, so it is admin-only; " +
+			"set roles=… or keys=… in the plugin's http declaration"
+	}
+}
+
+// handlerOwner returns the plugin name that registered h. The gateway stores the
+// owner alongside the handler precisely so this does not have to be guessed from
+// the path prefix.
+func handlerOwner(h *lua.HTTPHandler) string { return h.Plugin }
+
+// segmentCount is how deep a route is, used for longest-match. A path with more
+// segments wins, so /health/deep beats /health when both are registered.
+func segmentCount(p string) int { return strings.Count(p, "/") }
+
+// matchHandlerPath matches a route pattern against a concrete path, capturing
+// ":name" segments.
+//
+// Returns ok=false on a segment-count mismatch or a missing value for a
+// param — a route must not match with an empty capture, or two different URLs
+// would reach the same handler indistinguishably.
+func matchHandlerPath(pattern, path string) (map[string]string, bool) {
+	pp := strings.Split(strings.Trim(pattern, "/"), "/")
+	sp := strings.Split(strings.Trim(path, "/"), "/")
+	if len(pp) != len(sp) {
+		return nil, false
+	}
+	var params map[string]string
+	for i := range pp {
+		if strings.HasPrefix(pp[i], ":") {
+			name := pp[i][1:]
+			if sp[i] == "" {
+				return nil, false
+			}
+			if params == nil {
+				params = map[string]string{}
+			}
+			params[name] = sp[i]
+			continue
+		}
+		if pp[i] != sp[i] {
+			return nil, false
+		}
+	}
+	return params, true
+}
+
+func flattenQuery(q map[string][]string) map[string]string {
+	out := make(map[string]string, len(q))
+	for k, v := range q {
+		if len(v) > 0 {
+			out[k] = v[0]
+		}
+	}
+	return out
+}
+
+func flattenHeader(h http.Header) map[string]string {
+	out := make(map[string]string, len(h))
+	for k := range h {
+		out[strings.ToLower(k)] = h.Get(k)
+	}
+	return out
+}
+
+func readBody(r *http.Request) (string, error) {
+	if r.Body == nil {
+		return "", nil
+	}
+	defer r.Body.Close()
+	// A cap, because the handler runs inside the gateway process: an unbounded
+	// body from an authenticated caller is a memory exhaustion vector.
+	const maxBody = 4 << 20
+	buf := make([]byte, 0, 4096)
+	tmp := make([]byte, 32*1024)
+	for {
+		n, err := r.Body.Read(tmp)
+		if n > 0 {
+			if len(buf)+n > maxBody {
+				return "", &bodyTooLarge{}
+			}
+			buf = append(buf, tmp[:n]...)
+		}
+		if err != nil {
+			break
+		}
+	}
+	return string(buf), nil
+}
+
+type bodyTooLarge struct{}
+
+func (bodyTooLarge) Error() string { return "request body exceeds 4 MiB" }
+
+// limitBody is kept for the numeric form some callers want in a log line.
+var _ = strconv.Itoa

internal/gateway/plugin_wiring_test.go

@@ -0,0 +1,626 @@
+package gateway
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+	"llmsproxy/internal/core"
+)
+
+// newFailingServer serves handler and returns its test server.
+func newFailingServer(h http.Handler) *httptest.Server { return httptest.NewServer(h) }
+
+// THE TEST THAT WAS MISSING.
+//
+// Every other plugin test calls Plugins.Fire() directly. That proves the Lua
+// runtime works, but it says nothing about whether the GATEWAY actually fires
+// the stages — the wiring in handleChat / singleChat / writeRec could be
+// deleted and the whole suite would stay green while no hook ever ran in
+// production.
+//
+// So these tests drive real HTTP requests through the real handler and assert
+// on what the plugin observed.
+
+// observed is what a spy plugin recorded. It is written from Lua, so the
+// counters live in plugin state and are read back through the state API.
+type spyPayload struct {
+	Stage  string
+	Type   string
+	Model  string
+	Source string
+	OK     bool
+	Tokens int64
+}
+
+// spyPlugin returns Lua source for a plugin that appends every stage it sees to
+// `state.seen`. It is generated per test so each test gets a clean log.
+func spyPlugin() string {
+	return `
+local plugin = { name = "spy", version = "1.0.0" }
+plugin.state = { seen = {} }
+plugin.hooks = {
+  request_start = "on_start",
+  routed = "on_routed",
+  request_end = "on_end",
+}
+local function rec(p, stage)
+  table.insert(plugin.state.seen, {
+    stage = stage, type = p.type or "", model = p.model or "",
+    source = p.source or "", ok = p.ok and true or false,
+    tokens = p.prompt_tokens or 0,
+  })
+end
+function plugin.on_start(p) rec(p, "request_start") return nil end
+function plugin.on_routed(p) rec(p, "routed") return nil end
+function plugin.on_end(p) rec(p, "request_end") return nil end
+return plugin
+`
+}
+
+// gatewayWithSpy boots a gateway with one mock upstream and the spy plugin
+// loaded, and returns the gateway plus the upstream URL.
+func gatewayWithSpy(t *testing.T) (*Gateway, string) {
+	t.Helper()
+	up := mockUpstream()
+	t.Cleanup(up.Close)
+
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	cfgBody := "listen: :0\n" +
+		"adapter_dir: " + filepath.Join(dir, "adapters") + "\n" +
+		"plugin_dir: " + filepath.Join(dir, "plugins") + "\n" +
+		"runtime_file: " + filepath.Join(dir, "runtime.json") + "\n" +
+		"gateway_keys:\n  - sk-test\n"
+	if err := os.WriteFile(cfgPath, []byte(cfgBody), 0600); err != nil {
+		t.Fatal(err)
+	}
+	cfg, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	cfg.Sources = []config.Source{{
+		Name:    "spyup",
+		BaseURL: up.URL,
+		Adapter: "openai",
+		APIKey:  "sk-x",
+		Models:  []config.Model{{ID: "spy-model", Kind: "chat"}},
+	}}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(c.Close)
+	if err := c.Plugins().LoadSource("spy", spyPlugin()); err != nil {
+		t.Fatalf("load spy: %v", err)
+	}
+	g, err := New(c)
+	if err != nil {
+		t.Fatalf("gateway: %v", err)
+	}
+	return g, up.URL
+}
+
+// seenStages reads the spy's log out of its published state.
+func seenStages(t *testing.T, g *Gateway) []map[string]interface{} {
+	t.Helper()
+	raw := g.core.Plugins().State("spy")
+	if raw == nil {
+		return nil
+	}
+	b, err := json.Marshal(raw)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var st struct {
+		Seen []map[string]interface{} `json:"seen"`
+	}
+	if err := json.Unmarshal(b, &st); err != nil {
+		t.Fatal(err)
+	}
+	return st.Seen
+}
+
+func stageNames(seen []map[string]interface{}) []string {
+	out := make([]string, 0, len(seen))
+	for _, s := range seen {
+		out = append(out, s["stage"].(string))
+	}
+	return out
+}
+
+// TestHooksFireOnRealDirectChat: a plain non-streaming chat through the real
+// handler must produce request_start, routed and request_end — in that order,
+// with the real source and model filled in.
+func TestHooksFireOnRealDirectChat(t *testing.T) {
+	g, _ := gatewayWithSpy(t)
+	rr := doReq(t, g, http.MethodPost, "/v1/chat/completions",
+		`{"model":"spy-model","messages":[{"role":"user","content":"hi"}]}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("chat status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	seen := seenStages(t, g)
+	got := stageNames(seen)
+	want := []string{"request_start", "routed", "request_end"}
+	if len(got) != 3 {
+		t.Fatalf("stages = %v, want %v (a hook that never fires is the failure "+
+			"this test exists for)", got, want)
+	}
+	for i := range want {
+		if got[i] != want[i] {
+			t.Fatalf("stage order = %v, want %v", got, want)
+		}
+	}
+	// The payload must be real data, not empty scaffolding.
+	end := seen[2]
+	if end["source"] != "spyup" {
+		t.Errorf("request_end source = %v, want spyup", end["source"])
+	}
+	if end["model"] != "spy-model" {
+		t.Errorf("request_end model = %v, want spy-model", end["model"])
+	}
+	if end["ok"] != true {
+		t.Errorf("request_end ok = %v, want true", end["ok"])
+	}
+	if toInt(t, end["tokens"]) <= 0 {
+		t.Errorf("request_end prompt_tokens = %v, want the mock upstream's real usage", end["tokens"])
+	}
+}
+
+// TestHooksFireOnRealStreamChat: streaming is a DIFFERENT code path (the record
+// is written from a deferred call after the stream ends), so it needs its own
+// assertion. A hook that only worked on the non-streaming path would be a real
+// bug for agent clients, which are the main streaming users.
+func TestHooksFireOnRealStreamChat(t *testing.T) {
+	g, _ := gatewayWithSpy(t)
+	rr := doReq(t, g, http.MethodPost, "/v1/chat/completions",
+		`{"model":"spy-model","stream":true,"messages":[{"role":"user","content":"hi"}]}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("stream status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	got := stageNames(seenStages(t, g))
+	if len(got) != 3 {
+		t.Fatalf("stream stages = %v, want all three", got)
+	}
+	if got[0] != "request_start" || got[2] != "request_end" {
+		t.Errorf("stream stage order = %v", got)
+	}
+}
+
+// TestHooksFireOnAutoRequest: the AUTO chain path is a third code path. It also
+// must fire, and routed must report tier -2 (resolved by the chain) rather than
+// the direct path's -1.
+func TestHooksFireOnAutoRequest(t *testing.T) {
+	g, _ := gatewayWithSpy(t)
+	// Register the model in the AUTO chain via the admin API.
+	put := doReq(t, g, http.MethodPut, "/api/auto", `{"rules":[{"model":"spy-model"}]}`)
+	if put.Code != http.StatusOK {
+		t.Fatalf("save auto: %d %s", put.Code, put.Body.String())
+	}
+	rr := doReq(t, g, http.MethodPost, "/v1/chat/completions",
+		`{"model":"AUTO","messages":[{"role":"user","content":"hi"}]}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("auto chat status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	seen := seenStages(t, g)
+	if len(seen) != 3 {
+		t.Fatalf("AUTO stages = %v, want three", stageNames(seen))
+	}
+	if seen[0]["model"] != "AUTO" {
+		t.Errorf("request_start model = %v, want the requested \"AUTO\"", seen[0]["model"])
+	}
+	if seen[1]["model"] != "spy-model" {
+		t.Errorf("routed model = %v, want the RESOLVED model", seen[1]["model"])
+	}
+}
+
+// TestHooksFireOnFailedRequest is the one that matters most for accounting: a
+// request that 500s upstream must still reach request_end, or a billing plugin
+// would never see failed traffic.
+func TestHooksFireOnFailedRequest(t *testing.T) {
+	// Point at an upstream that always fails.
+	bad := http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusInternalServerError)
+		_, _ = w.Write([]byte(`{"error":"boom"}`))
+	})
+	up := newFailingServer(bad)
+	defer up.Close()
+
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	cfgBody := "listen: :0\n" +
+		"adapter_dir: " + filepath.Join(dir, "adapters") + "\n" +
+		"plugin_dir: " + filepath.Join(dir, "plugins") + "\n" +
+		"runtime_file: " + filepath.Join(dir, "runtime.json") + "\n" +
+		"gateway_keys:\n  - sk-test\n"
+	os.WriteFile(cfgPath, []byte(cfgBody), 0600)
+	cfg, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	cfg.Sources = []config.Source{{
+		Name: "badup", BaseURL: up.URL, Adapter: "openai", APIKey: "sk-x",
+		Models: []config.Model{{ID: "bad-model", Kind: "chat"}},
+	}}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	defer c.Close()
+	if err := c.Plugins().LoadSource("spy", spyPlugin()); err != nil {
+		t.Fatal(err)
+	}
+	g, err := New(c)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	rr := doReq(t, g, http.MethodPost, "/v1/chat/completions",
+		`{"model":"bad-model","messages":[{"role":"user","content":"hi"}]}`)
+	if rr.Code == http.StatusOK {
+		t.Fatal("expected the upstream failure to surface as a non-200")
+	}
+	seen := seenStages(t, g)
+	// request_start must fire; routed must NOT (no source was ever selected);
+	// request_end MUST (otherwise billing misses failed traffic).
+	got := stageNames(seen)
+	if len(got) == 0 || got[0] != "request_start" {
+		t.Fatalf("stages = %v, want request_start first", got)
+	}
+	hasEnd := false
+	for _, s := range seen {
+		if s["stage"] == "request_end" {
+			hasEnd = true
+			if s["ok"] != false {
+				t.Error("request_end reported ok=true for a failed request")
+			}
+		}
+	}
+	if !hasEnd {
+		t.Error("a FAILED request never reached request_end; a billing plugin " +
+			"would not see it at all")
+	}
+	if contains(got, "routed") {
+		t.Error("routed fired for a request that never selected a source")
+	}
+}
+
+// TestBrokenPluginDoesNotBreakForwarding is the safety property, asserted
+// end-to-end: a plugin that throws in EVERY stage must not stop a chat from
+// being served.
+func TestBrokenPluginDoesNotBreakForwarding(t *testing.T) {
+	up := mockUpstream()
+	defer up.Close()
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	os.WriteFile(cfgPath, []byte("listen: :0\n"+
+		"adapter_dir: "+filepath.Join(dir, "adapters")+"\n"+
+		"plugin_dir: "+filepath.Join(dir, "plugins")+"\n"+
+		"runtime_file: "+filepath.Join(dir, "runtime.json")+"\n"+
+		"gateway_keys:\n  - sk-test\n"), 0600)
+	cfg, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	cfg.Sources = []config.Source{{
+		Name: "u", BaseURL: up.URL, Adapter: "openai", APIKey: "sk-x",
+		Models: []config.Model{{ID: "m", Kind: "chat"}},
+	}}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer c.Close()
+	broken := `
+local plugin = { name = "broken", version = "1.0.0" }
+plugin.hooks = { request_start = "f", routed = "f", request_end = "f" }
+function plugin.f(p) error("plugin is broken") end
+return plugin
+`
+	if err := c.Plugins().LoadSource("broken", broken); err != nil {
+		t.Fatalf("load broken: %v", err)
+	}
+	g, err := New(c)
+	if err != nil {
+		t.Fatal(err)
+	}
+	rr := doReq(t, g, http.MethodPost, "/v1/chat/completions",
+		`{"model":"m","messages":[{"role":"user","content":"hi"}]}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("a broken plugin broke forwarding: status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	if !strings.Contains(rr.Body.String(), "pong") {
+		t.Errorf("the response is not the upstream's: %s", rr.Body.String())
+	}
+	// And the failure must be visible rather than silent.
+	if len(c.Plugins().HookErrors()) == 0 {
+		t.Error("a plugin that threw in every stage left no error record")
+	}
+}
+
+// TestPluginListIsServedWithStages is the operator-facing check: an admin
+// debugging a plugin needs to see which stages are live and whether the hook
+// file is where the kernel thinks it is.
+func TestPluginListIsServedWithStages(t *testing.T) {
+	g, _ := gatewayWithSpy(t)
+	rr := doReq(t, g, http.MethodGet, "/api/plugins", "")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("status=%d", rr.Code)
+	}
+	for _, want := range []string{"spy", "request_start", "routed", "request_end", "plugin_dir"} {
+		if !strings.Contains(rr.Body.String(), want) {
+			t.Errorf("/api/plugins response lacks %q", want)
+		}
+	}
+}
+
+func toInt(t *testing.T, v interface{}) int64 {
+	t.Helper()
+	f, ok := v.(float64)
+	if !ok {
+		return 0
+	}
+	return int64(f)
+}
+
+func contains(list []string, s string) bool {
+	for _, x := range list {
+		if x == s {
+			return true
+		}
+	}
+	return false
+}
+
+// imageUpstream serves an OpenAI-shaped image response.
+func imageUpstream() *httptest.Server {
+	return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"created":1,"data":[{"b64_json":"aGk="}]}`))
+	}))
+}
+
+// TestHooksFireOnRealImageRequest: the image path is a THIRD entry point. It
+// has its own handler and its own scheduler call, so "chat works" says nothing
+// about it — and image traffic is billable, so a cost plugin that never sees
+// it under-reports.
+func TestHooksFireOnRealImageRequest(t *testing.T) {
+	up := imageUpstream()
+	defer up.Close()
+
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	os.WriteFile(cfgPath, []byte("listen: :0\n"+
+		"adapter_dir: "+filepath.Join(dir, "adapters")+"\n"+
+		"plugin_dir: "+filepath.Join(dir, "plugins")+"\n"+
+		"runtime_file: "+filepath.Join(dir, "runtime.json")+"\n"+
+		"gateway_keys:\n  - sk-test\n"), 0600)
+	cfg, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	cfg.Sources = []config.Source{{
+		Name: "imgup", BaseURL: up.URL, Adapter: "openai", APIKey: "sk-x",
+		Models: []config.Model{{ID: "img-model", Kind: "image"}},
+	}}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	defer c.Close()
+	if err := c.Plugins().LoadSource("spy", spyPlugin()); err != nil {
+		t.Fatal(err)
+	}
+	g, err := New(c)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	rr := doReq(t, g, http.MethodPost, "/v1/images/generations",
+		`{"model":"img-model","prompt":"a cat"}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("image status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	seen := seenStages(t, g)
+	got := stageNames(seen)
+	if len(got) != 3 {
+		t.Fatalf("image stages = %v, want all three (image is a separate handler)", got)
+	}
+	if seen[0]["type"] != "image" {
+		t.Errorf("request_start type = %v, want \"image\"", seen[0]["type"])
+	}
+	if seen[1]["source"] != "imgup" {
+		t.Errorf("routed source = %v, want imgup", seen[1]["source"])
+	}
+}
+
+// TestRejectedChatStillFiresRequestStart pins the ordering decision: request_start
+// fires BEFORE the quota/scope gates, so a plugin auditing traffic sees rejected
+// requests too. A plugin that only ever saw served requests could not compute a
+// true request rate.
+func TestRejectedChatStillFiresRequestStart(t *testing.T) {
+	up := mockUpstream()
+	defer up.Close()
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	os.WriteFile(cfgPath, []byte("listen: :0\n"+
+		"adapter_dir: "+filepath.Join(dir, "adapters")+"\n"+
+		"plugin_dir: "+filepath.Join(dir, "plugins")+"\n"+
+		"runtime_file: "+filepath.Join(dir, "runtime.json")+"\n"+
+		"gateway_keys:\n  - sk-test\n"), 0600)
+	cfg, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	cfg.Sources = []config.Source{{
+		Name: "u", BaseURL: up.URL, Adapter: "openai", APIKey: "sk-x",
+		Models: []config.Model{{ID: "m", Kind: "chat"}},
+	}}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer c.Close()
+	if err := c.Plugins().LoadSource("spy", spyPlugin()); err != nil {
+		t.Fatal(err)
+	}
+	g, err := New(c)
+	if err != nil {
+		t.Fatal(err)
+	}
+	// A model the gateway does not know => 404, rejected before any routing.
+	rr := doReq(t, g, http.MethodPost, "/v1/chat/completions",
+		`{"model":"no-such-model","messages":[{"role":"user","content":"hi"}]}`)
+	if rr.Code != http.StatusNotFound {
+		t.Fatalf("expected 404 for an unknown model, got %d", rr.Code)
+	}
+	got := stageNames(seenStages(t, g))
+	if !contains(got, "request_start") {
+		t.Errorf("stages = %v; a REJECTED request must still fire request_start so a "+
+			"plugin can count real traffic, not just served traffic", got)
+	}
+}
+
+// TestChainStepReachesPluginOnDegradation is the end-to-end proof for the
+// AUTO trace: a request that had to drop from tier 1 to tier 2 must be visible
+// to a plugin as a tier_skip followed by a selected, and request_end must carry
+// tier_served=2.
+//
+// Before the trace existed the plugin saw only tier=-2 ("resolved by the
+// chain") and could not tell a degradation from a clean tier-1 hit — which is
+// the whole question a priority chain exists to answer.
+func TestChainStepReachesPluginOnDegradation(t *testing.T) {
+	// tier 1's source always fails, so the walk must drop to tier 2.
+	bad := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.WriteHeader(http.StatusInternalServerError)
+		_, _ = w.Write([]byte(`{"error":"boom"}`))
+	}))
+	defer bad.Close()
+	good := mockUpstream()
+	defer good.Close()
+
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	os.WriteFile(cfgPath, []byte("listen: :0\n"+
+		"adapter_dir: "+filepath.Join(dir, "adapters")+"\n"+
+		"plugin_dir: "+filepath.Join(dir, "plugins")+"\n"+
+		"runtime_file: "+filepath.Join(dir, "runtime.json")+"\n"+
+		"gateway_keys:\n  - sk-test\n"), 0600)
+	cfg, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	cfg.Sources = []config.Source{
+		{Name: "t1", BaseURL: bad.URL, Adapter: "openai", APIKey: "sk-x",
+			Models: []config.Model{{ID: "hi-tier", Kind: "chat"}}},
+		{Name: "t2", BaseURL: good.URL, Adapter: "openai", APIKey: "sk-x",
+			Models: []config.Model{{ID: "lo-tier", Kind: "chat"}}},
+	}
+	if err := cfg.ApplyDefaults(); err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer c.Close()
+
+	// A spy that records chain_step events too.
+	sp := `
+local plugin = { name = "walker", version = "1.0.0" }
+plugin.state = { steps = {}, ends = {} }
+plugin.hooks = { chain_step = "step", request_end = "fin" }
+function plugin.step(p)
+  table.insert(plugin.state.steps, { kind = p.kind, tier = p.tier, source = p.source, model = p.model, reason = p.reason })
+  return nil
+end
+function plugin.fin(p)
+  plugin.state.ends[#plugin.state.ends + 1] = {
+    tier_served = p.tier_served, degraded = p.degraded,
+    walk = p.chain_walk, source = p.source, model = p.model,
+  }
+  return nil
+end
+return plugin
+`
+	if err := c.Plugins().LoadSource("walker", sp); err != nil {
+		t.Fatal(err)
+	}
+	g, err := New(c)
+	if err != nil {
+		t.Fatal(err)
+	}
+	// Two tiers, both in the chain.
+	if rr := doReq(t, g, http.MethodPut, "/api/auto",
+		`{"rules":[{"model":"hi-tier","source":"t1","tier":1},{"model":"lo-tier","source":"t2","tier":2}]}`); rr.Code != 200 {
+		t.Fatalf("save auto: %d %s", rr.Code, rr.Body.String())
+	}
+
+	rr := doReq(t, g, http.MethodPost, "/v1/chat/completions",
+		`{"model":"AUTO","messages":[{"role":"user","content":"hi"}]}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("chat = %d %s", rr.Code, rr.Body.String())
+	}
+
+	raw := c.Plugins().State("walker")
+	b, _ := json.Marshal(raw)
+	var st struct {
+		Steps []struct {
+			Kind   string `json:"kind"`
+			Tier   int    `json:"tier"`
+			Source string `json:"source"`
+			Model  string `json:"model"`
+		} `json:"steps"`
+		Ends []struct {
+			TierServed int    `json:"tier_served"`
+			Degraded   bool   `json:"degraded"`
+			Source     string `json:"source"`
+			Model      string `json:"model"`
+		} `json:"ends"`
+	}
+	if err := json.Unmarshal(b, &st); err != nil {
+		t.Fatalf("decode: %v (%s)", err, string(b))
+	}
+	if len(st.Steps) < 2 {
+		t.Fatalf("chain_step events = %+v, want at least a slot_fail and a selected", st.Steps)
+	}
+	if st.Steps[0].Kind != "slot_fail" || st.Steps[0].Tier != 1 {
+		t.Errorf("first step = %+v, want slot_fail on tier 1", st.Steps[0])
+	}
+	last := st.Steps[len(st.Steps)-1]
+	if last.Kind != "selected" || last.Tier != 2 {
+		t.Errorf("last step = %+v, want selected on tier 2", last)
+	}
+	if len(st.Ends) != 1 {
+		t.Fatalf("request_end count = %d, want 1", len(st.Ends))
+	}
+	if st.Ends[0].TierServed != 2 {
+		t.Errorf("tier_served = %d, want 2", st.Ends[0].TierServed)
+	}
+	if !st.Ends[0].Degraded {
+		t.Error("degraded = false, but the request dropped a tier")
+	}
+	if st.Ends[0].Model != "lo-tier" {
+		t.Errorf("served model = %q, want lo-tier", st.Ends[0].Model)
+	}
+}

internal/gateway/plugins_api.go

@@ -0,0 +1,619 @@
+package gateway
+
+import (
+	"encoding/json"
+	"errors"
+	"fmt"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+
+	"llmsproxy/internal/lua"
+)
+
+var (
+	errPluginName  = errors.New("plugin name must be non-empty and contain no path separator or dot")
+	errNoPluginDir = errors.New("no plugin_dir configured; set plugin_dir in config.yaml to enable plugins")
+)
+
+// Plugin management API.
+//
+//	GET    /api/plugins                  list loaded plugins + their hooks/UI/errors
+//	POST   /api/plugins                  upload/replace one plugin (.lua), hot-applied
+//	DELETE /api/plugins/{name}           remove a plugin
+//	GET    /api/plugins/{name}/state     the plugin's own published state
+//	PUT    /api/plugins/{name}/state     replace that state (admin only)
+//
+// Why state is a first-class endpoint: request_end is the LAST stage, so a
+// hook's return value has no downstream consumer inside the gateway. A plugin
+// that accumulates numbers (the billing plugin does exactly this) therefore
+// keeps them in its own Lua state and serves them here, which is what its UI
+// component fetches. This keeps plugin data clearly separated from the
+// gateway's own stats — see the accounting note in docs/plugins.md: the
+// gateway's quota accounting stays authoritative, a plugin only reports.
+
+// handlePluginUI serves the merged UI extensions the WebUI injects at boot.
+//
+// It is a single GET (not per-plugin) because the browser needs ALL extensions
+// before it can build the sidebar: a page contributed by one plugin and an
+// element contributed by another land in the same payload, and fetching them
+// separately would mean the sidebar has to be rebuilt as each arrives.
+//
+// Served to any authenticated role: the WebUI is authenticated before it asks,
+// and a plugin's own widgets need to render for the user whose key they show.
+func (g *Gateway) handlePluginUI(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodGet {
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "use GET")
+		return
+	}
+	ps := g.core.Plugins()
+	if ps == nil {
+		writeJSON(w, http.StatusOK, map[string]interface{}{"ui": lua.UIExtension{}})
+		return
+	}
+	// AllStages, not a hand-written list: an earlier version enumerated the
+	// three stages literally here, so when chain_step was added it was silently
+	// missing from this response — a plugin author reading the discovery
+	// document would have believed the stage did not exist. Deriving it from
+	// the one place that defines the order is the whole point of having it.
+	writeJSON(w, http.StatusOK, map[string]interface{}{
+		"ui":     ps.UI(),
+		"stages": pipelineStageNames(),
+	})
+}
+
+// stageNames returns the pipeline stage names in firing order, for the
+// discovery payload and for tests.
+func pipelineStageNames() []string {
+	out := make([]string, 0, len(lua.AllStages))
+	for _, s := range lua.AllStages {
+		out = append(out, string(s))
+	}
+	return out
+}
+
+func (g *Gateway) handlePluginsAPI(w http.ResponseWriter, r *http.Request) {
+	ps := g.core.Plugins()
+	if ps == nil {
+		writeJSON(w, http.StatusOK, map[string]interface{}{"plugins": []interface{}{}})
+		return
+	}
+	path := strings.TrimPrefix(r.URL.Path, "/api/plugins")
+	path = strings.Trim(path, "/")
+
+	// /api/plugins/{name}/state
+	if strings.HasSuffix(path, "/state") {
+		name := strings.TrimSuffix(path, "/state")
+		if name == "" {
+			writeError(w, http.StatusBadRequest, "invalid_request", "plugin name required")
+			return
+		}
+		g.handlePluginState(w, r, name)
+		return
+	}
+
+	// /api/plugins/{name}/http/{path}/access — the operator's view of, and edit
+	// surface for, which keys an endpoint answers.
+	if i := strings.Index(path, "/http/"); i >= 0 {
+		name := path[:i]
+		rest := path[i+len("/http/"):]
+		if strings.HasSuffix(rest, "/access") {
+			epPath := strings.TrimSuffix(rest, "/access")
+			g.handleEndpointAccess(w, r, name, epPath)
+			return
+		}
+	}
+
+	// /api/plugins/{name}/call/{fn}
+	if i := strings.Index(path, "/call/"); i >= 0 {
+		name := path[:i]
+		fn := path[i+len("/call/"):]
+		if name == "" || fn == "" {
+			writeError(w, http.StatusBadRequest, "invalid_request", "call needs a plugin and a function name")
+			return
+		}
+		g.handlePluginCall(w, r, name, fn)
+		return
+	}
+
+	// /api/plugins/{name}/fanout/{fn} — one call per argument set, in
+	// parallel, from one request. See handlePluginFanout for why the
+	// request-per-item form cannot be made concurrent from the client.
+	if i := strings.Index(path, "/fanout/"); i >= 0 {
+		name := path[:i]
+		fn := path[i+len("/fanout/"):]
+		if name == "" || fn == "" {
+			writeError(w, http.StatusBadRequest, "invalid_request", "fanout needs a plugin and a function name")
+			return
+		}
+		g.handlePluginFanout(w, r, name, fn)
+		return
+	}
+
+	if reqRole(r.Context()) != "admin" {
+		writeError(w, http.StatusForbidden, "forbidden", "admin role required")
+		return
+	}
+
+	if path == "" {
+		switch r.Method {
+		case http.MethodGet:
+			writeJSON(w, http.StatusOK, map[string]interface{}{
+				"plugins":     ps.List(),
+				"on_disk":     ps.OnDisk(),
+				"hook_errors": ps.HookErrors(),
+				"plugin_dir":  g.pluginDir(),
+				// Visibility overrides and the endpoint access table, so the
+				// plugin page renders what is IN FORCE rather than making the
+				// operator read .lua files or config.yaml to find out.
+				"visibility":  ps.VisibilityOverrides(),
+				"http_access": ps.EndpointAccess(),
+			})
+		case http.MethodPost:
+			var body struct {
+				Name string `json:"name"`
+				Code string `json:"code"`
+			}
+			if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+				writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+				return
+			}
+			if err := g.installPlugin(body.Name, body.Code); err != nil {
+				writeError(w, http.StatusBadRequest, "plugin_error", err.Error())
+				return
+			}
+			writeJSON(w, http.StatusOK, map[string]interface{}{"ok": true, "name": body.Name})
+		default:
+			writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "")
+		}
+		return
+	}
+
+	// /api/plugins/{name} — source read (GET), enable/disable (PUT), delete.
+	// Multi-segment paths (…/state) are handled earlier and never reach here.
+	if !strings.Contains(path, "/") {
+		switch r.Method {
+		case http.MethodGet:
+			// Reading the SOURCE (not the runtime state) is what an editor
+			// needs; the state endpoint is /state and returns accumulated data
+			// instead. Mixing the two would make a "save what I read"
+			// round-trip impossible.
+			code, err := g.readPluginSource(path)
+			if err != nil {
+				writeError(w, http.StatusNotFound, "not_found", err.Error())
+				return
+			}
+			writeJSON(w, http.StatusOK, map[string]interface{}{
+				"name": path, "code": code,
+				"enabled": g.core.Plugins().Enabled(path),
+			})
+		case http.MethodPut:
+			// Enable / disable. The intent rides in the body rather than the
+			// verb, because "enable" and "disable" are the same resource and
+			// PUT /{name} is the one the docs advertise.
+			var body struct {
+				Enabled *bool `json:"enabled"`
+			}
+			if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+				writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+				return
+			}
+			if body.Enabled == nil {
+				writeError(w, http.StatusBadRequest, "invalid_request", `body must be {"enabled":true|false}`)
+				return
+			}
+			if err := g.core.Plugins().SetEnabled(path, *body.Enabled); err != nil {
+				writeError(w, http.StatusBadRequest, "plugin_error", err.Error())
+				return
+			}
+			writeJSON(w, http.StatusOK, map[string]interface{}{
+				"ok": true, "name": path, "enabled": *body.Enabled,
+			})
+		case http.MethodDelete:
+			if err := g.removePlugin(path); err != nil {
+				writeError(w, http.StatusBadRequest, "plugin_error", err.Error())
+				return
+			}
+			writeJSON(w, http.StatusOK, map[string]interface{}{"ok": true})
+		default:
+			writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "")
+		}
+		return
+	}
+
+	writeError(w, http.StatusNotFound, "not_found", "unknown plugin sub-resource: "+path)
+}
+
+// readPluginSource returns a plugin's file contents for the editor.
+func (g *Gateway) readPluginSource(name string) (string, error) {
+	if err := validPluginName(name); err != nil {
+		return "", err
+	}
+	if g.pluginDir() == "" {
+		return "", errNoPluginDir
+	}
+	b, err := os.ReadFile(filepath.Join(g.pluginDir(), name+".lua"))
+	if err != nil {
+		return "", fmt.Errorf("no plugin source named %q", name)
+	}
+	return string(b), nil
+}
+
+// validPluginName rejects anything that could escape the plugin directory.
+// Shared by install / remove / read so the check cannot drift between them.
+func validPluginName(name string) error {
+	if name == "" || strings.ContainsAny(name, `/\.`) {
+		return errPluginName
+	}
+	return nil
+}
+
+// handlePluginState serves GET (read state) and PUT (replace state).
+func (g *Gateway) handlePluginState(w http.ResponseWriter, r *http.Request, name string) {
+	ps := g.core.Plugins()
+	switch r.Method {
+	case http.MethodGet:
+		// The plugin's audience gates its DATA as well as its page. Hiding the UI
+		// while leaving GET /state open would put the numbers one curl away, which
+		// is the leak the audience field exists to close — and billing's state is
+		// a global ledger, not the caller's own usage.
+		//
+		// A plugin that declares no audience is unchanged (every role may read
+		// it), so a single-purpose widget keeps working without edits.
+		role := reqRole(r.Context())
+		if !ps.StateVisibleTo(name, role) {
+			writeError(w, http.StatusForbidden, "forbidden",
+				"plugin "+name+" is not visible to role "+role)
+			return
+		}
+		// prices rides along: the billing rules editor must show the table
+		// that is actually in effect, not a reconstruction from config. It is
+		// absent for plugins with no prices (every plugin but billing).
+		state := ps.State(name)
+		// A plugin may SPLIT its state: global buckets stay deployment-wide while
+		// per-caller buckets are narrowed to the authenticated key. Role
+		// visibility alone cannot express this — the same role is allowed to read
+		// the plugin, just not another caller's row.
+		if scope := ps.StateScopeFor(name); scope != nil && role != "admin" {
+			state = splitStateForKey(state, scope, reqKey(r.Context()))
+		}
+		payload := map[string]interface{}{
+			"plugin": name,
+			"state":  state,
+		}
+		if pr := ps.Prices(name); pr != nil {
+			payload["prices"] = pr
+		}
+		writeJSON(w, http.StatusOK, payload)
+	case http.MethodPut:
+		if reqRole(r.Context()) != "admin" {
+			writeError(w, http.StatusForbidden, "forbidden", "admin role required")
+			return
+		}
+		var state interface{}
+		if err := json.NewDecoder(r.Body).Decode(&state); err != nil {
+			writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+			return
+		}
+		if err := ps.SetState(name, state); err != nil {
+			writeError(w, http.StatusBadRequest, "plugin_error", err.Error())
+			return
+		}
+		writeJSON(w, http.StatusOK, map[string]interface{}{"ok": true})
+	default:
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "")
+	}
+}
+
+// Call invokes a named function on a plugin's Lua state and returns its result.
+//
+// This exists because plugin pages otherwise have NO way to reach their own
+// plugin code: pluginAPI offers state read/write and a couple of callbacks, and
+// that is enough for billing (whose numbers arrive on hooks during request
+// forwarding) but useless for a plugin that must do work on demand. A quota
+// page has to ask the vendor when the operator clicks Refresh, and no hook
+// fires at that moment.
+//
+// It is deliberately narrow:
+//   - The function must be one the plugin explicitly lists in plugin.exports.
+//     Anything else is refused, so this endpoint cannot become a generic "run
+//     arbitrary Lua with my credentials" oracle over a plugin's private state.
+//   - The caller must be admin, and the result is JSON — no Lua value is ever
+//     handed back to Go as a live closure.
+//   - Errors come back as (nil, message) from the plugin, matching every other
+//     plugin entry point, and surface as a 502 rather than raising.
+func (g *Gateway) handlePluginCall(w http.ResponseWriter, r *http.Request, name, fn string) {
+	if reqRole(r.Context()) != "admin" {
+		writeError(w, http.StatusForbidden, "forbidden", "admin role required")
+		return
+	}
+	if r.Method != http.MethodPost {
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "")
+		return
+	}
+	var args map[string]interface{}
+	if err := json.NewDecoder(r.Body).Decode(&args); err != nil {
+		writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+		return
+	}
+	result, err := g.core.Plugins().Call(name, fn, args)
+	if err != nil {
+		writeError(w, http.StatusBadGateway, "plugin_error", err.Error())
+		return
+	}
+	writeJSON(w, http.StatusOK, map[string]interface{}{"result": result})
+}
+
+// handlePluginFanout runs one plugin export per element of a list, in parallel,
+// from a SINGLE request.
+//
+// Why this exists: the quota page has twenty cards, and it used to fire twenty
+// browser requests at /call/. They arrived concurrently and then serialised
+// inside Plugins.Call, which holds the plugin mutex across the Lua function —
+// including its network wait. Measured on the live gateway: 20 probes took 29s of
+// wall clock while a single probe with no contention took 1.4–3.3s, so the page
+// sat on "probing…" for half a minute. The browser-side concurrency was real and
+// made no difference.
+//
+// A fan-out cannot be assembled from the client either: any request-per-item
+// arrangement is exactly the shape that queues. The parallelism has to happen
+// where the fan-out is received.
+func (g *Gateway) handlePluginFanout(w http.ResponseWriter, r *http.Request, name, fn string) {
+	if reqRole(r.Context()) != "admin" {
+		writeError(w, http.StatusForbidden, "forbidden", "admin role required")
+		return
+	}
+	if r.Method != http.MethodPost {
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "")
+		return
+	}
+	var req struct {
+		// Args is a LIST of argument objects, one per call. A bare object is
+		// rejected rather than coerced: "args" meaning "one call" and "args"
+		// meaning "many" depending on its shape is how a caller ends up probing
+		// one source and believing it probed twenty.
+		Args []map[string]interface{} `json:"args"`
+	}
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+		return
+	}
+	if len(req.Args) == 0 {
+		writeError(w, http.StatusBadRequest, "invalid_request", "args must be a non-empty list of argument objects")
+		return
+	}
+	// The page legitimately has one card per configured source. The bound is not
+	// arbitrary: it stops one request from asking for thousands of VMs, and it is
+	// enforced before any work starts rather than discovered shard by shard.
+	if len(req.Args) > fanoutMaxArgs {
+		writeError(w, http.StatusBadRequest, "invalid_request",
+			fmt.Sprintf("args has %d entries, limit is %d", len(req.Args), fanoutMaxArgs))
+		return
+	}
+	results, err := g.core.Plugins().ShardCall(name, fn, req.Args)
+	if err != nil {
+		writeError(w, http.StatusBadGateway, "plugin_error", err.Error())
+		return
+	}
+	// A shard that failed contributes a null in its slot. That is deliberate and
+	// is NOT the same as the whole request failing: one unreachable upstream must
+	// not turn a page of twenty cards into a single error.
+	writeJSON(w, http.StatusOK, map[string]interface{}{"results": results})
+}
+
+// fanoutMaxArgs bounds one fan-out request. Twenty is the measured real source
+// count on the gateway this was built for; the headroom is for operators with
+// more sources configured.
+const fanoutMaxArgs = 64
+
+func (g *Gateway) pluginDir() string {
+	if c := g.core.Config(); c != nil {
+		return c.PluginDir
+	}
+	return ""
+}
+
+// installPlugin writes a plugin to disk and hot-loads it. A syntax error is
+// returned to the caller AND the file is left on disk so the operator can fix
+// it, matching how adapters behave (the file is authoritative once present).
+func (g *Gateway) installPlugin(name, code string) error {
+	if err := validPluginName(name); err != nil {
+		return err
+	}
+	dir := g.pluginDir()
+	if dir == "" {
+		return errNoPluginDir
+	}
+	if err := os.MkdirAll(dir, 0755); err != nil {
+		return err
+	}
+	if err := os.WriteFile(filepath.Join(dir, name+".lua"), []byte(code), 0644); err != nil {
+		return err
+	}
+	return g.core.Plugins().LoadSource(name, code)
+}
+
+func (g *Gateway) removePlugin(name string) error {
+	if err := validPluginName(name); err != nil {
+		return err
+	}
+	if g.pluginDir() == "" {
+		return errNoPluginDir
+	}
+	_ = os.Remove(filepath.Join(g.pluginDir(), name+".lua"))
+	return g.core.Plugins().Unload(name)
+}
+
+// handleEndpointAccess serves the operator's view and edit surface for one
+// plugin endpoint's reachable keys.
+//
+//	GET    …/http/{path}/access   the declared rules plus any override
+//	PUT    …/http/{path}/access   body {"access": "admin,@homeagent"}
+//	DELETE …/http/{path}/access   clear the override
+//
+// The spec string is a comma list: a bare word is a role, "@name" is a gateway
+// key's configured name, and the literal "public" opens it to any
+// authenticated key. That is the same grammar the visibility override uses, and
+// keeping one grammar means an operator learns it once.
+//
+// Key names are validated against the configured keys BEFORE being stored. An
+// allowlist naming a key that does not exist is a silent no-op — every caller
+// is refused — and the operator would have no way to tell that apart from the
+// endpoint simply working.
+func (g *Gateway) handleEndpointAccess(w http.ResponseWriter, r *http.Request, name, epPath string) {
+	if reqRole(r.Context()) != "admin" {
+		writeError(w, http.StatusForbidden, "forbidden", "admin role required")
+		return
+	}
+	ps := g.core.Plugins()
+
+	// epPath "*" is the WHOLE-PLUGIN visibility override, not an endpoint. It
+	// shares this route (and therefore this store) on purpose: an operator
+	// setting "who can see this plugin" and "who can call this endpoint" is
+	// asking the same question at two granularities, and two stores would let
+	// them disagree. SetEndpointAccess cannot serve it — it validates that the
+	// endpoint exists, which "*" by definition does not.
+	if epPath == pluginWholeOverride {
+		g.handlePluginVisibility(w, r, name)
+		return
+	}
+
+	switch r.Method {
+	case http.MethodGet:
+		for _, e := range ps.EndpointAccess() {
+			if e.Plugin == name && e.Path == epPath {
+				writeJSON(w, http.StatusOK, map[string]interface{}{"endpoint": e})
+				return
+			}
+		}
+		writeError(w, http.StatusNotFound, "not_found", "no such endpoint")
+	case http.MethodPut, http.MethodDelete:
+		spec := ""
+		if r.Method == http.MethodPut {
+			var body struct {
+				Access string `json:"access"`
+			}
+			if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+				writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+				return
+			}
+			spec = body.Access
+		}
+		if err := ps.ValidateAccessSpec(spec); err != nil {
+			writeError(w, http.StatusBadRequest, "invalid_request", err.Error())
+			return
+		}
+		if err := ps.SetEndpointAccess(name, epPath, spec); err != nil {
+			writeError(w, http.StatusBadRequest, "invalid_request", err.Error())
+			return
+		}
+		writeJSON(w, http.StatusOK, map[string]interface{}{"ok": true, "access": spec})
+	default:
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", "")
+	}
+}
+
+// splitStateForKey rewrites a plugin state so that a non-admin caller sees only
+// the buckets that belong to them.
+//
+// The plugin declares which buckets are global and which are per-key. A per-key
+// bucket is keyed by the MASKED key ("***a1b2c3" — the same value the audit log
+// uses), so the caller's own row is found by masking the presented secret and
+// matching the last six characters.
+//
+// Keys that do not match are DROPPED, not zeroed: a zeroed row would render as
+// "you spent nothing" and, worse, would leak that a row exists for someone else.
+func splitStateForKey(state interface{}, scope *lua.StateScope, key string) interface{} {
+	m, ok := state.(map[string]interface{})
+	if !ok || scope == nil || key == "" {
+		return state
+	}
+	masked := maskKey(key)
+	out := make(map[string]interface{}, len(m))
+	for k, v := range m {
+		if !scope.IsGlobal(k) {
+			// Per-key bucket: keep only the caller's rows, at every nesting level
+			// the scope mentions (by_day_key is {day: {masked: bucket}}).
+			out[k] = filterKeyBuckets(v, masked, scope.DepthOf(k))
+			continue
+		}
+		out[k] = v
+	}
+	return out
+}
+
+// filterKeyBuckets narrows a keyed structure to one masked key.
+//
+// depth 1 is {masked: bucket}; depth 2 is {day: {masked: bucket}}. The level is
+// declared rather than sniffed, because both shapes are dicts and sniffing would
+// confuse an empty bucket with a nested one.
+func filterKeyBuckets(v interface{}, masked string, depth int) interface{} {
+	switch depth {
+	case 0:
+		return v // not a per-key bucket; leave alone
+	case 1:
+		m, ok := v.(map[string]interface{})
+		if !ok {
+			return v
+		}
+		if row, ok := m[masked]; ok {
+			return map[string]interface{}{masked: row}
+		}
+		return map[string]interface{}{}
+	default:
+		m, ok := v.(map[string]interface{})
+		if !ok {
+			return v
+		}
+		out := make(map[string]interface{}, len(m))
+		for day, inner := range m {
+			im, ok := inner.(map[string]interface{})
+			if !ok {
+				continue
+			}
+			if row, ok := im[masked]; ok {
+				out[day] = map[string]interface{}{masked: row}
+			}
+		}
+		return out
+	}
+}
+
+// pluginWholeOverride is the epPath that addresses a plugin's overall visibility
+// rather than one of its endpoints.
+const pluginWholeOverride = "*"
+
+// handlePluginVisibility is the whole-plugin half of the access API. The
+// audience string uses the same vocabulary as an endpoint spec (roles, @key
+// names, public) so the UI can drive both from one input.
+func (g *Gateway) handlePluginVisibility(w http.ResponseWriter, r *http.Request, name string) {
+	ps := g.core.Plugins()
+	switch r.Method {
+	case http.MethodGet:
+		writeJSON(w, http.StatusOK, map[string]interface{}{
+			"plugin":     name,
+			"visibility": ps.VisibilityOverrides()[name],
+		})
+	case http.MethodPut, http.MethodDelete:
+		spec := ""
+		if r.Method == http.MethodPut {
+			var body struct {
+				Access string `json:"access"`
+			}
+			if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+				writeError(w, http.StatusBadRequest, "invalid_request", "invalid json: "+err.Error())
+				return
+			}
+			spec = body.Access
+		}
+		if err := ps.ValidateAccessSpec(spec); err != nil {
+			writeError(w, http.StatusBadRequest, "invalid_request", err.Error())
+			return
+		}
+		ps.SetPluginVisibility(name, spec)
+		writeJSON(w, http.StatusOK, map[string]interface{}{"ok": true, "visibility": spec})
+	default:
+		writeError(w, http.StatusMethodNotAllowed, "method_not_allowed", r.Method+" is not allowed here")
+	}
+}

internal/gateway/plugins_api_test.go

@@ -0,0 +1,263 @@
+package gateway
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+	"llmsproxy/internal/core"
+	"llmsproxy/internal/lua"
+)
+
+// gatewayWithBilling boots a gateway with the bundled billing plugin loaded, so
+// the UI-injection endpoint is exercised against a real plugin rather than a
+// hand-written stub. The other plugin tests in this package assert on the
+// WebUI source; this one asserts on the HTTP contract the browser consumes.
+func gatewayWithBilling(t *testing.T) *Gateway {
+	t.Helper()
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	body := "listen: :0\n" +
+		"adapter_dir: " + filepath.Join(dir, "adapters") + "\n" +
+		"plugin_dir: " + filepath.Join(dir, "plugins") + "\n" +
+		"runtime_file: " + filepath.Join(dir, "runtime.json") + "\n" +
+		"gateway_keys:\n  - sk-test\n"
+	if err := os.WriteFile(cfgPath, []byte(body), 0600); err != nil {
+		t.Fatal(err)
+	}
+	cfg, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	c, err := core.NewFromConfig(cfg)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(c.Close)
+	// Load the shipped plugin explicitly: seeding only runs for a directory that
+	// does not exist yet, and this test wants a known plugin regardless.
+	src, err := lua.ReadBundledPlugin("billing")
+	if err != nil {
+		t.Fatalf("read bundled billing: %v", err)
+	}
+	if err := c.Plugins().LoadSource("billing", src); err != nil {
+		t.Fatalf("load billing: %v", err)
+	}
+	g, err := New(c)
+	if err != nil {
+		t.Fatalf("gateway: %v", err)
+	}
+	return g
+}
+
+// TestUIInjectServesPluginUI: GET /api/ui-inject is the single call the WebUI
+// makes at boot, and it must carry BOTH a contributed page and contributed
+// elements — the browser builds the sidebar from the page and mounts the
+// elements into existing panes from the same payload, so a partial response
+// would produce a page with no body or a missing tile.
+func TestUIInjectServesPluginUI(t *testing.T) {
+	g := gatewayWithBilling(t)
+	rr := doReq(t, g, http.MethodGet, "/api/ui-inject", "")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	var view struct {
+		// Decoded into the real types so the test cannot drift from the wire
+		// contract. An inline copy missed the `pages` field when the payload
+		// shape changed and failed to compile, which is at least loud — but the
+		// same copy also went on asserting the OLD single-page shape for a
+		// release, silently.
+		UI     lua.UIExtension `json:"ui"`
+		Stages []string        `json:"stages"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &view); err != nil {
+		t.Fatalf("decode: %v", err)
+	}
+	// The payload carries every page in one list; ui.page is no longer a
+	// separate slot (it was, and a second plugin contributing a page overwrote
+	// whatever was there).
+	var billing *lua.UIPage
+	for i, pg := range view.UI.Pages {
+		if pg != nil && pg.PageID == "billing" {
+			billing = view.UI.Pages[i]
+		}
+	}
+	if billing == nil {
+		t.Fatalf("no billing page in the inject payload; pages=%d", len(view.UI.Pages))
+	}
+	if !strings.Contains(billing.Mount, "billing-root") {
+		t.Error("the page mount came back empty")
+	}
+	if len(view.UI.Elements) == 0 {
+		t.Error("billing contributes an element to the status page but it is missing")
+	}
+	for _, e := range view.UI.Elements {
+		if e.Target != "status" {
+			t.Errorf("element target = %q, want \"status\"", e.Target)
+		}
+	}
+	// Derived from AllStages, not hardcoded: the previous assertion of "3"
+	// is exactly what let chain_step go missing from this payload unnoticed.
+	if len(view.Stages) != len(lua.AllStages) {
+		t.Errorf("stages = %v, want %d (one per AllStages entry)", view.Stages, len(lua.AllStages))
+	}
+	for i, st := range lua.AllStages {
+		if i >= len(view.Stages) || view.Stages[i] != string(st) {
+			t.Errorf("stages[%d] = %v, want %q", i, view.Stages, string(st))
+		}
+	}
+	if !containsStr(view.Stages, string(lua.StageChainStep)) {
+		t.Error("the discovery payload does not advertise chain_step; a plugin " +
+			"author would conclude the stage does not exist")
+	}
+}
+
+// TestUIInjectIsEmptyWithoutPlugins: a gateway with no plugins must still answer
+// 200 with an empty (not missing, not null) payload. The WebUI calls this
+// unconditionally at boot, so a 404 or a null `ui` would break every dashboard.
+func TestUIInjectIsEmptyWithoutPlugins(t *testing.T) {
+	g := newTestGateway(t)
+	rr := doReq(t, g, http.MethodGet, "/api/ui-inject", "")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	if !strings.Contains(rr.Body.String(), `"ui"`) {
+		t.Error("no ui key in the response; the WebUI would have nothing to read")
+	}
+}
+
+// containsStr reports whether list has s.
+func containsStr(list []string, s string) bool {
+	for _, x := range list {
+		if x == s {
+			return true
+		}
+	}
+	return false
+}
+
+// newRecorderFor pushes a request through the full handler chain.
+func newRecorderFor(t *testing.T, g *Gateway, req *http.Request) *httptest.ResponseRecorder {
+	t.Helper()
+	rr := httptest.NewRecorder()
+	g.Handler().ServeHTTP(rr, req)
+	return rr
+}
+
+// TestPluginsListAndStateAPI covers the management surface the plugin docs
+// promise: listing, and reading a plugin's own published state.
+func TestPluginsListAndStateAPI(t *testing.T) {
+	g := gatewayWithBilling(t)
+	rr := doReq(t, g, http.MethodGet, "/api/plugins", "")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("GET /api/plugins = %d: %s", rr.Code, rr.Body.String())
+	}
+	for _, want := range []string{"billing", "hook_errors", "plugin_dir", "request_end"} {
+		if !strings.Contains(rr.Body.String(), want) {
+			t.Errorf("/api/plugins response lacks %q", want)
+		}
+	}
+	// state read: the plugin published its (empty) state, so the key exists.
+	rr = doReq(t, g, http.MethodGet, "/api/plugins/billing/state", "")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("GET state = %d: %s", rr.Code, rr.Body.String())
+	}
+	if !strings.Contains(rr.Body.String(), `"total"`) {
+		t.Errorf("billing state lacks the total bucket: %s", rr.Body.String())
+	}
+}
+
+// TestPluginStatePUTIsAdminOnly: only the WRITE side is gated. A user key must
+// be able to READ its own billing widget's data, but must not be able to
+// rewrite the price table.
+func TestPluginStatePUTIsAdminOnly(t *testing.T) {
+	g := gatewayWithBilling(t)
+
+	// Build a user-role key and remember its secret.
+	rec, err := g.core.CreateKey("viewer", "user", nil, "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	userKey := rec.Key
+
+	// A user key may read the state.
+	req, _ := http.NewRequest(http.MethodGet, "/api/plugins/billing/state", nil)
+	req.Header.Set("Authorization", "Bearer "+userKey)
+	rr := newRecorderFor(t, g, req)
+	if rr.Code != http.StatusOK {
+		t.Errorf("user GET state = %d, want 200 (the billing widget must render for users)", rr.Code)
+	}
+
+	// A user key may NOT write it.
+	put, _ := http.NewRequest(http.MethodPut, "/api/plugins/billing/state",
+		strings.NewReader(`{"prices":{"default":{"prompt":0}}}`))
+	put.Header.Set("Authorization", "Bearer "+userKey)
+	put.Header.Set("Content-Type", "application/json")
+	prr := newRecorderFor(t, g, put)
+	if prr.Code != http.StatusForbidden {
+		t.Errorf("user PUT state = %d, want 403 (a user must not rewrite the price table)", prr.Code)
+	}
+
+	// An admin key may.
+	adm := doReq(t, g, http.MethodPut, "/api/plugins/billing/state",
+		`{"prices":{"default":{"prompt":1e-6}}}`)
+	if adm.Code != http.StatusOK {
+		t.Errorf("admin PUT state = %d: %s", adm.Code, adm.Body.String())
+	}
+}
+
+// TestPluginsListCarriesAccessTables: the plugin page renders visibility
+// overrides and the endpoint access table from /api/plugins. Without them in the
+// response the operator has to open a .lua or config.yaml to learn who can see
+// what, which is exactly what the UI is meant to avoid.
+func TestPluginsListCarriesAccessTables(t *testing.T) {
+	g := gatewayWithCallPlugin(t)
+	rr := doReq(t, g, http.MethodGet, "/api/plugins", "")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("GET /api/plugins = %d", rr.Code)
+	}
+	var out struct {
+		Visibility map[string]string     `json:"visibility"`
+		HTTPAccess []map[string]interface{} `json:"http_access"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &out); err != nil {
+		t.Fatalf("decode: %v (body=%s)", err, rr.Body.String())
+	}
+	// The calltest plugin registers no endpoints, so the table is present but
+	// empty — the KEY is that the field exists at all. A missing field decodes to
+	// nil and the UI silently renders nothing, which looks identical to "this
+	// plugin has no endpoints".
+	if !strings.Contains(rr.Body.String(), "http_access") {
+		t.Error("/api/plugins response lacks the http_access field")
+	}
+	if !strings.Contains(rr.Body.String(), "visibility") {
+		t.Error("/api/plugins response lacks the visibility field")
+	}
+}
+
+// TestPluginWholeVisibilityViaAccessAPI covers the "*" path: the plugin page
+// edits a plugin's overall audience through the same route it edits endpoints
+// with. Before this existed the UI had a visibility selector whose save 404'd,
+// because SetEndpointAccess validates that the endpoint exists and "*" does not.
+func TestPluginWholeVisibilityViaAccessAPI(t *testing.T) {
+	g := gatewayWithCallPlugin(t)
+	rr := doReq(t, g, http.MethodPut, "/api/plugins/calltest/http/*/access",
+		`{"access":"admin, user"}`)
+	if rr.Code != http.StatusOK {
+		t.Fatalf("PUT */access = %d: %s", rr.Code, rr.Body.String())
+	}
+	if !strings.Contains(rr.Body.String(), `"ok":true`) {
+		t.Errorf("unexpected body: %s", rr.Body.String())
+	}
+	// A user role is a legitimate audience; a nonsense one is not.
+	bad := doReq(t, g, http.MethodPut, "/api/plugins/calltest/http/*/access",
+		`{"access":"../../etc"}`)
+	if bad.Code != http.StatusBadRequest {
+		t.Errorf("traversal-shaped audience = %d, want 400: %s", bad.Code, bad.Body.String())
+	}
+}

internal/gateway/server.go

@@ -115,7 +115,7 @@ func (g *Gateway) loginRecord(ip string, ok bool, now int64) {
 	}
 }
 
-func New(c *core.Core, gatewayKeys []string) (*Gateway, error) {
+func New(c *core.Core) (*Gateway, error) {
 	sub, err := fs.Sub(uiFS, "ui")
 	if err != nil {
 		return nil, err
@@ -222,10 +222,30 @@ func (g *Gateway) routes(w http.ResponseWriter, r *http.Request) {
 		g.handleModels(w, r)
 	case r.URL.Path == "/api/adapters" || strings.HasPrefix(r.URL.Path, "/api/adapters/"):
 		g.handleAdaptersAPI(w, r)
+	case r.URL.Path == "/api/v1" || strings.HasPrefix(r.URL.Path, "/api/v1/"):
+		g.apiV1Routes(w, r)
+	// Must precede the /api/sources prefix match: a generic HasPrefix on
+	// "/api/sources/" would otherwise swallow "/api/sources/discover" and the
+	// operator would get "source name required in path" instead of a catalog.
+	case r.URL.Path == "/api/sources/discover" || strings.HasPrefix(r.URL.Path, "/api/sources/discover/"):
+		g.handleSourceDiscoverAPI(w, r)
 	case r.URL.Path == "/api/sources" || strings.HasPrefix(r.URL.Path, "/api/sources/"):
 		g.handleSourcesAPI(w, r)
 	case r.URL.Path == "/api/source_templates" || strings.HasPrefix(r.URL.Path, "/api/source_templates/"):
 		g.handleSourceTemplatesAPI(w, r)
+	case r.URL.Path == "/api/ui-inject" || strings.HasPrefix(r.URL.Path, "/api/ui-inject/"):
+		g.handlePluginUI(w, r)
+	case r.URL.Path == "/api/plugins/billing/rules":
+		// Checked BEFORE the generic /api/plugins/ prefix: the plugin router
+		// treats anything after the name as a plugin sub-resource and would
+		// 404 on "rules" instead of reaching the billing editor.
+		g.handleBillingRules(w, r)
+	case strings.HasPrefix(r.URL.Path, pluginHTTPPrefix):
+		// Before the generic plugin router: these paths belong to a PLUGIN, not to
+		// the plugin manager, and "http" is a valid sub-resource name otherwise.
+		g.handlePluginHTTP(w, r)
+	case r.URL.Path == "/api/plugins" || strings.HasPrefix(r.URL.Path, "/api/plugins/"):
+		g.handlePluginsAPI(w, r)
 	case r.URL.Path == "/api/chat":
 		g.handleChat(w, r)
 	case r.URL.Path == "/api/status":
@@ -294,20 +314,43 @@ func (g *Gateway) auth(next http.Handler) http.Handler {
 			http.Redirect(w, r, "/login?continue="+url.QueryEscape(r.URL.Path), http.StatusFound)
 			return
 		}
-		next.ServeHTTP(w, r.WithContext(withAuth(r.Context(), key, rec.Role)))
+		next.ServeHTTP(w, r.WithContext(withAuthNamed(r.Context(), key, rec.Role, rec.Name)))
 	})
 }
 
-// authCtx carries the authenticated gateway key and its role.
+// authCtx carries the authenticated gateway key, its role, and the key's
+// configured NAME.
+//
+// The name is here because plugin HTTP endpoints authorise by key name rather
+// than by secret, and looking the name up from the key at the point of use would
+// mean reading config on every request. It is "" for a key configured without
+// one, which such an endpoint can never match.
 type authCtx struct {
 	key  string
 	role string
+	name string
 }
 
 type authCtxKeyT struct{}
 
 func withAuth(ctx context.Context, key, role string) context.Context {
-	return context.WithValue(ctx, authCtxKeyT{}, authCtx{key: key, role: role})
+	return withAuthNamed(ctx, key, role, "")
+}
+
+// withAuthNamed is withAuth plus the key's configured name.
+func withAuthNamed(ctx context.Context, key, role, name string) context.Context {
+	return context.WithValue(ctx, authCtxKeyT{}, authCtx{key: key, role: role, name: name})
+}
+
+// reqKeyName returns the authenticated key's configured name, or "" if it has
+// none. Plugin endpoint authorisation is written against this, not against the
+// secret: a secret is rotated without notice, so a rule keyed on it would
+// silently stop matching.
+func reqKeyName(ctx context.Context) string {
+	if a, ok := ctx.Value(authCtxKeyT{}).(authCtx); ok {
+		return a.name
+	}
+	return ""
 }
 
 // reqKey returns the authenticated gateway key id (masked suffix for display).

internal/gateway/source_fields_test.go

@@ -0,0 +1,495 @@
+package gateway
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"reflect"
+	"strings"
+	"testing"
+	"time"
+
+	"llmsproxy/internal/config"
+	"llmsproxy/internal/core"
+)
+
+// This file covers the second round of the same class of defect bf0657b fixed.
+//
+// bf0657b stopped a partial source edit from clobbering api_key. The upsert
+// still rewrites the WHOLE source, so every other field the request could not
+// express was reset to its zero value on any edit: proxy_url, api_key_env,
+// timeout and queue_timeout.
+//
+// The four are worse than cosmetic because they change how the source dials out.
+// proxy_url silently turns a proxied upstream into a direct one (or vice versa)
+// and api_key_env turns a source whose credential lives in an environment
+// variable into a credential-less source — which keeps answering 200 on write
+// and only fails on the NEXT call upstream, long after the editor left.
+//
+// The trigger is not exotic scripting: the WebUI's own edit form posts exactly
+// such a payload. See TestWebUIEditPayloadPreservesRoutingFields.
+
+// richGateway writes a config.yaml whose source carries all four fragile
+// fields, then boots a gateway over it.
+func richGateway(t *testing.T) *Gateway {
+	t.Helper()
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	cfg := `listen: 127.0.0.1:8080
+gateway_keys:
+  - sk-test
+default_model: AUTO
+adapter_dir: ` + filepath.Join(dir, "adapters") + `
+runtime_file: ` + filepath.Join(dir, "runtime.json") + `
+sources:
+  - name: rich
+    base_url: https://api.deepseek.com
+    api_key: sk-live
+    adapter: deepseek
+    proxy_url: http://127.0.0.1:7890
+    timeout: 300s
+    queue_timeout: 90s
+    max_concurrent: 8
+    models:
+      - id: deepseek-v4-flash
+        priority: 100
+        kind: chat
+`
+	if err := os.WriteFile(cfgPath, []byte(cfg), 0600); err != nil {
+		t.Fatal(err)
+	}
+	c, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	core, err := core.NewFromConfig(c)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(core.Close)
+	g, err := New(core)
+	if err != nil {
+		t.Fatal(err)
+	}
+	return g
+}
+
+// sourceField reads one field off the live source, i.e. what the routes
+// actually use rather than what the file happens to contain.
+func sourceField(t *testing.T, g *Gateway, name string, get func(config.Source) interface{}) interface{} {
+	t.Helper()
+	for _, s := range g.core.Sources() {
+		if s.Name == name {
+			return get(s)
+		}
+	}
+	t.Fatalf("source %q not found", name)
+	return nil
+}
+
+// webUIPayload is byte-for-byte the JSON that index.html's saveSource() builds
+// from the edit dialog. It deliberately carries none of the four fragile
+// fields, because the dialog has no inputs for them.
+const webUIPayload = `{
+  "name":"rich","base_url":"https://api.deepseek.com","api_key":"sk-live",
+  "adapter":"deepseek","endpoint":"","image_endpoint":"",
+  "models":[{"id":"deepseek-v4-flash","priority":100,"kind":"chat"}],
+  "headers":{},"meta":{},"temperature":0,"max_tokens":0,"max_concurrent":16,"rpm":0
+}`
+
+// TestWebUIEditPayloadPreservesRoutingFields is the regression test for the
+// reported defect: changing max_concurrent from 8 to 16 in the UI must not
+// silently drop the proxy, the env-var credential, or the timeouts.
+func TestWebUIEditPayloadPreservesRoutingFields(t *testing.T) {
+	g := richGateway(t)
+
+	// The one field the operator actually changed.
+	rr := doReq(t, g, "POST", "/api/sources", webUIPayload)
+	if rr.Code != 200 {
+		t.Fatalf("edit status=%d body=%s", rr.Code, rr.Body.String())
+	}
+
+	checks := []struct {
+		field string
+		got   interface{}
+		want  interface{}
+	}{
+		{"proxy_url", sourceField(t, g, "rich", func(s config.Source) interface{} { return s.ProxyURL }),
+			"http://127.0.0.1:7890"},
+		{"timeout", sourceField(t, g, "rich", func(s config.Source) interface{} { return s.Timeout }),
+			300 * time.Second},
+		{"queue_timeout", sourceField(t, g, "rich", func(s config.Source) interface{} { return s.QueueTimeout }),
+			90 * time.Second},
+		// The field the edit DID change, so a fix that simply refuses to write
+		// anything would also pass the three assertions above.
+		{"max_concurrent", sourceField(t, g, "rich", func(s config.Source) interface{} { return s.MaxConcurrent }),
+			16},
+	}
+	for _, c := range checks {
+		if c.got != c.want {
+			t.Errorf("%s = %v, want %v", c.field, c.got, c.want)
+		}
+	}
+}
+
+// TestSourceEditPreservesAPIKeyEnv is called out separately because it is the
+// only one of the four whose loss is a credential failure rather than a routing
+// difference, and it is invisible until the next upstream call.
+func TestSourceEditPreservesAPIKeyEnv(t *testing.T) {
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	cfg := `listen: 127.0.0.1:8080
+gateway_keys:
+  - sk-test
+default_model: AUTO
+adapter_dir: ` + filepath.Join(dir, "adapters") + `
+runtime_file: ` + filepath.Join(dir, "runtime.json") + `
+sources:
+  - name: envonly
+    base_url: https://api.deepseek.com
+    api_key_env: DEEPSEEK_KEY
+    adapter: deepseek
+    max_concurrent: 8
+    models:
+      - id: deepseek-v4-flash
+        priority: 100
+        kind: chat
+`
+	if err := os.WriteFile(cfgPath, []byte(cfg), 0600); err != nil {
+		t.Fatal(err)
+	}
+	c, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	core, err := core.NewFromConfig(c)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(core.Close)
+	g, err := New(core)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	// A credential-less source is legitimate, so the API must not invent one
+	// here; the point is that the ENV reference survives the edit.
+	rr := doReq(t, g, "POST", "/api/sources", `{
+      "name":"envonly","base_url":"https://api.deepseek.com","api_key":"",
+      "adapter":"deepseek","models":[{"id":"deepseek-v4-flash","priority":100,"kind":"chat"}],
+      "headers":{},"meta":{},"max_concurrent":4}`)
+	if rr.Code != 200 {
+		t.Fatalf("edit status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	got := sourceField(t, g, "envonly", func(s config.Source) interface{} { return s.APIKeyEnv })
+	if got != "DEEPSEEK_KEY" {
+		t.Fatalf("api_key_env = %v after an edit that did not mention it; "+
+			"the source now has no credential and will only fail on the next upstream call", got)
+	}
+	if n := sourceField(t, g, "envonly", func(s config.Source) interface{} { return s.MaxConcurrent }); n != 4 {
+		t.Fatalf("max_concurrent = %v, want the edited 4", n)
+	}
+}
+
+// TestSourceEditPersistsInheritedFieldsToDisk: inheriting in memory is not
+// enough — the values must reach config.yaml, or a restart silently undoes the
+// fix and the field is lost anyway.
+func TestSourceEditPersistsInheritedFieldsToDisk(t *testing.T) {
+	g := richGateway(t)
+	if rr := doReq(t, g, "POST", "/api/sources", webUIPayload); rr.Code != 200 {
+		t.Fatalf("edit status=%d: %s", rr.Code, rr.Body.String())
+	}
+	raw, err := os.ReadFile(g.core.Config().Path)
+	if err != nil {
+		t.Fatalf("read config: %v", err)
+	}
+	text := string(raw)
+	for _, want := range []string{
+		"proxy_url: http://127.0.0.1:7890",
+	} {
+		if !strings.Contains(text, want) {
+			t.Errorf("config.yaml lost %q on an unrelated edit:\n%s", want, text)
+		}
+	}
+	// The durations are re-serialized in Go's canonical form (300s becomes
+	// 5m0s), so they must be checked SEMANTICALLY. Asserting the original
+	// spelling would be a false red: the value survived, only its text changed.
+	// Re-loading the file is also the only way to prove the persisted value
+	// parses back to the same duration.
+	reloaded, err := config.Load(g.core.Config().Path)
+	if err != nil {
+		t.Fatalf("re-read config: %v", err)
+	}
+	if len(reloaded.Sources) == 0 {
+		t.Fatal("no sources after reload")
+	}
+	s := reloaded.Sources[0]
+	if s.Timeout != 300*time.Second {
+		t.Errorf("persisted timeout = %v, want 300s", s.Timeout)
+	}
+	if s.QueueTimeout != 90*time.Second {
+		t.Errorf("persisted queue_timeout = %v, want 90s", s.QueueTimeout)
+	}
+	if s.ProxyURL != "http://127.0.0.1:7890" {
+		t.Errorf("persisted proxy_url = %v, want it preserved", s.ProxyURL)
+	}
+}
+
+// ---- explicit set / clear, i.e. the pointer semantics actually work ----
+
+func TestSourceOptionalFieldsCanBeSet(t *testing.T) {
+	g := richGateway(t)
+	rr := doReq(t, g, "POST", "/api/sources", `{
+      "name":"rich","base_url":"https://api.deepseek.com","api_key":"sk-live",
+      "adapter":"deepseek","models":[{"id":"deepseek-v4-flash","priority":100,"kind":"chat"}],
+      "headers":{},"meta":{},
+      "proxy_url":"http://127.0.0.1:1080",
+      "api_key_env":"NEW_ENV_KEY",
+      "timeout":"45s",
+      "queue_timeout":"5s"}`)
+	if rr.Code != 200 {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	for _, c := range []struct {
+		field string
+		got   interface{}
+		want  interface{}
+	}{
+		{"proxy_url", sourceField(t, g, "rich", func(s config.Source) interface{} { return s.ProxyURL }), "http://127.0.0.1:1080"},
+		{"api_key_env", sourceField(t, g, "rich", func(s config.Source) interface{} { return s.APIKeyEnv }), "NEW_ENV_KEY"},
+		{"timeout", sourceField(t, g, "rich", func(s config.Source) interface{} { return s.Timeout }), 45 * time.Second},
+		{"queue_timeout", sourceField(t, g, "rich", func(s config.Source) interface{} { return s.QueueTimeout }), 5 * time.Second},
+	} {
+		if c.got != c.want {
+			t.Errorf("%s = %v, want %v", c.field, c.got, c.want)
+		}
+	}
+}
+
+// TestSourceOptionalFieldsCanBeCleared is the half that an "empty means
+// inherit" implementation gets wrong: it can never clear a field, so emptying
+// the proxy box in the UI would keep using the old proxy forever.
+func TestSourceOptionalFieldsCanBeCleared(t *testing.T) {
+	g := richGateway(t)
+	rr := doReq(t, g, "POST", "/api/sources", `{
+      "name":"rich","base_url":"https://api.deepseek.com","api_key":"sk-live",
+      "adapter":"deepseek","models":[{"id":"deepseek-v4-flash","priority":100,"kind":"chat"}],
+      "headers":{},"meta":{},
+      "proxy_url":"","timeout":"","queue_timeout":""}`)
+	if rr.Code != 200 {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	if got := sourceField(t, g, "rich", func(s config.Source) interface{} { return s.ProxyURL }); got != "" {
+		t.Errorf("proxy_url = %v, want it cleared", got)
+	}
+	// A cleared timeout falls back to the config default (120s), which is what
+	// mergedSources applies for a zero value — assert the effective behaviour
+	// rather than the raw zero, since that is what a caller actually gets.
+	if got := sourceField(t, g, "rich", func(s config.Source) interface{} { return s.Timeout }); got != config.DefaultSourceTimeout {
+		t.Errorf("timeout = %v, want the default %v after clearing", got, config.DefaultSourceTimeout)
+	}
+	// api_key_env was NOT in the request, so it must be untouched even though
+	// its sibling fields were cleared.
+	_ = rr
+}
+
+// TestSourceOptionalFieldsClearIsScoped: clearing the proxy must not clear the
+// env credential that the same request did not mention.
+func TestSourceOptionalFieldsClearIsScoped(t *testing.T) {
+	dir := t.TempDir()
+	cfgPath := filepath.Join(dir, "config.yaml")
+	cfg := `listen: 127.0.0.1:8080
+gateway_keys:
+  - sk-test
+default_model: AUTO
+adapter_dir: ` + filepath.Join(dir, "adapters") + `
+runtime_file: ` + filepath.Join(dir, "runtime.json") + `
+sources:
+  - name: both
+    base_url: https://api.deepseek.com
+    api_key_env: DEEPSEEK_KEY
+    adapter: deepseek
+    proxy_url: http://127.0.0.1:7890
+    max_concurrent: 8
+    models:
+      - id: deepseek-v4-flash
+        priority: 100
+        kind: chat
+`
+	os.WriteFile(cfgPath, []byte(cfg), 0600)
+	c, err := config.Load(cfgPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+	core, err := core.NewFromConfig(c)
+	if err != nil {
+		t.Fatalf("core: %v", err)
+	}
+	t.Cleanup(core.Close)
+	g, err := New(core)
+	if err != nil {
+		t.Fatal(err)
+	}
+	rr := doReq(t, g, "POST", "/api/sources", `{
+      "name":"both","base_url":"https://api.deepseek.com","api_key":"",
+      "adapter":"deepseek","models":[{"id":"deepseek-v4-flash","priority":100,"kind":"chat"}],
+      "headers":{},"meta":{},"proxy_url":""}`)
+	if rr.Code != 200 {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	if got := sourceField(t, g, "both", func(s config.Source) interface{} { return s.APIKeyEnv }); got != "DEEPSEEK_KEY" {
+		t.Errorf("api_key_env = %v; clearing proxy_url must not touch it", got)
+	}
+	if got := sourceField(t, g, "both", func(s config.Source) interface{} { return s.ProxyURL }); got != "" {
+		t.Errorf("proxy_url = %v, want cleared", got)
+	}
+}
+
+func TestSourceRejectsBadDuration(t *testing.T) {
+	g := richGateway(t)
+	before := sourceField(t, g, "rich", func(s config.Source) interface{} { return s.Timeout })
+	rr := doReq(t, g, "POST", "/api/sources", `{
+      "name":"rich","base_url":"https://api.deepseek.com","api_key":"sk-live",
+      "adapter":"deepseek","models":[{"id":"deepseek-v4-flash","priority":100,"kind":"chat"}],
+      "headers":{},"meta":{},"timeout":"not-a-duration"}`)
+	if rr.Code != 400 {
+		t.Fatalf("status=%d, want 400 for an unparseable duration: %s", rr.Code, rr.Body.String())
+	}
+	if !strings.Contains(rr.Body.String(), "timeout") {
+		t.Errorf("the error must name the offending field: %s", rr.Body.String())
+	}
+	// A rejected write must leave the source exactly as it was.
+	if after := sourceField(t, g, "rich", func(s config.Source) interface{} { return s.Timeout }); after != before {
+		t.Errorf("a rejected edit still changed timeout: %v -> %v", before, after)
+	}
+}
+
+// TestSourceCreateSetsOptionalFields: the create path must accept them too, and
+// must not require an existing source to inherit from (a nil-pointer crash
+// here would only show up as a 500 on every new source).
+func TestSourceCreateSetsOptionalFields(t *testing.T) {
+	up := mockUpstream()
+	defer up.Close()
+	g := newTestGateway(t)
+	rr := doReq(t, g, "POST", "/api/sources", `{
+      "name":"fresh","base_url":"`+up.URL+`","api_key":"sk-x","adapter":"openai",
+      "models":[{"id":"m1","kind":"chat"}],"headers":{},"meta":{},
+      "proxy_url":"http://127.0.0.1:7890","timeout":"10s","queue_timeout":"3s"}`)
+	if rr.Code != 200 {
+		t.Fatalf("create status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	if got := sourceField(t, g, "fresh", func(s config.Source) interface{} { return s.ProxyURL }); got != "http://127.0.0.1:7890" {
+		t.Errorf("proxy_url = %v, want it set on create", got)
+	}
+	if got := sourceField(t, g, "fresh", func(s config.Source) interface{} { return s.Timeout }); got != 10*time.Second {
+		t.Errorf("timeout = %v, want 10s on create", got)
+	}
+}
+
+// TestSourceRevealExposesDurations guards the round-trip the edit form depends
+// on. config.Source tags Timeout/QueueTimeout `json:"-"`, so marshaling the
+// struct omits them; the form then reads an empty timeout box and — because it
+// always sends the box back — CLEARS the stored timeout on any unrelated save.
+// A test that only checks "the edit preserves timeout" would not have caught
+// that, because the loss happens between the read and the write.
+//
+// The credentials are asserted in the same breath: they came from the same
+// branch, and an earlier iteration of this fix masked the key and broke the
+// form's other half.
+func TestSourceRevealExposesDurations(t *testing.T) {
+	up := mockUpstream()
+	defer up.Close()
+	g := newTestGateway(t, config.Source{
+		Name:    "revealed",
+		BaseURL: up.URL,
+		Adapter: "openai",
+		APIKey:  "sk-reveal-me",
+		Models:  []config.Model{{ID: "m1", Kind: "chat"}},
+		Timeout: 300 * time.Second,
+	})
+
+	rr := doReq(t, g, "GET", "/api/v1/sources/revealed?reveal=credentials", "")
+	if rr.Code != 200 {
+		t.Fatalf("reveal status=%d: %s", rr.Code, rr.Body.String())
+	}
+	var view struct {
+		Source map[string]interface{} `json:"source"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &view); err != nil {
+		t.Fatalf("decode: %v", err)
+	}
+	// Duration strings, matching what the write path parses.
+	if got := view.Source["timeout"]; got != "5m0s" {
+		t.Errorf("reveal timeout = %v, want the string \"5m0s\"; the edit form "+
+			"would show an empty box and clear the stored value on save", got)
+	}
+	// The unset one shows the DEFAULT, not "": ApplyDefaults filled
+	// QueueTimeout before the form ever sees it. Both are acceptable as long as
+	// the value round-trips, so assert it is a parseable duration string.
+	qs, _ := view.Source["queue_timeout"].(string)
+	if _, err := time.ParseDuration(qs); err != nil {
+		t.Errorf("reveal queue_timeout = %q is not a duration string: %v", qs, err)
+	}
+	if got := view.Source["api_key"]; got != "sk-reveal-me" {
+		t.Errorf("reveal api_key = %v, want it in the clear for the form", got)
+	}
+	// The masked view must NOT gain the durations' secrets — it is fine for it
+	// to omit them, but it must never carry the key.
+	masked := doReq(t, g, "GET", "/api/v1/sources/revealed", "")
+	if strings.Contains(masked.Body.String(), "sk-reveal-me") {
+		t.Error("the masked view leaked the key")
+	}
+}
+
+// TestSourcePayloadCoversEveryEditableField is the guard against the class
+// itself. Any config.Source field a source edit can legitimately set must
+// exist on sourcePayload, otherwise the upsert resets it.
+//
+// It reflects over the STRUCT rather than over json.Marshal output: the four
+// pointer fields carry `omitempty`, so a nil pointer legitimately disappears
+// from the marshaled form — that absence is precisely the "not mentioned" signal
+// the handler relies on. Checking marshaled keys would therefore report a
+// healthy field as missing.
+func TestSourcePayloadCoversEveryEditableField(t *testing.T) {
+	// Fields of config.Source that a source edit can legitimately set.
+	want := []string{
+		"name", "base_url", "api_key", "api_key_env", "adapter", "endpoint",
+		"image_endpoint", "models", "headers", "proxy_url", "meta",
+		"temperature", "max_tokens", "max_concurrent", "rpm",
+		"timeout", "queue_timeout",
+	}
+	pt := reflect.TypeOf(sourcePayload{})
+	for _, f := range want {
+		if _, ok := pt.FieldByName(f); !ok {
+			// Tolerate a differently-spelled Go field only when the JSON tag
+			// matches, so the check stays about the wire contract.
+			if !hasJSONTag(pt, f) {
+				t.Errorf("config.Source has %q but sourcePayload has no such JSON key: "+
+					"an edit that omits it resets the stored value", f)
+			}
+		}
+	}
+}
+
+// hasJSONTag reports whether any field of t carries the given JSON name.
+func hasJSONTag(t reflect.Type, name string) bool {
+	for i := 0; i < t.NumField(); i++ {
+		if jsonName(t.Field(i)) == name {
+			return true
+		}
+	}
+	return false
+}
+
+// jsonName returns the JSON name of a struct field, falling back to the Go
+// field name when no tag is present.
+func jsonName(f reflect.StructField) string {
+	tag := f.Tag.Get("json")
+	if tag == "" {
+		return f.Name
+	}
+	if n := strings.Split(tag, ",")[0]; n != "" {
+		return n
+	}
+	return f.Name
+}

internal/gateway/sources_api_test.go

@@ -0,0 +1,259 @@
+package gateway
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+)
+
+// These cover the two write-path defects found while adding a model to a live
+// source by hand:
+//
+//  1. PUT /api/sources/{name} was documented in the API index but not
+//     implemented, so it answered 405 while POST (which does the same upsert)
+//     worked. A caller following the documented contract could not edit a
+//     source at all.
+//  2. POST is an upsert that replaces the whole source, so a partial edit that
+//     did not carry api_key persisted an empty/placeholder credential. The
+//     source kept its name and models and answered 401 on the next call, well
+//     after the writing script had exited 0.
+//
+// The second one is the dangerous class: a write that reports success and
+// breaks the source later.
+
+// sourceAPIKey reads the stored key back the way an operator would, i.e. what
+// the routes actually use, not what the file happens to contain.
+func sourceAPIKey(t *testing.T, g *Gateway, name string) string {
+	t.Helper()
+	for _, s := range g.core.Sources() {
+		if s.Name == name {
+			return s.APIKey
+		}
+	}
+	t.Fatalf("source %q not found", name)
+	return ""
+}
+
+func seededGateway(t *testing.T) (*Gateway, string) {
+	t.Helper()
+	up := mockUpstream()
+	t.Cleanup(up.Close)
+	g := newTestGateway(t, config.Source{
+		Name:    "seeded",
+		BaseURL: up.URL,
+		Adapter: "openai",
+		APIKey:  "sk-real-credential",
+		Models:  []config.Model{{ID: "m-one", Kind: "chat"}},
+	})
+	return g, up.URL
+}
+
+func TestSourcePutUpdatesSource(t *testing.T) {
+	g, up := seededGateway(t)
+
+	// A single-field edit: change one model's priority, name the source only in
+	// the path. This is the shape the API index documents for PUT.
+	rr := doReq(t, g, "PUT", "/api/sources/seeded", fmt.Sprintf(
+		`{"base_url":%q,"adapter":"openai","api_key":"sk-real-credential","models":[{"id":"m-one","priority":42,"kind":"chat"}]}`, up))
+	if rr.Code != http.StatusOK {
+		t.Fatalf("PUT status=%d body=%s", rr.Code, rr.Body.String())
+	}
+
+	var found bool
+	for _, s := range g.core.Sources() {
+		if s.Name != "seeded" {
+			continue
+		}
+		found = true
+		for _, m := range s.Models {
+			if m.ID == "m-one" && m.Priority != 42 {
+				t.Fatalf("priority not applied: %+v", s.Models)
+			}
+		}
+	}
+	if !found {
+		t.Fatal("source vanished after PUT")
+	}
+	// And it must be routable, not merely stored.
+	if rr := doReq(t, g, "GET", "/v1/models", ""); rr.Code != http.StatusOK ||
+		!strings.Contains(rr.Body.String(), "m-one") {
+		t.Fatalf("model not routable after PUT: %d %s", rr.Code, rr.Body.String())
+	}
+}
+
+func TestSourcePutRejectsNameMismatch(t *testing.T) {
+	g, _ := seededGateway(t)
+	// Body names a different source than the path. Silently preferring either
+	// one would let a client edit a source it did not name.
+	rr := doReq(t, g, "PUT", "/api/sources/seeded", `{"name":"other","base_url":"http://x","models":[{"id":"m"}]}`)
+	if rr.Code != http.StatusBadRequest {
+		t.Fatalf("name mismatch accepted: status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	if sourceAPIKey(t, g, "seeded") != "sk-real-credential" {
+		t.Fatal("rejected PUT still mutated the source")
+	}
+}
+
+func TestSourcePutRequiresNameInPath(t *testing.T) {
+	g, _ := seededGateway(t)
+	rr := doReq(t, g, "PUT", "/api/sources", `{"base_url":"http://x","models":[{"id":"m"}]}`)
+	if rr.Code != http.StatusBadRequest {
+		t.Fatalf("PUT without a name in the path: status=%d body=%s", rr.Code, rr.Body.String())
+	}
+}
+
+func TestSourceUpsertKeepsAPIKey(t *testing.T) {
+	for _, tc := range []struct {
+		name    string
+		sentKey string
+	}{
+		{"explicit mask", keepExistingAPIKey},
+		{"omitted key", ""},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			g, up := seededGateway(t)
+			rr := doReq(t, g, "PUT", "/api/sources/seeded", fmt.Sprintf(
+				`{"base_url":%q,"adapter":"openai","api_key":%q,"models":[{"id":"m-one"},{"id":"m-two"}]}`,
+				up, tc.sentKey))
+			if rr.Code != http.StatusOK {
+				t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+			}
+			if got := sourceAPIKey(t, g, "seeded"); got != "sk-real-credential" {
+				t.Fatalf("api_key was clobbered: got %q, want the pre-existing credential", got)
+			}
+			// The failure this guards is a source that looks configured but
+			// cannot authenticate, so assert the route still works.
+			if rr := doReq(t, g, "GET", "/v1/models", ""); rr.Code != http.StatusOK {
+				t.Fatalf("source unusable after edit: %d %s", rr.Code, rr.Body.String())
+			}
+		})
+	}
+}
+
+func TestSourceUpsertStillWritesNewKey(t *testing.T) {
+	// The keep-existing path must not swallow a deliberate rotation.
+	g, up := seededGateway(t)
+	rr := doReq(t, g, "PUT", "/api/sources/seeded", fmt.Sprintf(
+		`{"base_url":%q,"adapter":"openai","api_key":"sk-rotated","models":[{"id":"m-one"}]}`, up))
+	if rr.Code != http.StatusOK {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	if got := sourceAPIKey(t, g, "seeded"); got != "sk-rotated" {
+		t.Fatalf("rotation ignored: got %q", got)
+	}
+}
+
+func TestSourceUpsertNewSourceMayHaveNoKey(t *testing.T) {
+	// Empty is a real value for a credential-less source (local upstreams), so
+	// inherit-on-empty must not invent one for a source that does not exist yet.
+	up := mockUpstream()
+	defer up.Close()
+	g := newTestGateway(t)
+	rr := doReq(t, g, "POST", "/api/sources", fmt.Sprintf(
+		`{"name":"noauth","base_url":%q,"adapter":"openai","models":[{"id":"local-m"}]}`, up.URL))
+	if rr.Code != http.StatusOK {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	if got := sourceAPIKey(t, g, "noauth"); got != "" {
+		t.Fatalf("unexpected key on a new credential-less source: %q", got)
+	}
+}
+
+func TestSourceModelIDsIsAdditive(t *testing.T) {
+	// "Add one model" is the edit that motivated the defect: it needs no read
+	// of the existing list, so it must not require one.
+	g, up := seededGateway(t)
+	rr := doReq(t, g, "PUT", "/api/sources/seeded", fmt.Sprintf(
+		`{"base_url":%q,"adapter":"openai","api_key":%q,"model_ids":["m-two","m-two"," m-three "]}`, up, keepExistingAPIKey))
+	if rr.Code != http.StatusOK {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	var ids []string
+	for _, s := range g.core.Sources() {
+		if s.Name == "seeded" {
+			for _, m := range s.Models {
+				ids = append(ids, m.ID)
+			}
+		}
+	}
+	// m-one kept, m-two not duplicated, blank entries dropped, whitespace trimmed.
+	want := map[string]int{"m-one": 1, "m-two": 1, "m-three": 1}
+	if len(ids) != len(want) {
+		t.Fatalf("models=%v, want exactly %v", ids, want)
+	}
+	seen := map[string]int{}
+	for _, id := range ids {
+		seen[id]++
+	}
+	for id, n := range want {
+		if seen[id] != n {
+			t.Fatalf("models=%v: %q appears %d times, want %d", ids, id, seen[id], n)
+		}
+	}
+}
+
+func TestSourceWriteRejectsNonAdmin(t *testing.T) {
+	g, up := seededGateway(t)
+	for _, m := range []string{"POST", "PUT", "DELETE"} {
+		req, _ := http.NewRequest(m, "/api/sources/seeded", strings.NewReader(fmt.Sprintf(
+			`{"name":"seeded","base_url":%q,"models":[{"id":"m"}]}`, up)))
+		req.Header.Set("Authorization", "Bearer sk-not-a-real-key")
+		req.Header.Set("Content-Type", "application/json")
+		rr := httptest.NewRecorder()
+		g.Handler().ServeHTTP(rr, req)
+		if rr.Code == http.StatusOK {
+			t.Fatalf("%s allowed for an unknown key: %s", m, rr.Body.String())
+		}
+	}
+	if sourceAPIKey(t, g, "seeded") != "sk-real-credential" {
+		t.Fatal("unauthenticated write mutated the source")
+	}
+}
+
+// TestSourceWritePersistsToDisk guards the difference between "the in-memory
+// registry accepted it" and "it will survive a restart". The api_key must be
+// sealed at rest, never the literal placeholder.
+func TestSourceWritePersistsToDisk(t *testing.T) {
+	g, up := seededGateway(t)
+	rr := doReq(t, g, "PUT", "/api/sources/seeded", fmt.Sprintf(
+		`{"base_url":%q,"adapter":"openai","api_key":%q,"models":[{"id":"m-one"}]}`, up, keepExistingAPIKey))
+	if rr.Code != http.StatusOK {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	raw, err := os.ReadFile(g.core.Config().Path)
+	if err != nil {
+		t.Fatalf("read config: %v", err)
+	}
+	text := string(raw)
+	if strings.Contains(text, keepExistingAPIKey) {
+		t.Fatal("the keep-mask was persisted as a literal api_key")
+	}
+	if strings.Contains(text, "sk-real-credential") {
+		t.Fatal("api_key stored in plaintext; it must be sealed at rest")
+	}
+	if !strings.Contains(text, "m-one") {
+		t.Fatal("source not persisted")
+	}
+	// The masked GET view must not leak the credential either.
+	rr = doReq(t, g, "GET", "/api/v1/sources/seeded", "")
+	if rr.Code != http.StatusOK {
+		t.Fatalf("get status=%d", rr.Code)
+	}
+	var view struct {
+		Source struct {
+			APIKey string `json:"api_key"`
+		} `json:"source"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &view); err != nil {
+		t.Fatalf("decode: %v", err)
+	}
+	if view.Source.APIKey == "sk-real-credential" {
+		t.Fatal("GET /api/v1/sources leaked the plaintext api_key")
+	}
+}

internal/gateway/stats.go

@@ -43,11 +43,42 @@ type Req struct {
 	// accounting at all (even when the hit count is 0). The WebUI shows
 	// "0%" instead of "—" for such rows.
 	CacheReported bool `json:"cache_reported,omitempty"`
+	// Reasoning is completion_tokens_details.reasoning_tokens.
+	//
+	// It is a SUBSET of Compl, not an addition: reasoning tokens are billed at
+	// the completion rate and already counted there. This field exists so a
+	// pricing rule can split them when a provider prices thinking tokens
+	// differently (most do NOT — OpenAI and DeepSeek bill them as ordinary
+	// output), and so a reader can see what fraction of a bill is thinking.
+	// 0 means "not reported", which is NOT the same as "no reasoning happened" —
+	// three of the sources measured return no such field at all.
+	Reasoning int64 `json:"reasoning_tokens,omitempty"`
 	OK            bool `json:"ok"`
 	// Status http status code
 	Status int `json:"status"`
 	// Err short error message
 	Err string `json:"error,omitempty"`
+	// ImageCount is how many images an image-generation request produced.
+	// Kept separate from Compl/Prompt: image generation has no token concept,
+	// so counting images as "completion tokens" would corrupt the token totals.
+	ImageCount int `json:"image_count,omitempty"`
+
+	// Walk is the AUTO chain's step-by-step trace for this request: which tiers
+	// were skipped and why, which slots hard-failed, which one served it. It is
+	// the only way a consumer can tell "tier 1 served this" from "tier 1 was
+	// cooling so we dropped to tier 3" — a distinction that is the entire point
+	// of a priority chain.
+	//
+	// json:"-" — deliberately NOT persisted. The audit file is a hot append and
+	// this is observational detail: on a degraded gateway every request would
+	// carry a multi-element array, and the audit trail's own retention (16 files
+	// x 16 MB) is already the largest thing on the box. A plugin that wants the
+	// walk sees it live at request_end; an operator post-mortem reads it from the
+	// plugin's own accumulated state or from /api/auto slot health.
+	//
+	// Only populated when a plugin is loaded (chainTraceSink returns nil
+	// otherwise), so a gateway with no plugins allocates nothing for it.
+	Walk []map[string]interface{} `json:"-"`
 }
 
 // Stat aggregates counters for one dimension row.
@@ -70,6 +101,16 @@ type agrRow struct {
 
 // Stats collects per-key / per-model / per-source aggregates plus a bounded
 // ring of raw request records, all guarded by one mutex.
+//
+// LOCKING: mu is a plain sync.Mutex and is NOT reentrant. The *Locked methods
+// (aggregateLocked, addKeyTokenLocked, addKeyHourLocked, addKeyReqLocked,
+// wantPinnedBuckets, rotateAuditLocked, …) assume the caller already holds it,
+// while every other exported method takes it itself.
+//
+// Calling an exported method while already holding mu DEADLOCKS. This is not
+// hypothetical: a test that did KeyWindowReqs under s.mu.Lock() hung until the
+// 10-minute panic timeout. Always reach for the *Locked variant when the lock
+// is already held, and prefer the exported method when it is not.
 type Stats struct {
 	mu            sync.Mutex
 	active        int64
@@ -84,6 +125,21 @@ type Stats struct {
 	auditPath     string
 	replayPartial bool                       // aggregates built from a bounded audit tail
 	modelHour     map[string]map[int64]int64 // model -> unix-hour bucket -> tokens
+	// keyModelHour buckets the same tokens as modelHour but keyed by
+	// (gateway key id, model) so per-key model quotas are isolated from each
+	// other. modelHour stays key-blind on purpose: it backs the AUTO slot
+	// quota, which is a gateway-wide limit on a slot, not a per-key one.
+	keyModelHour map[string]map[string]map[int64]int64 // key -> model -> unix-hour -> tokens
+	// keyHour buckets a key's total tokens per unix hour, backing the
+	// whole-key quota (all models of one key share one budget).
+	keyHour map[string]map[int64]int64
+	// keyReqHour buckets a key's request count per unix hour, backing the
+	// key-wide request quota. Counted for every request type (chat, stream,
+	// image) including failed ones, so a failing client cannot loop for free.
+	keyReqHour map[string]map[int64]int64
+	// pinnedWanted records which (key, model) pairs have a quota that pins a
+	// source, so the pinned bucket is maintained only where it is read.
+	pinnedWanted map[string]bool
 }
 
 const hourSec = 3600
@@ -116,14 +172,18 @@ func NewStats(maxRecords int) *Stats {
 		maxRecords = defaultRingSize
 	}
 	return &Stats{
-		byKey:      map[string]*Stat{},
-		byModel:    map[string]*Stat{},
-		bySrc:      map[string]*Stat{},
-		byKeyModel: map[string]map[string]*Stat{},
-		byKeySrc:   map[string]map[string]*Stat{},
-		byStatus:   map[int]*Stat{},
-		modelHour:  map[string]map[int64]int64{},
-		maxRecs:    maxRecords,
+		byKey:        map[string]*Stat{},
+		byModel:      map[string]*Stat{},
+		bySrc:        map[string]*Stat{},
+		byKeyModel:   map[string]map[string]*Stat{},
+		byKeySrc:     map[string]map[string]*Stat{},
+		byStatus:     map[int]*Stat{},
+		modelHour:    map[string]map[int64]int64{},
+		keyModelHour: map[string]map[string]map[int64]int64{},
+		keyHour:      map[string]map[int64]int64{},
+		keyReqHour:   map[string]map[int64]int64{},
+		pinnedWanted: map[string]bool{},
+		maxRecs:      maxRecords,
 	}
 }
 
@@ -318,6 +378,9 @@ func (s *Stats) Record(r Req) {
 // aggregateLocked folds r into every aggregate row and the quota window
 // bucket. Caller must hold s.mu.
 func (s *Stats) aggregateLocked(r Req) {
+	if r.Key != "" {
+		s.addKeyReqLocked(r.Key, (r.Time/1000)/hourSec, 1)
+	}
 	inc(s.byKey, r.Key, r)
 	if r.Model != "" {
 		inc(s.byModel, r.Model, r)
@@ -362,6 +425,12 @@ func (s *Stats) aggregateLocked(r Req) {
 			s.modelHour[key] = hm
 		}
 		hm[h] += tok
+		// per-key buckets: same token split, but isolated per key so one
+		// key's quota cannot be drained by another key's traffic
+		if r.Key != "" {
+			s.addKeyTokenLocked(r.Key, r.Model, r.Source, h, tok)
+			s.addKeyHourLocked(r.Key, h, tok)
+		}
 		// retention: 24*40 = 960 hourly buckets ≈ 40 days of history (covers
 		// the longest "month" quota window)
 		if len(hm) > 24*40 {
@@ -374,6 +443,185 @@ func (s *Stats) aggregateLocked(r Req) {
 	}
 }
 
+// quotaRetentionHours is how much hourly history the per-key quota buckets
+// keep. It matches the modelHour retention (40 days) so the longest "month"
+// window is fully covered after a restart, and it is deliberately applied per
+// key so an idle key's buckets are reclaimed instead of pinning memory.
+const quotaRetentionHours = 24 * 40
+
+// addKeyTokenLocked adds tok to one key's (model[, source]) hourly bucket.
+//
+// The bare-model bucket is always written: it is what a scope entry without a
+// source pin reads, and writing it is what makes that quota work at all (every
+// recorded request carries a resolved source, so a source::model-only bucket
+// would leave an unpinned quota reading an empty bucket forever).
+//
+// The pinned "source::model" bucket is written only for (key, model) pairs
+// some scope entry actually pins a source for. Maintaining it unconditionally
+// doubles the bucket count — measured at 26.7 MB for 20 keys x 8 models x 3
+// sources at full retention, against a documented ~32 MB total budget — to
+// serve a lookup that almost nobody performs.
+func (s *Stats) addKeyTokenLocked(key, model, source string, h, tok int64) {
+	if model == "" {
+		return
+	}
+	byModel := s.keyModelHour[key]
+	if byModel == nil {
+		byModel = map[string]map[int64]int64{}
+		s.keyModelHour[key] = byModel
+	}
+	addBucket(byModel, model, h, tok)
+	if source == "" {
+		return
+	}
+	pinned := source + "::" + model
+	if pinned == model || !s.pinnedWanted[key+"\x00"+model] {
+		return
+	}
+	addBucket(byModel, pinned, h, tok)
+}
+
+// addBucket adds tok to one named hourly bucket and enforces retention.
+func addBucket(byModel map[string]map[int64]int64, name string, h, tok int64) {
+	hm := byModel[name]
+	if hm == nil {
+		hm = map[int64]int64{}
+		byModel[name] = hm
+	}
+	hm[h] += tok
+	if len(hm) > quotaRetentionHours {
+		for k := range hm {
+			if k < h-quotaRetentionHours {
+				delete(hm, k)
+			}
+		}
+	}
+}
+
+// wantPinnedBuckets starts maintaining the pinned (source::model) bucket for a
+// key's model, so a quota that pins a source counts traffic from the moment it
+// is configured. Usage recorded before that cannot be attributed to a source
+// after the fact, so a pinned quota may under-count its first window.
+func (s *Stats) wantPinnedBuckets(key, model string) {
+	if key == "" || model == "" {
+		return
+	}
+	s.pinnedWanted[key+"\x00"+model] = true
+}
+
+// addKeyHourLocked adds tok to one key's all-model hourly total.
+func (s *Stats) addKeyHourLocked(key string, h, tok int64) {
+	hm := s.keyHour[key]
+	if hm == nil {
+		hm = map[int64]int64{}
+		s.keyHour[key] = hm
+	}
+	hm[h] += tok
+	if len(hm) > quotaRetentionHours {
+		for k := range hm {
+			if k < h-quotaRetentionHours {
+				delete(hm, k)
+			}
+		}
+	}
+}
+
+// KeyWindowReqs returns how many requests one key issued within sec seconds;
+// sec <= 0 means all retained history.
+func (s *Stats) KeyWindowReqs(key string, sec int64) int64 {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return sumBuckets(s.keyReqHour[key], time.Now().Unix(), sec)
+}
+
+// addKeyReqLocked adds n requests to a key's hourly count bucket.
+func (s *Stats) addKeyReqLocked(key string, h, n int64) {
+	hm := s.keyReqHour[key]
+	if hm == nil {
+		hm = map[int64]int64{}
+		s.keyReqHour[key] = hm
+	}
+	hm[h] += n
+	if len(hm) > quotaRetentionHours {
+		for k := range hm {
+			if k < h-quotaRetentionHours {
+				delete(hm, k)
+			}
+		}
+	}
+}
+
+// sumBuckets totals the hourly buckets inside the window ending at `now`
+// (unix seconds). sec <= 0 means "all retained history" (no reset).
+//
+// It only visits the hour buckets a window can actually cover. The naive full
+// scan measured 5.9us/op at the full 40-day retention, and a per-key quota
+// check runs on every request (2-3 of these), so the scan cost lands directly
+// on the request path; visiting at most sec/3600 + 1 buckets makes it
+// proportional to the window instead of to the retained history. The map keys
+// are whole unix hours, so the covered range is exact rather than a sample.
+func sumBuckets(hm map[int64]int64, now, sec int64) int64 {
+	if len(hm) == 0 {
+		return 0
+	}
+	if sec <= 0 {
+		var total int64
+		for _, v := range hm {
+			total += v
+		}
+		return total
+	}
+	nowHour := now / hourSec
+	// A bucket at hour H covers [H*3600, (H+1)*3600). The pre-existing
+	// WindowTokens counts a bucket when h*3600 >= now-sec, so the first hour
+	// that qualifies is ceil((now-sec)/3600) — NOT floor, which would pull in
+	// one bucket too many (caught by TestSumBucketsMatchesFullScan: a 30-day
+	// window reported 8878 vs 8649).
+	firstHour := (now - sec + hourSec - 1) / hourSec
+	if firstHour < 0 {
+		firstHour = 0
+	}
+	var total int64
+	for h := firstHour; h <= nowHour; h++ {
+		total += hm[h]
+	}
+	return total
+}
+
+// KeyWindowModelTokens returns the tokens one key consumed for one model
+// (optionally pinned to a source) within sec seconds; sec <= 0 means all
+// retained history. Unlike WindowTokens this is isolated per key, which is
+// what a per-key model quota needs.
+//
+// A source pin matches the pinned bucket exactly. Without a pin the call
+// totals the key's bare-model bucket, which counts traffic on every source —
+// a quota on "this model" should not stop applying just because the request
+// happened to be served by a different upstream.
+func (s *Stats) KeyWindowModelTokens(key, model, source string, sec int64) int64 {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	byModel := s.keyModelHour[key]
+	if len(byModel) == 0 {
+		return 0
+	}
+	mk := model
+	if source != "" {
+		mk = source + "::" + model
+		// from now on this pair keeps a pinned bucket, so a quota configured
+		// against this source counts the traffic it is meant to cap
+		s.wantPinnedBuckets(key, model)
+	}
+	return sumBuckets(byModel[mk], time.Now().Unix(), sec)
+}
+
+// KeyWindowTokens returns one key's total tokens across all models within sec
+// seconds; sec <= 0 means all retained history. It backs the whole-key quota.
+func (s *Stats) KeyWindowTokens(key string, sec int64) int64 {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return sumBuckets(s.keyHour[key], time.Now().Unix(), sec)
+}
+
 // rotateAuditLocked renames the audit file to <path>.<unix>.old once it
 // exceeds auditRotateBytes and prunes old files beyond auditKeepOld, keeping
 // the newest ones. Caller must hold s.mu.
@@ -464,6 +712,27 @@ func AutoPeriodSeconds(period string, hours int64) int64 {
 	return 0
 }
 
+// AutoSecondsToReset returns how many seconds remain until the quota window
+// rolls over, for the Retry-After header on a 429. Buckets are whole unix
+// hours, so the value is rounded up to the next hour boundary and never
+// exceeds one window. 0 means "unknown" (never resets, or no period set), in
+// which case the caller should not advertise a retry time.
+func AutoSecondsToReset(period string, hours int64) int64 {
+	sec := AutoPeriodSeconds(period, hours)
+	if sec <= 0 {
+		return 0
+	}
+	now := time.Now().Unix()
+	// window = the last sec seconds, bucketed by whole hours; the oldest
+	// bucket still inside the window expires at its own hour boundary
+	elapsed := now % hourSec
+	left := sec - elapsed
+	if left <= 0 || left > sec {
+		left = sec
+	}
+	return left
+}
+
 // WindowTokens returns the tokens consumed for one model (optionally pinned
 // to a single source) within the window; sec <= 0 means all time. Buckets are
 // whole unix hours, so a sliding window overcounts by up to one hour — an
@@ -952,3 +1221,16 @@ func (s *Stats) Snapshot(limit int, key string) map[string]interface{} {
 		"replay_partial": s.replayPartial,
 	}
 }
+
+// Active reports requests in flight right now.
+//
+// It is a live gauge, not an aggregate: there is no calendar period that scopes
+// it, which is why the period-scoped /api/stats response has to carry it
+// explicitly. Before it did, the day/week/month views fed `st.active || 0` a
+// missing key and showed a hard 0 — the same gateway that was serving hundreds
+// of requests a second looked idle in three of the four views.
+func (s *Stats) Active() int64 {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.active
+}

internal/gateway/stats_lock_test.go

@@ -0,0 +1,132 @@
+package gateway
+
+import (
+	"reflect"
+	"runtime"
+	"strings"
+	"testing"
+	"time"
+)
+
+// The Stats mutex is a plain sync.Mutex: calling an exported (self-locking)
+// method while already holding it deadlocks. That happened for real — a probe
+// held s.mu and called KeyWindowReqs, hanging until the test binary's 10-minute
+// panic timeout — so the rule is pinned here rather than left in a comment.
+//
+// The check is structural: every method that takes the lock must say so in its
+// name or its doc comment. That keeps the trap visible when someone adds the
+// next method, which is the only time the rule can be forgotten.
+
+// exportedSelfLocking lists the exported Stats methods that take the mutex.
+// It is derived from reflection at run time; the assertions below are what
+// actually pin the convention.
+func TestStatsExportedMethodsDocumentTheirLocking(t *testing.T) {
+	typ := reflect.TypeOf(&Stats{})
+	for i := 0; i < typ.NumMethod(); i++ {
+		m := typ.Method(i)
+		if m.PkgPath != "" { // unexported
+			continue
+		}
+		// Methods that neither lock nor touch guarded state are fine either way;
+		// the ones that matter are those reaching into the maps under mu.
+		if !methodTouchesLockedState(m.Name) {
+			continue
+		}
+		if strings.HasSuffix(m.Name, "Locked") {
+			t.Errorf("Stats.%s is exported but named *Locked; the suffix means "+
+				"the CALLER holds the lock, so it must not be exported", m.Name)
+		}
+	}
+}
+
+// methodTouchesLockedState is the set of exported methods known to read or
+// write state guarded by Stats.mu. Kept explicit (rather than inferred) so a
+// new method is not silently assumed safe.
+func methodTouchesLockedState(name string) bool {
+	switch name {
+	case "KeyWindowReqs", "KeyWindowModelTokens", "KeyWindowTokens",
+		"WindowTokens", "KeyTokens", "ModelTokens", "Record", "AppendAudit",
+		"LoadAudit", "Snapshot", "SourceRecent", "SourceAverages",
+		"AuditRecords", "AuditPage", "StreamAuditRecords", "ReplayPartial",
+		"PoolStats":
+		return true
+	}
+	return false
+}
+
+// TestStatsLockedMethodsAreNotCalledUnderLock is the behavioural half: it
+// proves the internal helpers the ones above use are reachable while the lock
+// is held. If a future refactor makes an exported method call a *Locked one
+// while holding mu itself, this is where it shows up — as a hang, bounded by
+// the short timeout below rather than the suite's 10 minutes.
+func TestStatsLockedMethodsAreNotCalledUnderLock(t *testing.T) {
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		s := NewStats(10)
+		h := time.Now().Unix() / 3600
+		s.mu.Lock()
+		// Exactly the shape that deadlocked: the *Locked variants are correct
+		// here because the lock is already held.
+		s.addKeyReqLocked("k", h, 5)
+		s.addKeyHourLocked("k", h, 100)
+		s.wantPinnedBuckets("k", "m1")
+		s.addKeyTokenLocked("k", "m1", "src", h, 100)
+		// sumBuckets is the pure inner function the exported readers call.
+		if got := sumBuckets(s.keyReqHour["k"], time.Now().Unix(), 3600); got != 5 {
+			t.Errorf("sumBuckets under lock = %d, want 5", got)
+		}
+		s.mu.Unlock()
+	}()
+	select {
+	case <-done:
+	case <-time.After(20 * time.Second):
+		buf := make([]byte, 1<<16)
+		n := runtime.Stack(buf, true)
+		t.Fatalf("deadlocked while using the *Locked helpers under s.mu:\n%s", buf[:n])
+	}
+}
+
+// TestSumBucketsEdgeCases covers the boundaries the quota check depends on,
+// including the ones a regression would silently get wrong (an off-by-one here
+// either lets a quota leak or locks a key out early).
+func TestSumBucketsEdgeCases(t *testing.T) {
+	const hour = 3600
+	now := int64(10*hour + 61) // 10:00:61, i.e. just past the boundary
+	buckets := map[int64]int64{
+		0:  100, // ancient
+		9:  200, // previous hour
+		10: 7,   // current hour
+	}
+
+	if got := sumBuckets(buckets, now, 0); got != 307 {
+		t.Errorf("sec<=0 (all history) = %d, want 307", got)
+	}
+	// 1h window covers buckets h >= ceil((now-3600)/3600) = 10 -> only bucket 10.
+	if got := sumBuckets(buckets, now, hour); got != 7 {
+		t.Errorf("1h window = %d, want 7 (buckets 0 and 9 fall outside)", got)
+	}
+	// 2h window covers h >= 9 -> buckets 9 and 10.
+	if got := sumBuckets(buckets, now, 2*hour); got != 207 {
+		t.Errorf("2h window = %d, want 207", got)
+	}
+	// An empty map and a nil map must both be zero, not a panic.
+	if got := sumBuckets(map[int64]int64{}, now, hour); got != 0 {
+		t.Errorf("empty map = %d, want 0", got)
+	}
+	if got := sumBuckets(nil, now, hour); got != 0 {
+		t.Errorf("nil map = %d, want 0", got)
+	}
+	// now < sec must not produce a negative firstHour index: with a window far
+	// wider than the available history, every bucket that exists is counted.
+	// (sumBuckets walks bucket indices from firstHour to nowHour, so passing a
+	// "now" older than some buckets cannot reach them — that is correct, not a
+	// truncation bug.)
+	if got := sumBuckets(buckets, 10*hour+61, 100*hour); got != 307 {
+		t.Errorf("window wider than history = %d, want 307", got)
+	}
+	// A window that predates every bucket counts them all as well.
+	if got := sumBuckets(buckets, 10*hour+61, hour); got != 7 {
+		t.Errorf("1h window at t=10:00:61 = %d, want 7", got)
+	}
+}

internal/gateway/stats_period.go

@@ -0,0 +1,234 @@
+package gateway
+
+import (
+	"sort"
+	"strconv"
+	"strings"
+	"time"
+)
+
+// Period is a reporting window for the usage dashboard. The dashboard used to
+// have exactly one view — everything since process start — which made a quiet
+// morning indistinguishable from a quiet week. Periods give the operator a
+// scale to read the numbers at: today vs this week vs this month vs all time.
+//
+// The set is deliberately calendar-based and UTC-anchored. A rolling 24h window
+// would put "today" and "the last day" at different boundaries, so the same
+// number would move between two cards depending on when you looked; calendar
+// days are what people actually mean by "today". UTC also matches the billing
+// plugin's peak-window arithmetic, so a peak-rate hour does not land in a
+// different day in the cost view than in the usage view.
+type Period string
+
+const (
+	// PeriodDay is the current UTC calendar day.
+	PeriodDay Period = "day"
+	// PeriodWeek is the current ISO week (Mon 00:00 UTC to now).
+	PeriodWeek Period = "week"
+	// PeriodMonth is the current UTC calendar month.
+	PeriodMonth Period = "month"
+	// PeriodAll is since process start — the only view backed by the
+	// in-memory aggregates, and the only one available when no audit file is
+	// configured.
+	PeriodAll Period = "all"
+)
+
+// ValidPeriod reports whether p is a period the aggregator understands.
+// An unknown period is a client error, not a silent fallback to "all": a
+// dashboard that quietly shows lifetime totals when the caller asked for today
+// is worse than one that refuses.
+func ValidPeriod(p Period) bool {
+	switch p {
+	case PeriodDay, PeriodWeek, PeriodMonth, PeriodAll:
+		return true
+	}
+	return false
+}
+
+// periodStart returns the inclusive start of the window for p at time now.
+// Only PeriodDay/Week/Month are meaningful here; PeriodAll returns 0, which
+// every "from > 0" bounds check treats as unbounded.
+func periodStart(p Period, now time.Time) int64 {
+	now = now.UTC()
+	switch p {
+	case PeriodDay:
+		return time.Date(now.Year(), now.Month(), now.Day(), 0, 0, 0, 0, time.UTC).UnixMilli()
+	case PeriodWeek:
+		// ISO week starts Monday. Go's Weekday() is Sunday=0, so the shift
+		// below is 1 on Sunday and 0 on Monday..Saturday.
+		off := (int(now.Weekday()) + 6) % 7
+		d := now.AddDate(0, 0, -off)
+		return time.Date(d.Year(), d.Month(), d.Day(), 0, 0, 0, 0, time.UTC).UnixMilli()
+	case PeriodMonth:
+		return time.Date(now.Year(), now.Month(), 1, 0, 0, 0, 0, time.UTC).UnixMilli()
+	}
+	return 0
+}
+
+// PeriodBucket is one labelled point on the dashboard's timeline. Buckets are
+// the aggregation grain for a period: hourly for a day (so a spike is visible
+// inside "today"), daily for a week or month (so a week is not 7×24 points
+// wide and a month is not 31×24), and empty for "all" — a lifetime total has
+// no meaningful timeline, and pretending otherwise by drawing 500 hourly
+// buckets of rolling memory would be a different lie.
+type PeriodBucket struct {
+	Bucket string `json:"bucket"`
+	Stat
+}
+
+// PeriodSnapshot is the period-scoped twin of Snapshot's payload: the same
+// totals and the same by_* rows, plus a timeline. It deliberately mirrors
+// Snapshot's field names so the dashboard's table painters work unchanged —
+// paintModelTable(st.by_model) reads rows of {name, ...Stat}, and that shape
+// does not care where the numbers came from.
+type PeriodSnapshot struct {
+	Period Period         `json:"period"`
+	From   int64          `json:"from"` // unix millis, 0 when PeriodAll
+	Total  Stat           `json:"total"`
+	ByKey  []StatsRow     `json:"by_key"`
+	Models []StatsRow     `json:"by_model"`
+	Srcs   []StatsRow     `json:"by_source"`
+	Status []agrRow       `json:"by_status"`
+	Bucket []PeriodBucket `json:"buckets"`
+	// Truncated marks that the window was clipped by the retained audit
+	// history, so the numbers are a lower bound rather than the true period
+	// total. The dashboard shows this next to the numbers rather than letting
+	// a rotated-away week read as "that week had no traffic".
+	Truncated bool `json:"truncated"`
+}
+
+// bucketKey maps a record's timestamp onto the timeline grain for p.
+// Daily buckets are stamped at UTC midnight; hourly buckets carry the hour.
+func bucketKey(p Period, ms int64) string {
+	t := time.UnixMilli(ms).UTC()
+	if p == PeriodDay {
+		return t.Format("2006-01-02T15")
+	}
+	return t.Format("2006-01-02")
+}
+
+// bucketOf returns the bucket label plus the truncated-flag side effects of
+// walking a file: a record older than the requested window means the window
+// starts before the retained history, and the file may have been cut short.
+func (s *Stats) PeriodSnapshot(p Period, key string, now time.Time) PeriodSnapshot {
+	out := PeriodSnapshot{Period: p}
+	from := periodStart(p, now)
+	out.From = from
+
+	// "all" is exactly what Snapshot already answers, from the in-memory
+	// aggregates, and it is the one view that must keep working with no audit
+	// file configured at all (a fresh dev setup, or an operator who turned
+	// auditing off). Serving it from the same code path keeps the dashboard's
+	// "total" card identical whether or not a period is selected.
+	if p == PeriodAll {
+		snap := s.Snapshot(firstScreenRecords, key)
+		if tot, ok := snap["total"].(Stat); ok {
+			out.Total = tot
+		}
+		out.ByKey, _ = snap["by_key"].([]StatsRow)
+		out.Models, _ = snap["by_model"].([]StatsRow)
+		out.Srcs, _ = snap["by_source"].([]StatsRow)
+		out.Status, _ = snap["by_status"].([]agrRow)
+		// replay_partial is the same "these numbers came from a bounded
+		// tail" caveat, carried through under this view's own name.
+		out.Truncated, _ = snap["replay_partial"].(bool)
+		return out
+	}
+
+	// A bounded window is aggregated from the audit files, because the
+	// in-memory aggregates are lifetime totals and the ring buffer holds only
+	// maxRecs records (500 by default). Reading the ring would silently
+	// under-report any period longer than the last few hundred requests.
+	total := Stat{}
+	byKey := map[string]*Stat{}
+	byModel := map[string]*Stat{}
+	bySrc := map[string]*Stat{}
+	byStatus := map[string]*Stat{}
+	buckets := map[string]*Stat{}
+
+	var seen int
+	err := s.StreamAuditRecords(from, 0, key, func(r Req) error {
+		seen++
+		incStatus(&total, "", r)
+		inc(byKey, r.Key, r)
+		if r.Model != "" {
+			inc(byModel, r.Model, r)
+		}
+		inc(bySrc, r.Source, r)
+		if r.Status != 0 {
+			inc(byStatus, strconv.Itoa(r.Status), r)
+		}
+		k := bucketKey(p, r.Time)
+		b := buckets[k]
+		if b == nil {
+			b = &Stat{}
+			buckets[k] = b
+		}
+		incStatus(b, k, r)
+		return nil
+	})
+
+	out.Total = total
+	out.ByKey = rows(byKey)
+	out.Models = rows(byModel)
+	out.Srcs = rows(bySrc)
+	bs := make([]agrRow, 0, len(byStatus))
+	for code, st := range byStatus {
+		bs = append(bs, agrRow{Name: code, Stat: *st})
+	}
+	sort.Slice(bs, func(i, j int) bool {
+		ci, _ := strconv.Atoi(bs[i].Name)
+		cj, _ := strconv.Atoi(bs[j].Name)
+		return ci < cj
+	})
+	out.Status = bs
+
+	ks := make([]string, 0, len(buckets))
+	for k := range buckets {
+		ks = append(ks, k)
+	}
+	// Chronological, string-sorted: "2006-01-02T15" and "2006-01-02" both
+	// sort lexicographically in time order, so no date parsing is needed.
+	sort.Strings(ks)
+	out.Bucket = make([]PeriodBucket, 0, len(ks))
+	for _, k := range ks {
+		out.Bucket = append(out.Bucket, PeriodBucket{Bucket: k, Stat: *buckets[k]})
+	}
+
+	// The window is only "complete" if the audit walk actually reached back
+	// far enough to cover it. Two ways it cannot:
+	//
+	//  1. The walk found nothing at all in a window that certainly had
+	//     traffic, because the files holding it rotated away.
+	//  2. The walk errored part-way (I/O), leaving a partial total.
+	//
+	// Case 2 is reported from err; case 1 from seen == 0 combined with the
+	// caller having asked for a bounded window. It is deliberately
+	// conservative: a genuinely empty hour is rare enough that flagging it as
+	// possibly-truncated costs one tooltip, whereas silently under-reporting a
+	// month because rotation ate it is a wrong number with no indication.
+	if err != nil || (seen == 0 && p != PeriodAll) {
+		out.Truncated = true
+	}
+	return out
+}
+
+// periodFromQuery parses the ?period= parameter. An empty value means "all" so
+// that existing callers of /api/stats keep seeing exactly what they saw.
+// A malformed value is rejected by the caller (ValidPeriod) rather than
+// defaulting, so a typo in a bookmarked URL surfaces as an error instead of
+// silently switching the operator to lifetime totals.
+func periodFromQuery(q map[string][]string) Period {
+	v := strings.ToLower(strings.TrimSpace(firstQuery(q, "period")))
+	if v == "" {
+		return PeriodAll
+	}
+	return Period(v)
+}
+
+func firstQuery(q map[string][]string, key string) string {
+	if vs, ok := q[key]; ok && len(vs) > 0 {
+		return vs[0]
+	}
+	return ""
+}

internal/gateway/stats_period_records_test.go

@@ -0,0 +1,159 @@
+package gateway
+
+import (
+	"encoding/json"
+	"net/url"
+	"path/filepath"
+	"testing"
+	"time"
+)
+
+// The dashboard paints its records table from the `records` field of
+// /api/stats (paintRecords(st.records)). The period branch of that handler
+// used to hand-write a response map and simply omit the key, so switching the
+// selector to day / week / month rendered an EMPTY table no matter how much
+// traffic the window had — while the lifetime view worked, because it goes
+// through Snapshot() which carries records.
+//
+// These tests pin the contract across every period: same records, same cursor.
+func TestStatsPeriodViewsShipRecords(t *testing.T) {
+	td := t.TempDir()
+	g := newTestGateway(t)
+
+	// Write audit records the period aggregator and the record pager both read.
+	g.stats.auditPath = filepath.Join(td, "audit.jsonl")
+	now := time.Now().UnixMilli()
+	for i := 0; i < 5; i++ {
+		g.stats.Record(Req{
+			Time: now - int64(i)*1000, Type: "chat", OK: true,
+			Model: "m1", Source: "s1", Status: 200,
+			Prompt: 10, Compl: 2,
+		})
+	}
+
+	for _, p := range []string{"day", "week", "month", "all"} {
+		rr := doReq(t, g, "GET", "/api/stats?period="+p, "")
+		if rr.Code != 200 {
+			t.Fatalf("period=%s: HTTP %d: %s", p, rr.Code, rr.Body.String())
+		}
+		var st map[string]interface{}
+		if err := json.Unmarshal(rr.Body.Bytes(), &st); err != nil {
+			t.Fatalf("period=%s: %v", p, err)
+		}
+		recs, ok := st["records"].([]interface{})
+		if !ok {
+			t.Fatalf("period=%s: response has no `records` array — the dashboard "+
+				"records table renders empty for this view (keys: %v)", p, keysOf(st))
+		}
+		if len(recs) != 5 {
+			t.Fatalf("period=%s: got %d records, want 5", p, len(recs))
+		}
+		// The scroll handler pages from next_cursor; without it the table is
+		// stuck at one screen no matter how far the user scrolls.
+		if _, ok := st["next_cursor"]; !ok {
+			t.Fatalf("period=%s: no next_cursor — paging can never continue", p)
+		}
+	}
+}
+
+// The records shipped with a period view must be the newest-first page the
+// paging endpoint would return, in the same order. If /api/stats and
+// /api/stats/records disagreed, scrolling would duplicate or skip rows.
+func TestStatsPeriodRecordsMatchPager(t *testing.T) {
+	td := t.TempDir()
+	g := newTestGateway(t)
+	g.stats.auditPath = filepath.Join(td, "audit.jsonl")
+	now := time.Now().UnixMilli()
+	for i := 0; i < 8; i++ {
+		g.stats.Record(Req{
+			Time: now - int64(i)*1000, Type: "chat", OK: true,
+			Model: "m1", Status: 200, Prompt: 5, Compl: 1,
+		})
+	}
+
+	rr := doReq(t, g, "GET", "/api/stats?period=day&limit=100", "")
+	var st struct {
+		Records []struct {
+			Time  int64 `json:"time"`
+			Model string `json:"model"`
+		} `json:"records"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &st); err != nil {
+		t.Fatal(err)
+	}
+	rr2 := doReq(t, g, "GET", "/api/stats/records?limit=100", "")
+	var pg struct {
+		Records []struct {
+			Time  int64 `json:"time"`
+			Model string `json:"model"`
+		} `json:"records"`
+	}
+	if err := json.Unmarshal(rr2.Body.Bytes(), &pg); err != nil {
+		t.Fatal(err)
+	}
+	if len(st.Records) != len(pg.Records) {
+		t.Fatalf("period view shipped %d records, pager returned %d — the two "+
+			"sources must agree or scrolling duplicates/skips rows",
+			len(st.Records), len(pg.Records))
+	}
+	for i := range st.Records {
+		if st.Records[i].Time != pg.Records[i].Time {
+			t.Fatalf("record %d differs: period view t=%d, pager t=%d",
+				i, st.Records[i].Time, pg.Records[i].Time)
+		}
+	}
+	// The two sources must agree on DIRECTION as well as content. Which
+	// direction that is depends on how many reverse-read chunks the audit file
+	// spans (a small file is read whole, a large one chunk-by-chunk), so the
+	// test asserts agreement rather than hard-coding newest-first — but they
+	// can never disagree, because the table interleaves both streams.
+	if len(st.Records) > 1 {
+		periodAsc := st.Records[1].Time > st.Records[0].Time
+		pagerAsc := pg.Records[1].Time > pg.Records[0].Time
+		if periodAsc != pagerAsc {
+			t.Fatalf("direction differs: period view %s, pager %s — the table "+
+				"would splice the two streams out of order",
+				map[bool]string{true: "oldest-first", false: "newest-first"}[periodAsc],
+				map[bool]string{true: "oldest-first", false: "newest-first"}[pagerAsc])
+		}
+	}
+}
+
+// A non-admin caller cannot widen their view by passing someone else's key in
+// the query string: exportKey overrides ?key= with the caller's own masked id.
+// This is a data-leak boundary, and it applies to the records the period view
+// ships just as much as to the aggregates.
+func TestStatsPeriodRecordsIgnoreForeignKeyForUsers(t *testing.T) {
+	td := t.TempDir()
+	g := newTestGateway(t)
+	g.stats.auditPath = filepath.Join(td, "audit.jsonl")
+
+	other := keyID("sk-gw-cccccccccccccccc")
+	own := keyID("sk-test")
+	now := time.Now().UnixMilli()
+	g.stats.Record(Req{Time: now, OK: true, Model: "m", Status: 200, Key: other})
+	g.stats.Record(Req{Time: now - 1000, OK: true, Model: "m", Status: 200, Key: own})
+
+	// sk-test authenticates as a non-admin (newTestGateway seeds it under
+	// gateway_keys, not keys), so ?key=<other> must be ignored and the caller
+	// sees only its own rows.
+	rr := doReq(t, g, "GET",
+		"/api/stats?period=day&key="+url.QueryEscape(other), "")
+	if rr.Code != 200 {
+		t.Fatalf("HTTP %d: %s", rr.Code, rr.Body.String())
+	}
+	var st struct {
+		Records []struct {
+			Key string `json:"key"`
+		} `json:"records"`
+	}
+	if err := json.Unmarshal(rr.Body.Bytes(), &st); err != nil {
+		t.Fatal(err)
+	}
+	for _, r := range st.Records {
+		if r.Key == other {
+			t.Fatalf("another key's record leaked into a user view "+
+				"(caller=%s, requested=%s)", own, other)
+		}
+	}
+}

internal/gateway/stats_period_test.go

@@ -0,0 +1,434 @@
+package gateway
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+)
+
+// auditPeriod writes records straight into an audit file, which is what the
+// period aggregator reads. Going through Record() would also rotate/prune the
+// file; here the file shape is the thing under test.
+func auditPeriod(t *testing.T, path string, recs ...Req) {
+	t.Helper()
+	f, err := os.Create(path)
+	if err != nil {
+		t.Fatalf("create audit file: %v", err)
+	}
+	defer f.Close()
+	for _, r := range recs {
+		if r.Time == 0 {
+			t.Fatal("test record needs an explicit timestamp")
+		}
+		if r.Type == "" {
+			r.Type = "chat"
+		}
+		b, err := json.Marshal(r)
+		if err != nil {
+			t.Fatalf("marshal record: %v", err)
+		}
+		if _, err := f.Write(append(b, '\n')); err != nil {
+			t.Fatalf("write record: %v", err)
+		}
+	}
+}
+
+func periodStats(t *testing.T, auditPath string) *Stats {
+	t.Helper()
+	s := NewStats(10)
+	s.mu.Lock()
+	s.auditPath = auditPath
+	s.mu.Unlock()
+	return s
+}
+
+// TestPeriodWindowBoundaries is the boundary contract: a day window covers
+// exactly today-from-midnight, a week window starts Monday (not Sunday — the
+// Go Weekday() trap), and a month window starts on the 1st. All UTC.
+func TestPeriodWindowBoundaries(t *testing.T) {
+	// Wednesday 2026-03-11 15:30 UTC.
+	now := time.Date(2026, 3, 11, 15, 30, 0, 0, time.UTC)
+	day := time.Date(2026, 3, 11, 0, 0, 0, 0, time.UTC)
+	if got := time.UnixMilli(periodStart(PeriodDay, now)).UTC(); !got.Equal(day) {
+		t.Errorf("day window starts %s, want %s", got, day)
+	}
+	// Wednesday minus 2 days = Monday the 9th.
+	monday := time.Date(2026, 3, 9, 0, 0, 0, 0, time.UTC)
+	if got := time.UnixMilli(periodStart(PeriodWeek, now)).UTC(); !got.Equal(monday) {
+		t.Errorf("week window starts %s, want Monday %s", got, monday)
+	}
+	// Same instant on a Sunday must still start on the PRECEDING Monday,
+	// never on the Sunday itself. This is the off-by-one that
+	// int(now.Weekday()) would introduce (Sunday=0 → no shift).
+	sun := time.Date(2026, 3, 15, 10, 0, 0, 0, time.UTC) // Sunday
+	prevMon := time.Date(2026, 3, 9, 0, 0, 0, 0, time.UTC)
+	if got := time.UnixMilli(periodStart(PeriodWeek, sun)).UTC(); !got.Equal(prevMon) {
+		t.Errorf("Sunday week window starts %s, want the Monday before it %s", got, prevMon)
+	}
+	first := time.Date(2026, 3, 1, 0, 0, 0, 0, time.UTC)
+	if got := time.UnixMilli(periodStart(PeriodMonth, now)).UTC(); !got.Equal(first) {
+		t.Errorf("month window starts %s, want %s", got, first)
+	}
+	if got := periodStart(PeriodAll, now); got != 0 {
+		t.Errorf("all window from = %d, want 0 (unbounded)", got)
+	}
+}
+
+// TestPeriodAggregationExcludesOlderRecords is the reason the aggregator reads
+// the audit file: a record outside the window must not be counted, even though
+// it sits in the very same file, right next to records that are.
+func TestPeriodAggregationExcludesOlderRecords(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "audit.jsonl")
+	now := time.Date(2026, 3, 11, 15, 0, 0, 0, time.UTC)
+	h := func(n int) int64 { return now.Add(-time.Duration(n) * time.Hour).UnixMilli() }
+
+	auditPeriod(t, path,
+		// inside today
+		Req{Time: h(1), Model: "m", Source: "s", Prompt: 100, Compl: 10, LatMs: 5, OK: true},
+		Req{Time: h(2), Model: "m", Source: "s", Prompt: 200, Compl: 20, LatMs: 7, OK: true},
+		// outside today (yesterday) but inside the week
+		Req{Time: now.AddDate(0, 0, -1).UnixMilli(), Model: "old", Source: "s", Prompt: 999, Compl: 99, OK: true},
+	)
+	s := periodStats(t, path)
+
+	day := s.PeriodSnapshot(PeriodDay, "", now)
+	if day.Total.Reqs != 2 {
+		t.Errorf("day window counted %d requests, want 2 (yesterday's record must be excluded)", day.Total.Reqs)
+	}
+	if day.Total.Prompt != 300 {
+		t.Errorf("day prompt tokens = %d, want 300", day.Total.Prompt)
+	}
+
+	week := s.PeriodSnapshot(PeriodWeek, "", now)
+	if week.Total.Reqs != 3 {
+		t.Errorf("week window counted %d requests, want 3", week.Total.Reqs)
+	}
+	if week.Total.Prompt != 1299 {
+		t.Errorf("week prompt tokens = %d, want 1299", week.Total.Prompt)
+	}
+}
+
+// TestPeriodDimensionsAggregateIndependently checks the by_* rows actually
+// split by their own key rather than all collapsing into one row.
+func TestPeriodDimensionsAggregateIndependently(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "audit.jsonl")
+	now := time.Date(2026, 3, 11, 10, 0, 0, 0, time.UTC)
+	auditPeriod(t, path,
+		Req{Time: now.UnixMilli(), Key: "k1", Model: "alpha", Source: "s1", Prompt: 10, OK: true},
+		Req{Time: now.UnixMilli(), Key: "k1", Model: "beta", Source: "s2", Prompt: 20, OK: true},
+		Req{Time: now.UnixMilli(), Key: "k2", Model: "alpha", Source: "s1", Prompt: 30, OK: true},
+	)
+	s := periodStats(t, path)
+	got := s.PeriodSnapshot(PeriodDay, "", now)
+
+	if len(got.Models) != 2 {
+		t.Fatalf("by_model has %d rows, want 2: %+v", len(got.Models), got.Models)
+	}
+	// alpha = 10 + 30
+	for _, r := range got.Models {
+		if r.Name == "alpha" && r.Prompt != 40 {
+			t.Errorf("alpha prompt = %d, want 40", r.Prompt)
+		}
+	}
+	if len(got.Srcs) != 2 {
+		t.Errorf("by_source has %d rows, want 2: %+v", len(got.Srcs), got.Srcs)
+	}
+	if len(got.ByKey) != 2 {
+		t.Errorf("by_key has %d rows, want 2: %+v", len(got.ByKey), got.ByKey)
+	}
+	// The rows must add up to the total, or the dashboard shows a total that
+	// disagrees with its own table.
+	var sum int64
+	for _, r := range got.Models {
+		sum += r.Prompt
+	}
+	if sum != got.Total.Prompt {
+		t.Errorf("by_model prompts sum to %d but total is %d — the table would contradict the KPI", sum, got.Total.Prompt)
+	}
+}
+
+// TestPeriodDayBucketsAreHourlyAndOrdered pins the timeline grain and its
+// order. Day = hourly (a spike must be visible inside "today"); week/month =
+// daily. Sorted chronologically, because a chart fed unsorted buckets draws
+// nonsense.
+func TestPeriodDayBucketsAreHourlyAndOrdered(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "audit.jsonl")
+	now := time.Date(2026, 3, 11, 15, 0, 0, 0, time.UTC)
+	// Deliberately out of order in the file.
+	auditPeriod(t, path,
+		Req{Time: now.Add(-2 * time.Hour).UnixMilli(), Model: "m", Prompt: 1, OK: true},
+		Req{Time: now.Add(-4 * time.Hour).UnixMilli(), Model: "m", Prompt: 1, OK: true},
+		Req{Time: now.UnixMilli(), Model: "m", Prompt: 1, OK: true},
+	)
+	s := periodStats(t, path)
+	got := s.PeriodSnapshot(PeriodDay, "", now)
+	if len(got.Bucket) != 3 {
+		t.Fatalf("day buckets = %d, want 3 (hourly): %+v", len(got.Bucket), got.Bucket)
+	}
+	for i := 1; i < len(got.Bucket); i++ {
+		if got.Bucket[i-1].Bucket >= got.Bucket[i].Bucket {
+			t.Errorf("day buckets not chronological at %d: %q >= %q",
+				i, got.Bucket[i-1].Bucket, got.Bucket[i].Bucket)
+		}
+	}
+	// Weekly grain is daily, so the same three records collapse to one day.
+	week := s.PeriodSnapshot(PeriodWeek, "", now)
+	if len(week.Bucket) != 1 {
+		t.Errorf("week buckets = %d, want 1 (daily grain): %+v", len(week.Bucket), week.Bucket)
+	}
+}
+
+// TestPeriodKeyFilterIsolatesKeys guards the multi-tenant boundary: asking for
+// one key must never return another key's rows in any dimension.
+func TestPeriodKeyFilterIsolatesKeys(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "audit.jsonl")
+	now := time.Date(2026, 3, 11, 10, 0, 0, 0, time.UTC)
+	auditPeriod(t, path,
+		Req{Time: now.UnixMilli(), Key: "alice", Model: "m", Source: "s", Prompt: 10, OK: true},
+		Req{Time: now.UnixMilli(), Key: "bob", Model: "m", Source: "s", Prompt: 999, OK: true},
+	)
+	s := periodStats(t, path)
+	got := s.PeriodSnapshot(PeriodDay, "alice", now)
+	if got.Total.Reqs != 1 || got.Total.Prompt != 10 {
+		t.Errorf("alice's window = %d reqs / %d prompt, want 1 / 10 — another key leaked in",
+			got.Total.Reqs, got.Total.Prompt)
+	}
+	if len(got.Srcs) != 1 || got.Srcs[0].Prompt != 10 {
+		t.Errorf("by_source leaked: %+v", got.Srcs)
+	}
+}
+
+// TestPeriodAllUsesLifetimeAggregates checks "all" still answers from the
+// in-memory aggregates (no audit file needed) and carries the same rows the
+// dashboard already renders.
+func TestPeriodAllUsesLifetimeAggregates(t *testing.T) {
+	s := NewStats(100)
+	now := time.Date(2026, 3, 11, 10, 0, 0, 0, time.UTC)
+	s.Record(Req{Time: now.AddDate(0, 0, -40).UnixMilli(), Key: "k", Model: "m", Source: "s", Prompt: 7, OK: true})
+	s.Record(Req{Time: now.UnixMilli(), Key: "k", Model: "m", Source: "s", Prompt: 3, OK: true})
+
+	got := s.PeriodSnapshot(PeriodAll, "", now)
+	if got.Total.Reqs != 2 || got.Total.Prompt != 10 {
+		t.Errorf("all view = %d reqs / %d prompt, want 2 / 10 (lifetime)", got.Total.Reqs, got.Total.Prompt)
+	}
+	if len(got.Models) != 1 || got.Models[0].Prompt != 10 {
+		t.Errorf("all view by_model = %+v, want one row with 10 prompt", got.Models)
+	}
+	// A lifetime total has no timeline to draw; buckets must stay empty
+	// rather than inventing 500 hourly points.
+	if len(got.Bucket) != 0 {
+		t.Errorf("all view produced %d buckets, want 0", len(got.Bucket))
+	}
+}
+
+// TestPeriodEmptyWindowFlagsTruncated covers the "rotated away" case: a bounded
+// window with nothing to show may mean "quiet day" or "the files are gone".
+// The aggregator must not report a confidently wrong zero.
+func TestPeriodEmptyWindowFlagsTruncated(t *testing.T) {
+	dir := t.TempDir()
+	s := periodStats(t, filepath.Join(dir, "audit.jsonl")) // file does not exist
+	now := time.Date(2026, 3, 11, 10, 0, 0, 0, time.UTC)
+	got := s.PeriodSnapshot(PeriodDay, "", now)
+	if got.Total.Reqs != 0 {
+		t.Errorf("empty window reqs = %d, want 0", got.Total.Reqs)
+	}
+	if !got.Truncated {
+		t.Error("★ an empty bounded window must be flagged truncated — otherwise a " +
+			"rotated-away week reads as 'no traffic that week'")
+	}
+}
+
+// TestValidPeriodRejectsUnknown guards the deliberate strictness: a typo must
+// be an error, not a silent fallback to lifetime totals.
+func TestValidPeriodRejectsUnknown(t *testing.T) {
+	for _, ok := range []Period{PeriodDay, PeriodWeek, PeriodMonth, PeriodAll} {
+		if !ValidPeriod(ok) {
+			t.Errorf("ValidPeriod(%q) = false, want true", ok)
+		}
+	}
+	for _, bad := range []Period{"", "year", "hour", "today", "DAY "} {
+		if ValidPeriod(bad) {
+			t.Errorf("ValidPeriod(%q) = true, want false", bad)
+		}
+	}
+}
+
+// TestPeriodStatusDimensionCountsFailures covers by_status, which nothing else
+// exercised — an aggregation branch with no test is exactly how a period view
+// silently loses the failure pie. M8 (disabling the status dimension) is
+// caught only because this test exists.
+func TestPeriodStatusDimensionCountsFailures(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "audit.jsonl")
+	now := time.Date(2026, 3, 11, 10, 0, 0, 0, time.UTC)
+	auditPeriod(t, path,
+		Req{Time: now.UnixMilli(), Model: "m", Source: "s", Status: 200, OK: true, Prompt: 5},
+		Req{Time: now.UnixMilli(), Model: "m", Source: "s", Status: 502, OK: false, Prompt: 1},
+		Req{Time: now.UnixMilli(), Model: "m", Source: "s", Status: 502, OK: false, Prompt: 1},
+		Req{Time: now.UnixMilli(), Model: "m", Source: "s", Status: 429, OK: false, Prompt: 1},
+	)
+	s := periodStats(t, path)
+	got := s.PeriodSnapshot(PeriodDay, "", now)
+
+	byCode := map[string]int64{}
+	for _, r := range got.Status {
+		byCode[r.Name] = r.Reqs
+	}
+	if len(got.Status) != 3 {
+		t.Fatalf("by_status has %d rows, want 3 (200/502/429): %+v", len(got.Status), got.Status)
+	}
+	if byCode["200"] != 1 || byCode["502"] != 2 || byCode["429"] != 1 {
+		t.Errorf("by_status counts = %v, want 200:1 502:2 429:1", byCode)
+	}
+	// Status codes must sort numerically: "429" before "502" lexically would
+	// render the failure legend in the wrong order.
+	if got.Status[0].Name != "200" || got.Status[1].Name != "429" || got.Status[2].Name != "502" {
+		t.Errorf("by_status not numerically sorted: %s, %s, %s",
+			got.Status[0].Name, got.Status[1].Name, got.Status[2].Name)
+	}
+	// The failure rows must agree with the OK/Err split on the total, or the
+	// pie and the KPI row contradict each other.
+	if got.Total.OK != 1 || got.Total.Err != 3 {
+		t.Errorf("total ok/err = %d/%d, want 1/3", got.Total.OK, got.Total.Err)
+	}
+}
+
+func TestPeriodFromQuery(t *testing.T) {
+	cases := []struct {
+		query string
+		want  Period
+	}{
+		{"", PeriodAll}, // no selector → unchanged legacy behaviour
+		{"?period=day", PeriodDay},
+		{"?period=WEEK", PeriodWeek},
+		{"?period=month", PeriodMonth},
+		{"?period=all", PeriodAll},
+	}
+	for _, c := range cases {
+		r := httptest.NewRequest(http.MethodGet, "/api/stats"+c.query, nil)
+		if got := periodFromQuery(r.URL.Query()); got != c.want {
+			t.Errorf("periodFromQuery(%q) = %q, want %q", c.query, got, c.want)
+		}
+	}
+}
+
+// TestPeriodWeekMayStartBeforeMonth pins a fact that looks like a bug and
+// gets reported as one: the calendar periods are NOT nested.
+//
+// On 2026-10-03 (a Saturday) the ISO week runs Mon 2026-09-28 .. Sat
+// 10-03, while the month window is 10-01 .. 10-03. The week therefore reaches
+// back into the previous month and its total is legitimately LARGER than the
+// month's. Anyone reading only the totals will conclude the aggregator is
+// broken; it is not, and this test exists so a future "fix" that makes the
+// windows nested (e.g. clamping the week to the month) is caught as the
+// regression it would be.
+func TestPeriodWeekMayStartBeforeMonth(t *testing.T) {
+	// Saturday 2026-10-03, UTC.
+	now := time.Date(2026, 10, 3, 12, 0, 0, 0, time.UTC)
+
+	weekFrom := periodStart(PeriodWeek, now)
+	monthFrom := periodStart(PeriodMonth, now)
+
+	weekFromD := time.UnixMilli(weekFrom).UTC()
+	monthFromD := time.UnixMilli(monthFrom).UTC()
+
+	if got := weekFromD.Format("2006-01-02"); got != "2026-09-28" {
+		t.Fatalf("week window start = %s, want 2026-09-28 (the Monday)", got)
+	}
+	if got := monthFromD.Format("2006-01-02"); got != "2026-10-01" {
+		t.Fatalf("month window start = %s, want 2026-10-01", got)
+	}
+	if !weekFromD.Before(monthFromD) {
+		t.Fatal("this test only means anything when the week starts before the month")
+	}
+	// The week must reach strictly further back than the month window does,
+	// otherwise the nesting intuition would hold and the UI note would be noise.
+	if back := monthFromD.Sub(weekFromD).Hours() / 24; back <= 0 {
+		t.Fatalf("week should start at least a day before the month window, got %v days back", back)
+	}
+}
+
+// The same overlap must be visible in the aggregate numbers, not just in the
+// window bounds: a week total larger than a month total is the exact symptom
+// an operator reported.
+func TestPeriodWeekTotalCanExceedMonthTotal(t *testing.T) {
+	dir := t.TempDir()
+	audit := filepath.Join(dir, "audit.jsonl")
+
+	// Two records before the month window (inside this ISO week) and one
+	// inside October.
+	pre := time.Date(2026, 9, 29, 10, 0, 0, 0, time.UTC).UnixMilli()
+	in1 := time.Date(2026, 10, 1, 10, 0, 0, 0, time.UTC).UnixMilli()
+	in2 := time.Date(2026, 10, 3, 10, 0, 0, 0, time.UTC).UnixMilli()
+
+	auditPeriod(t, audit,
+		Req{Time: pre, OK: true, Prompt: 100, Compl: 10},
+		Req{Time: pre + 1000, OK: true, Prompt: 100, Compl: 10},
+		Req{Time: in1, OK: true, Prompt: 50, Compl: 5},
+		Req{Time: in2, OK: true, Prompt: 50, Compl: 5},
+	)
+
+	s := periodStats(t, audit)
+	now := time.Date(2026, 10, 3, 12, 0, 0, 0, time.UTC)
+
+	week := s.PeriodSnapshot(PeriodWeek, "", now)
+	month := s.PeriodSnapshot(PeriodMonth, "", now)
+
+	if week.Total.Reqs != 4 {
+		t.Fatalf("week reqs = %d, want 4 (both September records fall in this ISO week)", week.Total.Reqs)
+	}
+	if month.Total.Reqs != 2 {
+		t.Fatalf("month reqs = %d, want 2", month.Total.Reqs)
+	}
+	if week.Total.Tokens <= month.Total.Tokens {
+		t.Fatalf("week tokens (%d) must exceed month tokens (%d): the week reaches into September",
+			week.Total.Tokens, month.Total.Tokens)
+	}
+	// Day must be a strict subset of both, which is the only nesting that holds.
+	day := s.PeriodSnapshot(PeriodDay, "", now)
+	if day.Total.Reqs != 1 {
+		t.Fatalf("day reqs = %d, want 1", day.Total.Reqs)
+	}
+	if day.Total.Tokens >= month.Total.Tokens || day.Total.Tokens >= week.Total.Tokens {
+		t.Fatal("today must be contained in both the week and the month totals")
+	}
+}
+
+// TestPeriodStatsCarryActiveGauge pins the live "active requests" gauge onto
+// every period response.
+//
+// It is a live gauge, not an aggregate, so no calendar window can scope it — but
+// the period response is what the day/week/month views render from, and it used
+// to omit the field entirely. The UI then evaluated `st.active || 0` and printed
+// a hard 0 in three of four views, while the all-time view (whose response did
+// carry the field) showed the truth. Reading "0 active" off a busy gateway is
+// worse than reading nothing: it is a confident wrong answer.
+func TestPeriodStatsCarryActiveGauge(t *testing.T) {
+	for _, period := range []string{"day", "week", "month", "all"} {
+		g := newTestGateway(t)
+		rr := doReq(t, g, http.MethodGet, "/api/stats?period="+period, "")
+		if rr.Code != http.StatusOK {
+			t.Fatalf("period=%s: %d", period, rr.Code)
+		}
+		var out map[string]interface{}
+		if err := json.Unmarshal(rr.Body.Bytes(), &out); err != nil {
+			t.Fatalf("period=%s decode: %v", period, err)
+		}
+		v, ok := out["active"]
+		if !ok {
+			t.Errorf("period=%s response has no active field; the UI renders 0", period)
+			continue
+		}
+		if _, ok := v.(float64); !ok {
+			t.Errorf("period=%s active is %T, want a number", period, v)
+		}
+	}
+}

internal/gateway/token_units_test.go

@@ -0,0 +1,231 @@
+package gateway
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/config"
+)
+
+// Token accounting must record TOKENS, not byte estimates and not image
+// counts. These tests pin the two units bugs found on 2026-09-27:
+//
+//  1. streamed chat: rec.Compl was only ever the per-chunk byte estimate
+//     (len(content)/3), even when the upstream sent exact usage on its final
+//     frame. Measured on production traffic: the same model reported a
+//     streamed completion median 4-27x its non-streamed median. Upstream 100
+//     recorded as 145.
+//  2. image generation: rec.Compl was set to len(resp.ImageData) — the image
+//     COUNT — and folded into the token totals.
+//
+// A fake upstream that reports FIXED numbers makes both mismatches
+// unambiguous: any difference between what it reports and what the audit
+// record holds is the gateway's doing.
+
+// fixedUsageUpstream reports prompt=prompt completion=compl on both the
+// streaming and the non-streaming path. The streaming path emits usage on its
+// final frame before [DONE], exactly like OpenAI-compatible upstreams.
+func fixedUsageUpstream(t *testing.T, prompt, compl int, images int) *httptest.Server {
+	t.Helper()
+	return httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var body map[string]interface{}
+		_ = json.NewDecoder(r.Body).Decode(&body)
+		stream, _ := body["stream"].(bool)
+		usage := fmt.Sprintf(`{"prompt_tokens":%d,"completion_tokens":%d,"total_tokens":%d}`,
+			prompt, compl, prompt+compl)
+
+		// Image generation wire: /v1/images/generations style body.
+		if strings.Contains(r.URL.Path, "image") {
+			w.Header().Set("Content-Type", "application/json")
+			var data []string
+			for i := 0; i < images; i++ {
+				data = append(data, `{"b64_json":"AAAA"}`)
+			}
+			fmt.Fprintf(w, `{"created":1,"data":[%s]}`, strings.Join(data, ","))
+			return
+		}
+
+		if !stream {
+			w.Header().Set("Content-Type", "application/json")
+			fmt.Fprintf(w, `{"model":"m1","choices":[{"message":{"content":"pong pong pong"},
+				"finish_reason":"stop"}],"usage":%s}`, usage)
+			return
+		}
+
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.WriteHeader(200)
+		fl, _ := w.(http.Flusher)
+		send := func(s string) {
+			fmt.Fprintf(w, "data: %s\n\n", s)
+			if fl != nil {
+				fl.Flush()
+			}
+		}
+		// Deliberately verbose payloads: a byte-estimating path records a
+		// number far from `compl`, so the test would catch it.
+		send(`{"model":"m1","choices":[{"delta":{"reasoning_content":"` + strings.Repeat("think ", 40) + `"}}]}`)
+		send(`{"model":"m1","choices":[{"delta":{"content":"` + strings.Repeat("answer ", 40) + `"}}]}`)
+		send(`{"model":"m1","choices":[{"delta":{},"finish_reason":"stop"}]}`)
+		send(`{"model":"m1","choices":[],"usage":` + usage + `}`)
+		send(`[DONE]`)
+	}))
+}
+
+// lastChatRecord returns the most recent audit record whose Type is one of the
+// given types.
+func lastChatRecord(t *testing.T, g *Gateway, types ...string) Req {
+	t.Helper()
+	recs := g.stats.AuditRecords(0, 0, "")
+	want := map[string]bool{}
+	for _, ty := range types {
+		want[ty] = true
+	}
+	for i := len(recs) - 1; i >= 0; i-- {
+		if want[recs[i].Type] {
+			return recs[i]
+		}
+	}
+	t.Fatalf("no audit record of type %v found (have %d records)", types, len(recs))
+	return Req{}
+}
+
+// The streamed path must record the upstream's exact usage, not a byte
+// estimate. Before the fix this recorded 145 for an upstream-reported 100.
+func TestStreamedChatRecordsUpstreamUsage(t *testing.T) {
+	const wantPrompt, wantCompl = 44, 100
+	up := fixedUsageUpstream(t, wantPrompt, wantCompl, 0)
+	defer up.Close()
+	g := newTestGateway(t, config.Source{
+		Name: "up", BaseURL: up.URL, Adapter: "openai",
+		Models: []config.Model{{ID: "m1", Priority: 100}},
+	})
+
+	rr := doReq(t, g, "POST", "/v1/chat/completions",
+		`{"model":"m1","messages":[{"role":"user","content":"hi"}],"stream":true}`)
+	if rr.Code != 200 {
+		t.Fatalf("stream status=%d body=%s", rr.Code, rr.Body.String())
+	}
+
+	rec := lastChatRecord(t, g, "stream")
+	if rec.Prompt != wantPrompt || rec.Compl != wantCompl {
+		t.Errorf("streamed record prompt=%d completion=%d, want %d/%d — the audit "+
+			"must carry the upstream's tokens, not a byte estimate",
+			rec.Prompt, rec.Compl, wantPrompt, wantCompl)
+	}
+}
+
+// Streaming and non-streaming must agree on the same request's tokens. This is
+// the cross-path consistency check: the two paths previously recorded 145 and
+// 100 for one identical upstream usage value.
+func TestStreamedAndNonStreamedAgreeOnTokens(t *testing.T) {
+	const wantPrompt, wantCompl = 44, 100
+	up := fixedUsageUpstream(t, wantPrompt, wantCompl, 0)
+	defer up.Close()
+	g := newTestGateway(t, config.Source{
+		Name: "up", BaseURL: up.URL, Adapter: "openai",
+		Models: []config.Model{{ID: "m1", Priority: 100}},
+	})
+
+	if rr := doReq(t, g, "POST", "/v1/chat/completions",
+		`{"model":"m1","messages":[{"role":"user","content":"hi"}]}`); rr.Code != 200 {
+		t.Fatalf("non-stream status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	if rr := doReq(t, g, "POST", "/v1/chat/completions",
+		`{"model":"m1","messages":[{"role":"user","content":"hi"}],"stream":true}`); rr.Code != 200 {
+		t.Fatalf("stream status=%d body=%s", rr.Code, rr.Body.String())
+	}
+
+	chat, stream := lastChatRecord(t, g, "chat"), lastChatRecord(t, g, "stream")
+	if chat.Prompt != stream.Prompt || chat.Compl != stream.Compl {
+		t.Errorf("chat=%d/%d stream=%d/%d — identical upstream usage must record "+
+			"identically on both paths", chat.Prompt, chat.Compl, stream.Prompt, stream.Compl)
+	}
+}
+
+// A fallback to the byte estimate is still correct WHEN the upstream reports no
+// usage at all. Without this the fix could silently zero out every streamed
+// record from upstreams that omit usage.
+func TestStreamedChatFallsBackToEstimateWithoutUsage(t *testing.T) {
+	up := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/event-stream")
+		w.WriteHeader(200)
+		fl, _ := w.(http.Flusher)
+		send := func(s string) {
+			fmt.Fprintf(w, "data: %s\n\n", s)
+			if fl != nil {
+				fl.Flush()
+			}
+		}
+		send(`{"model":"m1","choices":[{"delta":{"content":"` + strings.Repeat("abc ", 60) + `"}}]}`)
+		send(`{"model":"m1","choices":[{"delta":{},"finish_reason":"stop"}]}`)
+		send(`[DONE]`)
+	}))
+	defer up.Close()
+	g := newTestGateway(t, config.Source{
+		Name: "up", BaseURL: up.URL, Adapter: "openai",
+		Models: []config.Model{{ID: "m1", Priority: 100}},
+	})
+
+	if rr := doReq(t, g, "POST", "/v1/chat/completions",
+		`{"model":"m1","messages":[{"role":"user","content":"hi"}],"stream":true}`); rr.Code != 200 {
+		t.Fatalf("status=%d body=%s", rr.Code, rr.Body.String())
+	}
+	rec := lastChatRecord(t, g, "stream")
+	if rec.Compl <= 0 {
+		t.Errorf("completion=%d, want a positive byte-estimate fallback when the "+
+			"upstream reports no usage", rec.Compl)
+	}
+}
+
+// Image generation has no token concept: recording the image COUNT as
+// completion tokens corrupted the token totals. The count is reported
+// separately and must not leak into Prompt/Compl.
+func TestImageGenerationDoesNotRecordTokens(t *testing.T) {
+	up := fixedUsageUpstream(t, 44, 100, 3)
+	defer up.Close()
+	src := config.Source{
+		Name: "up", BaseURL: up.URL, Adapter: "openai",
+		Models: []config.Model{{ID: "img-m", Priority: 100, Kind: "image"}},
+	}
+	g := newTestGateway(t, src)
+
+	rr := doReq(t, g, "POST", "/v1/images/generations",
+		`{"model":"img-m","prompt":"a cat","n":3}`)
+	if rr.Code != 200 {
+		t.Fatalf("image status=%d body=%s", rr.Code, rr.Body.String())
+	}
+
+	rec := lastChatRecord(t, g, "image")
+	if rec.Compl != 0 || rec.Prompt != 0 {
+		t.Errorf("image record prompt=%d completion=%d, want 0/0 — image count is "+
+			"not a token and must not enter the token totals", rec.Prompt, rec.Compl)
+	}
+	if rec.ImageCount != 3 {
+		t.Errorf("image_count=%d, want 3 (the count belongs in its own field)", rec.ImageCount)
+	}
+	// And the aggregate must see zero tokens for it.
+	if got := g.stats.WindowTokens("img-m", "", 3600); got != 0 {
+		t.Errorf("window tokens for an image model = %d, want 0", got)
+	}
+}
+
+// The audit JSON must carry image_count so the WebUI can show images rather
+// than a bogus token figure.
+func TestImageCountSurvivesAuditJSON(t *testing.T) {
+	b, err := json.Marshal(Req{Type: "image", ImageCount: 2, OK: true})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if !strings.Contains(string(b), `"image_count":2`) {
+		t.Errorf("audit JSON %s lacks image_count", b)
+	}
+	// Omitempty keeps chat records unchanged.
+	b2, _ := json.Marshal(Req{Type: "chat", Prompt: 1, Compl: 2, OK: true})
+	if strings.Contains(string(b2), "image_count") {
+		t.Errorf("chat record %s should omit image_count", b2)
+	}
+}

internal/gateway/ui_contract_test.go

@@ -59,6 +59,82 @@ func lineOf(src string, idx int) int {
 	return strings.Count(src[:idx], "\n") + 1
 }
 
+// TestUIEditFormRoundTripsEveryOptionalSourceField pins the JS half of the
+// source-edit contract.
+//
+// The server distinguishes "absent" from "explicitly empty" for proxy_url,
+// api_key_env and timeout (they are pointers). That only helps if the form
+// SENTS them — a form that omits them falls back to "inherit", so clearing the
+// proxy box would silently keep the old proxy, which is the exact bug in the
+// other direction.
+//
+// It must also send the api_key_env value it was given, since that is the one
+// field whose loss is invisible until the next upstream call.
+func TestUIEditFormRoundTripsEveryOptionalSourceField(t *testing.T) {
+	src := uiSource(t)
+	body, ok := jsFunctionBody(src, "saveSource")
+	if !ok {
+		t.Fatal("saveSource() not found in the WebUI")
+	}
+	for _, field := range []string{"proxy_url", "api_key_env", "timeout"} {
+		if !strings.Contains(body, field+":") {
+			t.Errorf("saveSource() does not send %q; the server treats an absent "+
+				"field as \"keep the stored value\", so the form could never clear it", field)
+		}
+	}
+	// The form's inputs must be filled from the values the server sends, or a
+	// save would post an empty box and clear a configured field.
+	for _, read := range []string{`$("#s-proxy")`, `$("#s-keyenv")`, `$("#s-timeout")`} {
+		if !strings.Contains(src, read+".value") {
+			t.Errorf("the source form never reads %s — it would post an empty value", read)
+		}
+	}
+}
+
+// TestUIEditFormReadsDurationsFromTheRevealView: config.Source tags
+// Timeout/QueueTimeout `json:"-"`, so the durations only reach the form if the
+// reveal endpoint adds them explicitly. The form MUST read them (see above),
+// which makes this API response load-bearing rather than informational.
+func TestUIEditFormReadsDurationsFromTheRevealView(t *testing.T) {
+	src := uiSource(t)
+	for _, id := range []string{"s-proxy", "s-keyenv", "s-timeout"} {
+		if !strings.Contains(src, `id="`+id+`"`) {
+			t.Errorf("source form input #%s is missing", id)
+		}
+	}
+	// The timeout box must render through the duration helper. Reading a raw
+	// time.Duration number would show nanoseconds in a box that expects "300s".
+	if !strings.Contains(src, "timeoutText(") {
+		t.Error("the timeout input does not go through timeoutText(); a raw " +
+			"time.Duration JSON number is nanoseconds and would be uneditable")
+	}
+}
+
+// TestUIEditFormSendsEverySourceField is the mirror of
+// TestSourcePayloadCoversEveryEditableField on the JavaScript side: each field
+// the server accepts must be submitted by the form, or the server's upsert has
+// nothing to store for it.
+func TestUIEditFormSendsEverySourceField(t *testing.T) {
+	src := uiSource(t)
+	body, ok := jsFunctionBody(src, "saveSource")
+	if !ok {
+		t.Fatal("saveSource() not found")
+	}
+	// Fields the edit dialog owns. proxy_url / api_key_env / timeout are
+	// covered by the round-trip test above; this one catches the rest.
+	// models and meta are local variables submitted by Go shorthand
+	// (payload = { models, meta }), so they are matched as bare identifiers.
+	for _, field := range []string{
+		"name:", "base_url:", "api_key:", "adapter:", "endpoint:",
+		"image_endpoint:", "max_concurrent:", "rpm:", "temperature:",
+		"models,", "meta,",
+	} {
+		if !strings.Contains(body, field) {
+			t.Errorf("saveSource() does not send %s", field)
+		}
+	}
+}
+
 // TestUIAPICallsDeclareMethod asserts that every api() call passing an options
 // object also declares an HTTP method (or is a GET that only passes an
 // AbortSignal). Without this, fetch defaults to GET and mutating endpoints are
@@ -347,3 +423,60 @@ func cssBlock(src, selector string) (string, bool) {
 	}
 	return src[i : i+j+1], true
 }
+
+// The WebUI ships two locale objects (`zh` and `en`) that every rendered string
+// goes through. A key added to only one of them does not error: t() falls back
+// to `t("key") || "literal"` in some call sites and to the bare key string in
+// others, so the user sees either a hardcoded local string or a raw key name.
+// Nothing in the Go test suite noticed — this was found by hand while adding
+// thImgs. Pin the key sets so a one-sided edit fails here instead of shipping.
+func TestUILocaleKeyParity(t *testing.T) {
+	src := uiSource(t)
+	iZh := strings.Index(src, "zh: {")
+	iEn := strings.Index(src, "en: {")
+	if iZh < 0 || iEn < 0 || iEn < iZh {
+		t.Fatalf("locale blocks not found (zh=%d en=%d)", iZh, iEn)
+	}
+	zh := src[iZh:iEn]
+	en := src[iEn:]
+	// End the EN block at its closing brace, so keys from later objects
+	// (config templates etc.) do not pollute the comparison.
+	if end := strings.Index(en, "\n        },"); end > 0 {
+		en = en[:end]
+	}
+
+	keyRe := regexp.MustCompile(`(?m)^\s*([A-Za-z_][A-Za-z0-9_]*)\s*:`)
+	collect := func(seg string) map[string]bool {
+		m := map[string]bool{}
+		for _, g := range keyRe.FindAllStringSubmatch(seg, -1) {
+			m[g[1]] = true
+		}
+		return m
+	}
+	zk, ek := collect(zh), collect(en)
+	// The locale markers themselves are the two block headers.
+	delete(zk, "zh")
+	delete(ek, "en")
+
+	var missingEN, missingZH []string
+	for k := range zk {
+		if !ek[k] {
+			missingEN = append(missingEN, k)
+		}
+	}
+	for k := range ek {
+		if !zk[k] {
+			missingZH = append(missingZH, k)
+		}
+	}
+	if len(missingEN) > 0 {
+		t.Errorf("translation keys present in zh but missing in en: %v", missingEN)
+	}
+	if len(missingZH) > 0 {
+		t.Errorf("translation keys present in en but missing in zh: %v", missingZH)
+	}
+	if len(zk) < 100 {
+		t.Errorf("only %d zh keys parsed — the block boundary regexp has drifted "+
+			"and this test is no longer checking anything", len(zk))
+	}
+}

internal/gateway/ui_image_page_test.go

@@ -0,0 +1,195 @@
+package gateway
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"llmsproxy/internal/provider"
+)
+
+// TestUI_ChatPageGroupsImageModels is a UI-contract guard for the bug that
+// started this: the chat page listed image models beside chat ones, and picking
+// one posted to /api/chat, which cannot serve it.
+//
+// The contract is asserted on the HTML source rather than by running a browser,
+// because the failure this must catch is a source-level one — an <option>
+// without `disabled`. A DOM assertion in a headless browser would pass right
+// up until someone removed the attribute, and then only if a CI machine had the
+// whole stack up.
+func TestUI_ChatPageGroupsImageModels(t *testing.T) {
+	ui := uiSource(t)
+
+	// The dropdown must read model_kinds, not the flat models list. Reading the
+	// flat list is what made chat and image models indistinguishable.
+	if !strings.Contains(ui, "model_kinds") {
+		t.Error("chat page never reads model_kinds; image models will be " +
+			"indistinguishable from chat models again")
+	}
+
+	// Image options must be rendered disabled. A plain String.replace over the
+	// joined HTML only disables the FIRST match — that bug shipped once here
+	// (one source disabled, five selectable) — so the assertion looks for the
+	// per-item map, not for a global replace.
+	if strings.Contains(ui, `imgOpts.join("").replace`) {
+		t.Error("chat page disables image options with String.replace over the " +
+			"joined HTML, which only rewrites the first option")
+	}
+	if !strings.Contains(ui, "disabled") {
+		t.Error("chat page never emits a disabled attribute for image models")
+	}
+}
+
+// TestUI_HasDedicatedImagePage pins the second half of the fix: image models
+// need somewhere they CAN be tested.
+func TestUI_HasDedicatedImagePage(t *testing.T) {
+	ui := uiSource(t)
+
+	for _, want := range []string{
+		`data-tab="image"`,         // sidebar entry
+		`id="tab-image"`,           // pane
+		`function renderImage`,     // renderer registered with refresh()
+		`tab === "image"`,          // dispatched
+		`/v1/images/generations`,   // the endpoint the page actually calls
+	} {
+		if !strings.Contains(ui, want) {
+			t.Errorf("image page is missing %q", want)
+		}
+	}
+
+	// The page must be reachable from TABS; goTab iterates TABS to toggle pane
+	// visibility, so a pane outside that list is present but permanently hidden —
+	// which reads as "the page is blank" because the content is in the DOM.
+	if !strings.Contains(ui, `"image"`) {
+		t.Error(`"image" does not appear in the TABS list`)
+	}
+}
+
+// TestUI_ImagePageCallsTheImageEndpoint is the discriminator: a page that
+// renders an image result but posts to the chat endpoint would produce exactly
+// the original failure, just prettier.
+func TestUI_ImagePageCallsTheImageEndpoint(t *testing.T) {
+	ui := uiSource(t)
+
+	i := strings.Index(ui, "async function sendImage(")
+	if i < 0 {
+		t.Fatal("sendImage() not found")
+	}
+	// Bound the search to the function body rather than the whole file, so a
+	// /v1/images/generations mention elsewhere cannot satisfy this.
+	body := ui[i:]
+	if end := strings.Index(body, "\n      function "); end > 0 {
+		body = body[:end]
+	}
+	if end := strings.Index(body, "\n      async function "); end > 0 {
+		if s := body[:end]; strings.Contains(s, "/v1/images/generations") {
+			body = s
+		} else {
+			body = body[:end]
+		}
+	}
+	if !strings.Contains(body, "/v1/images/generations") {
+		t.Error("sendImage() does not post to /v1/images/generations")
+	}
+}
+
+// TestSourceStatus_ExposesModelKindsToTheUI is the server half of the same
+// contract: the field the UI reads must actually be serialized.
+func TestSourceStatus_ExposesModelKindsToTheUI(t *testing.T) {
+	st := provider.SourceStatus{
+		Name:       "vela",
+		Models:     []string{"claude-sonnet-5.5", "gpt-image-2"},
+		ModelKinds: []provider.ModelKindInfo{
+			{ID: "claude-sonnet-5.5", Kind: "chat"},
+			{ID: "gpt-image-2", Kind: "image"},
+		},
+	}
+	b, err := json.Marshal(st)
+	if err != nil {
+		t.Fatal(err)
+	}
+	var got map[string]interface{}
+	if err := json.Unmarshal(b, &got); err != nil {
+		t.Fatal(err)
+	}
+	mk, ok := got["model_kinds"].([]interface{})
+	if !ok || len(mk) != 2 {
+		t.Fatalf("model_kinds = %v, want 2 entries; the UI cannot split the list without it", got["model_kinds"])
+	}
+	last := mk[1].(map[string]interface{})
+	if last["kind"] != "image" {
+		t.Errorf("model_kinds[1].kind = %v, want \"image\"", last["kind"])
+	}
+}
+
+// TestUI_BaseURLFieldExplainsWhereToCut keeps the second bug fixed: the URL
+// hint has to mention the concrete failure, not just say "enter the URL".
+func TestUI_BaseURLFieldExplainsWhereToCut(t *testing.T) {
+	ui := uiSource(t)
+	if !strings.Contains(ui, "urlHint") {
+		t.Fatal("no urlHint; the base_url field gives no guidance about /v1")
+	}
+	hint := ui[strings.Index(ui, "urlHint:"):]
+	if len(hint) > 700 {
+		hint = hint[:700]
+	}
+	// The trap is specific and reproducible: base_url ending in /v1 makes chat
+	// work and image generation 404. Saying only "enter the URL" reproduces the
+	// confusion the hint exists to remove.
+	for _, want := range []string{"/v1", "404"} {
+		if !strings.Contains(hint, want) {
+			t.Errorf("urlHint does not mention %q — the hint must name the concrete failure", want)
+		}
+	}
+}
+
+// TestImageEndpointComposesUnderRealConditions proves the hint describes a
+// reachable path rather than a plausible-sounding one.
+//
+// The case that matters: base_url already ends in /v1 (which is what an
+// operator pastes from an SDK example), so the RELATIVE image endpoint composes
+// to …/v1/images/generations and works, while the gateway's default ABSOLUTE
+// path would produce …/v1/v1/… and 404.
+//
+// The doubled form is served as a 404 here so the test would fail if the code
+// ever reverted to relying on the default.
+func TestImageEndpointComposesUnderRealConditions(t *testing.T) {
+	up := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		switch r.URL.Path {
+		case "/v1/images/generations":
+			_, _ = w.Write([]byte(`{"data":[{"url":"http://example.invalid/x.png"}]}`))
+		default:
+			w.WriteHeader(http.StatusNotFound)
+		}
+	}))
+	defer up.Close()
+
+	// base_url as an operator types it: host root + /v1.
+	base := up.URL + "/v1"
+
+	// What discovery writes when base carries a version segment.
+	relative := base + "/images/generations"
+	resp, err := http.Post(relative, "application/json", strings.NewReader(`{"model":"m"}`))
+	if err != nil {
+		t.Fatal(err)
+	}
+	resp.Body.Close()
+	if resp.StatusCode != http.StatusOK {
+		t.Errorf("POST %s = %d, want 200 — the form discovery writes must work", relative, resp.StatusCode)
+	}
+
+	// What the gateway's built-in default would produce for the same base_url.
+	doubled := base + "/v1/images/generations"
+	resp2, err := http.Post(doubled, "application/json", strings.NewReader(`{"model":"m"}`))
+	if err != nil {
+		t.Fatal(err)
+	}
+	resp2.Body.Close()
+	if resp2.StatusCode != http.StatusNotFound {
+		t.Errorf("POST %s = %d, want 404 — this test's premise is that the "+
+			"default image path doubles the version segment; if the upstream ever "+
+			"accepts it, composeImageEndpoint's rule needs rechecking", doubled, resp2.StatusCode)
+	}
+}

internal/gateway/ui_key_auto_test.go

@@ -0,0 +1,438 @@
+package gateway
+
+import (
+	"os"
+	"strings"
+	"testing"
+)
+
+// Per-key AUTO chain UI contract.
+//
+// The per-key editor deliberately REUSES the global lane canvas instead of
+// shipping a second, simpler editor. That is a design choice with teeth: a
+// reduced copy would lack drag ordering and per-slot quota/period/hours, and
+// the two would drift apart silently. These checks pin the sharing, because
+// the only symptom of un-sharing is "the per-key editor got worse" — no error,
+// just a smaller feature set that nobody notices until they use it.
+func TestUIPerKeyAutoWiring(t *testing.T) {
+	src := uiSource(t)
+
+	// The admin key list must offer the editor, otherwise the whole feature is
+	// unreachable from the UI. Assert the *call site* inside keyCanvasHtml,
+	// not just that the functions exist: removing the button leaves every
+	// definition intact, and that is exactly the regression to catch.
+	kcStart := strings.Index(src, "function keyCanvasHtml")
+	if kcStart < 0 {
+		t.Fatal("keyCanvasHtml is gone — cannot check the AUTO chain button")
+	}
+	kcEnd := strings.Index(src[kcStart:], "\n      }")
+	if kcEnd < 0 {
+		t.Fatal("could not delimit keyCanvasHtml")
+	}
+	if !strings.Contains(src[kcStart:kcStart+kcEnd], "openKeyAutoModal(") {
+		t.Fatal("per-key AUTO chain button is missing from the key card header")
+	}
+	for _, fn := range []string{
+		"function openKeyAutoModal",
+		"function keyAutoInherit",
+		"function keyAutoSave",
+		"function keyAutoClose",
+	} {
+		if !strings.Contains(src, fn) {
+			t.Fatalf("UI is missing %s — the editor cannot function", fn)
+		}
+	}
+}
+
+// The per-key editor must render the SAME canvas the global editor uses.
+// If this regresses, the per-key chain silently loses drag ordering and
+// per-slot quota/period/hours while still looking like it works.
+func TestUIPerKeyAutoReusesGlobalCanvas(t *testing.T) {
+	src := uiSource(t)
+
+	// One canvas implementation, two mount ids.
+	if !strings.Contains(src, "function renderSortEditor(") {
+		t.Fatal("no shared canvas renderer — the per-key editor must reuse renderSortEditor")
+	}
+	if !strings.Contains(src, `canvasId: "scr-canvas"`) {
+		t.Fatal("the global editor no longer mounts the shared canvas")
+	}
+	if !strings.Contains(src, `canvasId: "key-auto-canvas"`) {
+		t.Fatal("the per-key editor does not mount the shared canvas")
+	}
+	// Canvas lookup must consider both ids, or drag/drop silently no-ops on
+	// whichever surface is not #scr-canvas.
+	if !strings.Contains(src, "function sortCanvasEl(") ||
+		!strings.Contains(src, `getElementById("key-auto-canvas")`) {
+		t.Fatal("sortCanvasEl must resolve both canvas ids")
+	}
+	// The rules<->lanes conversion is shared too, so the per-key editor sends
+	// exactly the fields the server understands (quota/period/hours included).
+	for _, fn := range []string{"function buildLanes(", "function lanesToRules("} {
+		if !strings.Contains(src, fn) {
+			t.Fatalf("%s must be shared between the global and per-key editors", fn)
+		}
+	}
+}
+
+// Saving must route through persistAuto's scope switch, not a hand-rolled
+// request, or the two editors will serialize rules differently.
+func TestUIPerKeyAutoSavesThroughScope(t *testing.T) {
+	src := uiSource(t)
+	if !strings.Contains(src, "sortState.scope = key") {
+		t.Fatal("openKeyAutoModal must set sortState.scope so the save targets the key")
+	}
+	if !strings.Contains(src, "sortState.scope = null") {
+		t.Fatal("closing the modal must clear sortState.scope, or the GLOBAL editor would save to this key")
+	}
+	if !strings.Contains(src, "if (sortState.scope)") {
+		t.Fatal("persistAuto must branch on sortState.scope")
+	}
+	if !strings.Contains(src, `"/auto"`) {
+		t.Fatal("per-key editor must call the /api/keys/{key}/auto endpoint")
+	}
+	// An empty lane list is how "inherit the global chain" is expressed.
+	if !strings.Contains(src, "JSON.stringify({ auto: lanesToRules(sortState.lanes) })") {
+		t.Fatal("keyAutoSave must PUT lanesToRules(sortState.lanes) verbatim")
+	}
+}
+
+// Both languages need the strings, otherwise a non-Chinese admin sees raw
+// i18n keys in the modal.
+func TestUIPerKeyAutoHasBothLanguages(t *testing.T) {
+	src := uiSource(t)
+	for _, k := range []string{
+		"kAutoChainTitle", "kAutoChainHint", "kAutoChainInherit",
+		"kAutoChainEmpty", "kAutoChainGlobal", "kAutoChainBad",
+	} {
+		n := strings.Count(src, k+`: "`)
+		if n < 2 {
+			t.Fatalf("i18n key %s defined %d time(s), want both zh and en", k, n)
+		}
+	}
+}
+
+// The statistics dashboard must print the window it actually covers. Calendar
+// periods are not nested — "this week" starts on its Monday and therefore
+// reaches into the previous month early in a month, making its total LARGER
+// than "this month". That is correct arithmetic, but without the dates on
+// screen it reads as a bug, and it has been reported as one.
+func TestUIStatsShowsPeriodWindow(t *testing.T) {
+	src := uiSource(t)
+	if !strings.Contains(src, `id="period-range"`) {
+		t.Fatal("the stats dashboard has no element for the period window range")
+	}
+	if n := strings.Count(src, "periodRange"); n < 2 {
+		t.Fatalf("i18n key periodRange appears %d time(s), want both zh and en", n)
+	}
+}
+// Both canvases can exist in the DOM at once: the modal is mounted on top of
+// the settings tab, whose #scr-canvas stays alive behind it. Resolving the
+// canvas by document order (whichever comes first) painted the per-key chain
+// into the hidden background tab and left the dialog empty — and worse,
+// overwrote the global chain the user returns to. The lookup must follow
+// sortState.scope instead.
+func TestUISortCanvasResolvesByScope(t *testing.T) {
+	src := uiSource(t)
+	el := readFuncBody(t, src, "sortCanvasEl")
+	if !strings.Contains(el, "if (sortState.scope)") {
+		t.Fatal("sortCanvasEl must branch on sortState.scope, not document order")
+	}
+	if strings.Contains(el, `getElementById("scr-canvas")`) &&
+		strings.Index(el, `getElementById("scr-canvas")`) < strings.Index(el, "sortState.scope") {
+		t.Fatal("sortCanvasEl returns #scr-canvas before checking scope — the " +
+			"per-key editor paints into the global canvas")
+	}
+}
+
+// A successful save must close the dialog even when the slots do not resolve.
+// The write DID succeed; leaving the modal open makes the user click Save
+// again and again. This regressed once: the !slots branch returned early.
+func TestUIKeyAutoSaveClosesOnSuccess(t *testing.T) {
+	src := uiSource(t)
+	body := readFuncBody(t, src, "keyAutoSave")
+	// keyAutoClose() must be reached on every success path, not only in the
+	// happy branch.
+	iClose := strings.Index(body, "keyAutoClose()")
+	if iClose < 0 {
+		t.Fatal("keyAutoSave never calls keyAutoClose — the dialog would stay open")
+	}
+	iToast := strings.Index(body, "toast(")
+	if iToast >= 0 && iClose < iToast {
+		t.Fatal("keyAutoClose is called before the toast; the warning must be " +
+			"shown first or it is lost with the modal")
+	}
+	// An early `return` inside a branch bypasses keyAutoClose entirely, so
+	// merely finding the call is not enough: every `return` that sits between
+	// entering the try block and the close call is a path that leaves the
+	// dialog open after a successful write.
+	tryStart := strings.Index(body, "try {")
+	seg := body[tryStart:iClose]
+	for _, line := range strings.Split(seg, "\n") {
+		trimmed := strings.TrimSpace(line)
+		if trimmed == "return;" || strings.HasPrefix(trimmed, "return ") {
+			t.Fatal("a `return` before keyAutoClose leaves the dialog open after " +
+				"a successful save: " + trimmed)
+		}
+	}
+}
+
+// readFuncBody returns the source of a top-level function by name, using
+// brace matching so nested blocks do not truncate it.
+func readFuncBody(t *testing.T, src, name string) string {
+	t.Helper()
+	idx := strings.Index(src, "function "+name+"(")
+	if idx < 0 {
+		t.Fatalf("function %s not found", name)
+	}
+	start := strings.Index(src[idx:], "{")
+	if start < 0 {
+		t.Fatalf("function %s has no body", name)
+	}
+	i := idx + start
+	depth := 0
+	for j := i; j < len(src); j++ {
+		switch src[j] {
+		case '{':
+			depth++
+		case '}':
+			depth--
+			if depth == 0 {
+				return src[i : j+1]
+			}
+		}
+	}
+	t.Fatalf("function %s body is unterminated", name)
+	return ""
+}
+
+// The per-key dialog must not share the app-wide #modal-wrap id.
+//
+// Every dialog in this file uses that id, and more than one can be open at
+// once (the "add slot" picker opens ON TOP of this one). Handlers resolve
+// their own dialog via `.closest()` from their button, which works — but this
+// dialog is also looked up from the CANVAS side (sortCanvasEl, keyAutoClose,
+// keyAutoInherit), where there is no button to walk up from.
+// getElementById("#modal-wrap") then returns whichever dialog comes first in
+// the document, i.e. not the one being edited: Cancel closed the wrong box and
+// the canvas could resolve to the picker's.
+func TestUIKeyAutoModalHasOwnID(t *testing.T) {
+	src := uiSource(t)
+	body := readFuncBody(t, src, "openKeyAutoModal")
+	if !strings.Contains(body, `wrap.id = "key-auto-modal"`) {
+		t.Fatal("openKeyAutoModal must give its dialog its own id, not the shared " +
+			"#modal-wrap (another dialog may be open on top of it)")
+	}
+	close := readFuncBody(t, src, "keyAutoClose")
+	if strings.Contains(close, `"modal-wrap"`) {
+		t.Fatal("keyAutoClose resolves the shared #modal-wrap; it would close " +
+			"whichever dialog comes first in the document")
+	}
+	if !strings.Contains(close, `"key-auto-modal"`) {
+		t.Fatal("keyAutoClose must resolve the per-key dialog by its own id")
+	}
+}
+
+// Editing a chain inside the per-key dialog must NOT persist it. That dialog
+// has Save and Cancel, so auto-writing on add/drag/delete meant the change was
+// already committed before Save and Cancel could not undo it.
+//
+// The global page is the opposite: it has always applied edits immediately.
+func TestUIKeyAutoEditsDoNotAutoPersist(t *testing.T) {
+	src := uiSource(t)
+	body := readFuncBody(t, src, "afterChainEdit")
+	if !strings.Contains(body, "if (sortState.scope)") {
+		t.Fatal("afterChainEdit must branch on sortState.scope")
+	}
+	// The scoped branch must return before reaching persistAuto.
+	iScope := strings.Index(body, "if (sortState.scope)")
+	iReturn := strings.Index(body[iScope:], "return")
+	iPersist := strings.Index(body, "persistAuto()")
+	if iScope < 0 || iReturn < 0 || iPersist < 0 {
+		t.Fatal("afterChainEdit no longer has the expected shape")
+	}
+	if iScope+iReturn > iPersist {
+		t.Fatal("the scoped branch reaches persistAuto() — a per-key edit would " +
+			"be written before Save")
+	}
+	// And it must not call itself (a regex-driven rewrite did exactly that).
+	if strings.Contains(body, "afterChainEdit()") {
+		t.Fatal("afterChainEdit calls itself: infinite recursion")
+	}
+	// Every in-canvas edit goes through it, so none of them bypass the check.
+	for _, fn := range []string{"scrAddFromForm", "sortScopeSave", "scrDelSlot"} {
+		b := readFuncBody(t, src, fn)
+		if strings.Contains(b, "persistAuto()") {
+			t.Fatalf("%s calls persistAuto() directly, bypassing afterChainEdit", fn)
+		}
+	}
+}
+
+// The stats range line formats its end date from the LAST bucket label.
+// Bucket labels are hourly for the day view ("2026-10-03T09") and daily for
+// week/month ("2026-10-01"), so blindly appending "T00:00:00Z" to the hourly
+// form builds "2026-10-03T09T00:00:00Z" — Date rejects it, toISOString()
+// throws, and the assignment is lost. The symptom is an empty range line on
+// exactly the view people look at most, with no error anywhere.
+func TestUIStatsPeriodRangeHandlesHourlyBuckets(t *testing.T) {
+	src := uiSource(t)
+	el := readFuncBody(t, src, "paintStats")
+	// It must branch on the label shape instead of concatenating blindly.
+	if !strings.Contains(el, `T\d{2}$`) && !strings.Contains(el, `\d{2}-\d{2}T\d{2}`) {
+		t.Fatal("paintStats must recognise the hourly bucket label shape " +
+			"(YYYY-MM-DDTHH); concatenating a time onto it yields an invalid Date")
+	}
+	// And it must not contain the naive form that broke it.
+	if strings.Contains(el, `bucket + "T00:00:00Z"`) {
+		t.Fatal("paintStats still appends T00:00:00Z to the bucket label — " +
+			"this throws on the day view and leaves the range line empty")
+	}
+}
+
+// A key with no chain of its own must open on a COPY OF THE GLOBAL CHAIN, not
+// on an empty canvas. The realistic admin task is "this user gets the global chain
+// minus the sources they must not hit"; starting blank forces retyping every
+// tier against a 226-model picker. This is deliberately NOT the discovery
+// fallback (which dumps every model of every source) — it is the same rules
+// the global chain actually runs.
+func TestUIKeyAutoSeedsFromGlobalChain(t *testing.T) {
+	src := uiSource(t)
+	body := readFuncBody(t, src, "openKeyAutoModal")
+	if !strings.Contains(body, `api("/api/auto")`) {
+		t.Fatal("openKeyAutoModal must read the global chain to seed the editor")
+	}
+	if !strings.Contains(body, "seededFromGlobal") {
+		t.Fatal("the editor must remember whether its content came from the " +
+			"global chain, so the notice can say so")
+	}
+	// And it must not fall back to the discovery path that fills 226 models.
+	if strings.Contains(body, `buildLanes(idx, [], `) {
+		t.Fatal("seeding from an empty rule list triggers buildLanes discovery")
+	}
+}
+
+// The notice must distinguish "already this key's chain" from "a copy that
+// only becomes its chain on Save" — those two look identical on the canvas and
+// behave very differently if the admin hits Cancel.
+func TestUIKeyAutoNoticeDistinguishesSavedFromCopied(t *testing.T) {
+	src := uiSource(t)
+	body := readFuncBody(t, src, "keyAutoInheritNotice")
+	for _, k := range []string{"kAutoChainSeeded", "kAutoChainNewOwn", "kAutoChainWillInherit"} {
+		if !strings.Contains(body, k) {
+			t.Fatalf("the notice never uses %s — a copied chain and a saved one "+
+				"would look the same", k)
+		}
+	}
+	// Both languages, or a non-Chinese admin sees raw keys.
+	for _, k := range []string{"kAutoChainSeeded", "kAutoChainNewOwn", "kAutoChainWillInherit"} {
+		if n := strings.Count(src, k+`: "`); n < 2 {
+			t.Fatalf("i18n key %s defined %d time(s), want both zh and en", k, n)
+		}
+	}
+}
+
+// A key whose model scope excludes AUTO cannot use AUTO at all — the scope
+// filter runs before the chain — yet the editor happily lets an admin
+// configure a per-key chain. That combination is easy to create by accident
+// and produces a 403 with no visible cause, so the dialog must say so.
+//
+// We deliberately do NOT widen the scope automatically: silently granting a
+// model the operator did not ask for is worse than a loud warning.
+func TestUIKeyAutoWarnsOnScopeConflict(t *testing.T) {
+	src := uiSource(t)
+	if !strings.Contains(src, "function keyAutoScopeWarning(") {
+		t.Fatal("no scope-conflict check in the per-key editor")
+	}
+	body := readFuncBody(t, src, "keyAutoScopeWarning")
+	if !strings.Contains(body, "AUTO") {
+		t.Fatal("keyAutoScopeWarning must test whether the scope lists AUTO")
+	}
+	// It must not write anything back to the key: reporting only.
+	for _, forbidden := range []string{"PUT", "POST", "fetch(", "api("} {
+		if strings.Contains(body, forbidden) {
+			t.Fatalf("keyAutoScopeWarning calls %s — it must only report the "+
+				"conflict, never widen the scope itself", forbidden)
+		}
+	}
+	// Empty scope means unrestricted, so no warning is correct there.
+	if !strings.Contains(body, "if (!models.length)") {
+		t.Fatal("an unrestricted key (no models) must not be warned")
+	}
+	// Both languages.
+	if n := strings.Count(src, "kAutoChainScopeWarn"); n < 2 {
+		t.Fatalf("i18n key kAutoChainScopeWarn appears %d time(s), want zh+en", n)
+	}
+}
+
+// The scope conflict for AUTO must be reported by checkQuota, because that is
+// the check an AUTO request actually reaches first (it runs before the chain is
+// consulted). Putting the message on the later path made it dead code — the
+// request was rejected one check earlier with the generic "not allowed for this
+// key" wording, which sends an operator looking for a quota that isn't the
+// problem.
+func TestAutoScopeConflictMessageComesFromCheckQuota(t *testing.T) {
+	b, err := os.ReadFile("chat.go")
+	if err != nil {
+		t.Fatalf("read chat.go: %v", err)
+	}
+	src := string(b)
+	i := strings.Index(src, "func (g *Gateway) checkQuota(")
+	if i < 0 {
+		t.Fatal("checkQuota not found")
+	}
+	// Body = up to the next top-level func.
+	rest := src[i:]
+	if j := strings.Index(rest[1:], "\nfunc "); j > 0 {
+		rest = rest[:j+1]
+	}
+	if !strings.Contains(rest, "isAuto(model)") {
+		t.Fatal("checkQuota must special-case AUTO: a scope that omits AUTO is " +
+			"rejected there before anything else looks at the chain")
+	}
+	if !strings.Contains(rest, "not in this key's model scope") {
+		t.Fatal("checkQuota's AUTO branch must name the scope as the cause")
+	}
+}
+
+// renderSortEditor replaces its mount's innerHTML wholesale. The global
+// priority page used to hand it the .card that also holds the h2 toolbar, so
+// every open deleted the save/reset/add buttons: the lanes still rendered and
+// the drag worked, but there was no way to persist a change. The per-key modal
+// never showed this because its mount held only the canvas.
+//
+// The regression is invisible in the DOM when you only count lane elements, so
+// this asserts on the structure: the toolbar lives in a sibling of the mount,
+// not inside it.
+func TestGlobalSortToolbarSurvivesCanvasRender(t *testing.T) {
+	src := uiSource(t)
+
+	// The global editor must mount into a dedicated host, not the .card.
+	i := strings.Index(src, `$("#tab-sort").innerHTML = `)
+	if i < 0 {
+		t.Fatal("global sort markup not found")
+	}
+	chunk := src[i:min(len(src), i+1400)]
+	if !strings.Contains(chunk, `id="sort-mount"`) {
+		t.Fatal("global sort page must provide a dedicated #sort-mount for the canvas")
+	}
+	// The mount handed to renderSortEditor must be that host.
+	if !strings.Contains(src, `renderSortEditor($("#sort-mount")`) {
+		t.Fatal("renderSortEditor must be given #sort-mount; passing the .card " +
+			"wipes the h2 toolbar that holds the save button")
+	}
+	// The toolbar (save in particular) must be OUTSIDE #sort-mount.
+	close := strings.Index(chunk, "</div>`;")
+	if close < 0 {
+		t.Fatal("malformed #tab-sort template")
+	}
+	head := chunk[:close]
+	mountAt := strings.Index(head, `id="sort-mount"`)
+	saveAt := strings.Index(head, `onclick="saveSort()"`)
+	if saveAt < 0 {
+		t.Fatal("the global sort page has no save button at all")
+	}
+	if mountAt >= 0 && saveAt > mountAt {
+		t.Fatal("the save button sits inside the canvas mount and will be " +
+			"erased the moment the canvas renders")
+	}
+}

internal/gateway/ui_plugin_test.go

@@ -0,0 +1,413 @@
+package gateway
+
+import (
+	"strings"
+	"testing"
+)
+
+// The plugin UI injection is JavaScript inside the embedded index.html, and it
+// is the ONLY thing that turns a plugin's `ui` block into a visible page or
+// element. These tests pin the wiring on the JS side; the server side (what the
+// payload contains) is covered by TestUIInjectServesPluginUI and the lua
+// package's TestBillingPluginDeclaresUI.
+//
+// What makes this worth pinning: a missing hook here fails SILENTLY. The page
+// simply never appears, there is no error anywhere, and it looks like "the
+// plugin didn't declare a page" rather than "the UI forgot to inject it".
+
+// uiSource returns the embedded WebUI document.
+func uiSourceX(t *testing.T) string {
+	t.Helper()
+	return uiSource(t)
+}
+
+// TestUIFetchesPluginInjection: the boot sequence must ask the kernel what to
+// inject. Without this fetch the whole feature is inert.
+func TestUIFetchesPluginInjection(t *testing.T) {
+	src := uiSourceX(t)
+	if !strings.Contains(src, "/api/ui-inject") {
+		t.Error("the WebUI never calls /api/ui-inject; plugin pages and elements can never appear")
+	}
+}
+
+// TestUIInjectsBeforeFirstRender: injection must be awaited before the first
+// refresh, otherwise the sidebar is built without the plugin entry and the
+// first paint races the fetch. This is an ordering contract, so it is asserted
+// on the source order rather than trusted.
+func TestUIInjectsBeforeFirstRender(t *testing.T) {
+	src := uiSourceX(t)
+	iInject := strings.Index(src, "injectPluginUI()")
+	iRefresh := strings.LastIndex(src, `refresh("status")`)
+	if iInject < 0 {
+		t.Fatal("injectPluginUI() is never called")
+	}
+	if iRefresh < 0 {
+		t.Fatal("the boot sequence no longer calls refresh(\"status\")")
+	}
+	if iInject > iRefresh {
+		t.Error("injectPluginUI() is called after the first refresh; the sidebar " +
+			"and #main would be built before the plugin page exists")
+	}
+	// And it must be awaited, not fire-and-forget.
+	window := src[iInject:]
+	if !strings.Contains(window[:200], ".finally") && !strings.Contains(window[:200], "await") {
+		t.Error("injectPluginUI() is not awaited before refresh; a slow response " +
+			"would race the first paint")
+	}
+}
+
+// TestUIPluginScriptsRunAfterMarkup is the subtle one. Setting innerHTML with a
+// <script> tag does NOT execute it; appending via a template neither does. The
+// mount therefore has to be inserted first and its scripts re-created
+// afterwards, or a plugin's script runs before its own DOM exists — which is
+// exactly the "document.getElementById returns null" failure mode.
+func TestUIPluginScriptsRunAfterMarkup(t *testing.T) {
+	src := uiSourceX(t)
+	// A <template> is used to parse the mount without executing scripts...
+	if !strings.Contains(src, "createElement(\"template\")") {
+		t.Error("the mount is not parsed via <template>; scripts could execute before their DOM")
+	}
+	// ...and scripts are then re-created as fresh elements so they DO run.
+	if !strings.Contains(src, "document.createElement(\"script\")") {
+		t.Error("plugin <script> blocks are never re-created, so they never execute")
+	}
+	if !strings.Contains(src, "replaceWith(s)") {
+		t.Error("the original inert <script> is not replaced by an executable one")
+	}
+}
+
+// TestUIPluginAPISurface: the documented browser API must exist with the exact
+// names docs/plugins.md promises, since plugin authors code against it.
+func TestUIPluginAPISurface(t *testing.T) {
+	src := uiSourceX(t)
+	for _, member := range []string{"fetchState", "postState", "onTabShown"} {
+		if !strings.Contains(src, member+":") && !strings.Contains(src, member+"(") {
+			t.Errorf("window.pluginAPI.%s is missing; docs/plugins.md documents it", member)
+		}
+	}
+}
+
+// TestUIPluginPageBecomesRealTab: a plugin page must get a pane in #main AND a
+// sidebar button wired to goTab, otherwise the page is unreachable.
+func TestUIPluginPageBecomesRealTab(t *testing.T) {
+	src := uiSourceX(t)
+	// pane in #main
+	if !strings.Contains(src, `pane.id = "tab-" + id`) {
+		t.Error("no pane is created for a plugin page")
+	}
+	if !strings.Contains(src, "main.appendChild(pane)") {
+		t.Error("the plugin pane is not appended to #main")
+	}
+	// sidebar button wired to the tab router
+	if !strings.Contains(src, "btn.dataset.tab = id") {
+		t.Error("the sidebar button is not given a data-tab, so goTab() will not route to it")
+	}
+	if !strings.Contains(src, "btn.onclick = () => goTab(id)") {
+		t.Error("the sidebar button is not wired to goTab()")
+	}
+	// and the router must know about it
+	if !strings.Contains(src, "PLUGIN_PAGES.has(tab)") {
+		t.Error("refresh() does not route plugin pages, so opening one renders nothing")
+	}
+}
+
+// TestUIPluginElementsHonorAnchor: elements declare top / bottom / before:sel /
+// after:sel. Silently ignoring the anchor would put a "top" tile at the bottom
+// of the status page, which looks like a layout bug rather than a plugin bug.
+func TestUIPluginElementsHonorAnchor(t *testing.T) {
+	src := uiSourceX(t)
+	for _, anchor := range []string{`anchor === "top"`, `anchor.startsWith("before:")`, `"after:"`} {
+		if !strings.Contains(src, anchor) {
+			t.Errorf("the anchor form %s is not handled; elements would all land at the bottom", anchor)
+		}
+	}
+}
+
+// TestUIPluginInjectionFailureIsNonFatal: plugins are optional, so a failed
+// /api/ui-inject must still leave a working UI (the dashboard has to render).
+// Two places have to cooperate: the function swallows the fetch error, and the
+// caller catches anything that still escapes so refresh() always runs.
+func TestUIPluginInjectionFailureIsNonFatal(t *testing.T) {
+	src := uiSourceX(t)
+	// inside the function: the fetch is wrapped in try/catch
+	fnStart := strings.Index(src, "async function injectPluginUI()")
+	if fnStart < 0 {
+		t.Fatal("injectPluginUI() is not defined")
+	}
+	fn := src[fnStart:]
+	if !strings.Contains(fn, "plugins are optional; the UI must work without them") {
+		t.Error("injectPluginUI does not guard its own fetch failure")
+	}
+	// at the call site: the rejection cannot escape before the first render
+	// LastIndex, not Index: the DEFINITION of injectPluginUI also matches, and
+	// the definition has no .catch on it.
+	iCall := strings.LastIndex(src, "injectPluginUI()")
+	if iCall < 0 {
+		t.Fatal("injectPluginUI() is never called")
+	}
+	// Bound the window at len(src): the call site sits near EOF and a fixed
+	// slice overruns it (a panic in a test is worse than a skipped assertion).
+	end := iCall + 220
+	if end > len(src) {
+		end = len(src)
+	}
+	if !strings.Contains(src[iCall:end], ".catch") {
+		t.Error("a failed /api/ui-inject would reject before refresh(\"status\"), " +
+			"leaving the dashboard blank")
+	}
+}
+
+// ---- plugin management UI contract ---------------------------------------
+//
+// The management page is the operator's only way to take a broken plugin out
+// of the request path. Every one of these assertions guards a link that, if it
+// silently broke, would leave the gateway running with a plugin it cannot
+// disable — the worst kind of gap: everything looks fine and nothing is
+// reachable.
+
+func TestUIHasPluginTabAndPane(t *testing.T) {
+	src := uiSourceX(t)
+	if !strings.Contains(src, `data-tab="plugins"`) {
+		t.Error("no sidebar entry for the plugin page")
+	}
+	if !strings.Contains(src, `id="tab-plugins"`) {
+		t.Error("no #tab-plugins pane")
+	}
+	// The tab list is now a single constant; a new tab must be added there or
+	// goTab will not un-hide its pane.
+	if !strings.Contains(src, `const TABS = [`) {
+		t.Error("TABS is gone; the tab list went back to a duplicated literal")
+	}
+	for _, tn := range []string{"status", "chat", "keys", "sort", "sources", "adapters", "plugins"} {
+		if !strings.Contains(src, `"`+tn+`"`) {
+			t.Errorf("TABS is missing %q", tn)
+		}
+	}
+	// goTab must iterate TABS, not its own list.
+	if !strings.Contains(src, "TABS.forEach((tn) =>") {
+		t.Error("goTab does not iterate TABS")
+	}
+	if strings.Contains(src, `["status", "chat", "keys", "sort", "sources", "adapters"].forEach`) {
+		t.Error("a duplicated tab literal survived; it will drift from TABS")
+	}
+}
+
+func TestUIRendersPluginManagement(t *testing.T) {
+	src := uiSourceX(t)
+	body, ok := jsFunctionBody(src, "renderPlugins")
+	if !ok {
+		t.Fatal("renderPlugins() not found")
+	}
+	// It must read the DISK listing, not just the loaded set: a plugin that
+	// failed to compile is absent from the loaded set, and showing only the
+	// loaded set makes a syntax error look like "the plugin is not installed".
+	if !strings.Contains(body, "on_disk") {
+		t.Error("renderPlugins reads only the loaded set; a failed plugin would " +
+			"be invisible instead of shown with its error")
+	}
+	if !strings.Contains(body, "/api/plugins") {
+		t.Error("renderPlugins does not call /api/plugins")
+	}
+	// Hook errors must be surfaced: a plugin that throws in every stage leaves
+	// no other trace, so without this the symptom is "the feature just doesn't
+	// work".
+	if !strings.Contains(body, "hook_errors") {
+		t.Error("renderPlugins ignores hook_errors; a silently broken plugin is undebuggable")
+	}
+	// Enable / disable / remove / edit.
+	for _, fn := range []string{"togglePlugin", "delPlugin", "installPlugin", "editPlugin"} {
+		if _, ok := jsFunctionBody(src, fn); !ok {
+			t.Errorf("%s() is missing from the WebUI", fn)
+		}
+	}
+	// The toggle must go through the enable/disable endpoint, not delete.
+	tb, ok := jsFunctionBody(src, "togglePlugin")
+	if !ok {
+		t.Fatal("togglePlugin() missing")
+	}
+	if !strings.Contains(tb, `method: "PUT"`) {
+		t.Error("togglePlugin does not use PUT")
+	}
+	if !strings.Contains(tb, "enabled:") {
+		t.Error("togglePlugin does not send an \"enabled\" field")
+	}
+	// And the admin-only tab list must include plugins, or a non-admin would
+	// see a page whose every action 403s.
+	if !strings.Contains(src, `["sort", "sources", "adapters", "plugins"]`) {
+		t.Error("the admin-only tab list omits \"plugins\"; a user key would see a " +
+			"page full of actions that all fail with 403")
+	}
+}
+
+// TestUIBindDropzoneIsParameterised guards the refactor: the adapter and plugin
+// upload forms share one dropzone, so a hard-coded id would send a dropped
+// plugin file into the adapter name field.
+func TestUIBindDropzoneIsParameterised(t *testing.T) {
+	src := uiSourceX(t)
+	body, ok := jsFunctionBody(src, "bindDropzone")
+	if !ok {
+		t.Fatal("bindDropzone() not found")
+	}
+	if strings.Contains(body, `$("#dz")`) || strings.Contains(body, `$("#adp-name")`) {
+		t.Error("bindDropzone still hard-codes the adapter's element ids; the " +
+			"plugin form would write into the adapter form")
+	}
+	if !strings.Contains(body, "dzId") || !strings.Contains(body, "nameSel") {
+		t.Error("bindDropzone does not accept the ids to bind")
+	}
+	// Both forms must call it.
+	if !strings.Contains(src, `bindDropzone("pl-dz", "pl-file", "#pl-name", "#pl-code")`) {
+		t.Error("the plugin upload form does not use the parameterised dropzone")
+	}
+}
+
+// TestPluginElementsSurviveHostRebuild guards the defect that made plugin
+// elements look absent no matter how the injection was configured.
+//
+// renderStatus (and six other pages) assign pane.innerHTML wholesale. Anything a
+// plugin had mounted into that pane is destroyed by the assignment. The symptom
+// is silent and misleading: the plugin really did declare an element, the
+// payload really did arrive, and the element is still gone on the next repaint —
+// so the natural conclusion is "my plugin declared nothing", which sends you
+// looking in the wrong file.
+//
+// The fix is to re-mount after the rebuild. This test asserts the re-mount is
+// wired at the SINGLE place every renderer passes through, rather than leaving
+// it to be re-added per page.
+func TestPluginElementsSurviveHostRebuild(t *testing.T) {
+	html := uiSource(t)
+
+	if !strings.Contains(html, "function remountPluginElements") {
+		t.Fatal("remountPluginElements is not defined; nothing can re-attach a plugin element after a host rebuild")
+	}
+	if !strings.Contains(html, "const PLUGIN_MOUNT_HOOKS = []") {
+		t.Fatal("PLUGIN_MOUNT_HOOKS is not declared")
+	}
+	// The hook array must be declared BEFORE injectPluginUI pushes to it, and
+	// before refresh() calls into it. A use-before-declaration in a const
+	// block is a hard TDZ ReferenceError at first paint.
+	hookDecl := strings.Index(html, "const PLUGIN_MOUNT_HOOKS = []")
+	push := strings.Index(html, "PLUGIN_MOUNT_HOOKS.push")
+	use := strings.Index(html, "PLUGIN_MOUNT_HOOKS.forEach")
+	if hookDecl < 0 || push < 0 || use < 0 {
+		t.Fatal("the hook array is declared but never both filled and drained")
+	}
+	if hookDecl > push || hookDecl > use {
+		t.Error("PLUGIN_MOUNT_HOOKS is used before its declaration (const TDZ: first paint would throw)")
+	}
+
+	// refresh() is the chokepoint every renderer passes through.
+	rf := strings.Index(html, "function refresh(tab)")
+	if rf < 0 {
+		t.Fatal("refresh(tab) is gone")
+	}
+	body := html[rf:]
+	if i := strings.Index(body, "\n      }"); i > 0 {
+		body = body[:i]
+	}
+	if !strings.Contains(body, "remountPluginElements") {
+		t.Error("refresh() does not re-mount plugin elements: a page that rebuilds its DOM wipes them")
+	}
+
+	// A mount must be idempotent, or the widget's <script> runs again on every
+	// host repaint and its counters silently double.
+	if !strings.Contains(html, `data-idx="`) || !strings.Contains(html, "plugin-el[data-idx=") {
+		t.Error("element mounting is not guarded by a per-pane marker; re-mounting would re-run plugin scripts")
+	}
+}
+
+// TestPluginPagesAreReachableByGoTab guards a bug that only shows up in a
+// browser: goTab() iterated the hardcoded TABS list to toggle `hidden`, and a
+// plugin-contributed page is not in that list. The pane existed, the plugin had
+// filled it with real data, and clicking the sidebar entry changed nothing —
+// the `hidden` class was never removed.
+//
+// It reads exactly like "the page is blank" while the content sits in the DOM,
+// and no static check catches it: the injection is correct, the data is correct,
+// and the API returns 200.
+//
+// So this asserts the two structural facts that make it reachable: goTab
+// consults the plugin page set, and that set is declared before goTab runs (a
+// `const` further down the file would be a temporal-dead-zone ReferenceError on
+// the first click).
+func TestPluginPagesAreReachableByGoTab(t *testing.T) {
+	ui := uiSource(t)
+
+	goTab := strings.Index(ui, "function goTab(")
+	if goTab < 0 {
+		t.Fatal("goTab() is gone")
+	}
+	end := strings.Index(ui[goTab:], "\n      }")
+	if end < 0 {
+		t.Fatal("could not isolate goTab()")
+	}
+	body := ui[goTab : goTab+end]
+
+	if !strings.Contains(body, "PLUGIN_PAGES.forEach") {
+		t.Error("★ goTab() does not toggle plugin pages — a plugin page keeps its " +
+			"`hidden` class forever, so the sidebar entry does nothing")
+	}
+
+	// Declaration must precede the use, or the first click throws.
+	decl := strings.Index(ui, "var PLUGIN_PAGES = new Set()")
+	if decl < 0 {
+		// A const later in the file would also "work" only if nothing reads it
+		// first — make that explicit rather than silent.
+		if strings.Contains(ui, "const PLUGIN_PAGES = new Set()") {
+			t.Error("PLUGIN_PAGES is a `const` declared after goTab() reads it — " +
+				"temporal dead zone: the first click on a plugin page throws")
+		} else {
+			t.Error("PLUGIN_PAGES is not declared anywhere")
+		}
+		return
+	}
+	if decl > goTab {
+		t.Errorf("PLUGIN_PAGES is declared at %d but goTab() at %d reads it — "+
+			"declaration must come first", decl, goTab)
+	}
+}
+
+// TestPluginSidebarEntryCarriesATitle: a plugin tab must render icon AND label.
+//
+// Native tabs are `<svg>…</svg><span data-i="navStatus">状态</span>`; the plugin
+// branch set innerHTML to the icon alone, so every plugin page showed up as a
+// nameless icon in the sidebar. An operator cannot tell what an icon means
+// until they click it.
+func TestPluginSidebarEntryCarriesATitle(t *testing.T) {
+	html := uiSource(t)
+	anchor := "btn.innerHTML =\n              pluginIconHTML(pg.icon) +"
+	i := strings.Index(html, anchor)
+	if i < 0 {
+		t.Fatal("plugin sidebar button construction not found")
+	}
+	// Look at the statement the anchor opens, not the whole file.
+	stmt := html[i : i+400]
+	if strings.Contains(stmt, "esc(pg.title") {
+		return
+	}
+	t.Errorf("plugin sidebar button renders the icon only; a label is required. "+
+		"got: %.200s", stmt)
+}
+
+// TestPluginPageHasAccessEditor: the plugin page must let an operator configure
+// endpoint access. The APIs existed with zero UI entry point, which meant the
+// only way to change who could reach a plugin endpoint was to hand-edit
+// config.yaml — the thing the UI is meant to replace.
+func TestPluginPageHasAccessEditor(t *testing.T) {
+	src := uiSource(t)
+	for _, needle := range []string{
+		`renderPluginAccess`,    // called after rendering the plugin list
+		`savePluginAccess`,      // wired to a button
+		`/access`,               // hits the per-endpoint access API
+		`id="pl-access"`,        // container exists
+		`http_access`,           // reads the table from /api/plugins
+		`data-ep=`,              // one input per endpoint
+		`data-plvis=`,           // one visibility selector per plugin
+		`/http/*/access`,        // whole-plugin visibility shares the access API
+		`plVisCell`,             // renders the declared-vs-override distinction
+	} {
+		if !strings.Contains(src, needle) {
+			t.Errorf("plugin page lacks %q", needle)
+		}
+	}
+}

internal/gateway/ui_quota_contract_test.go

@@ -0,0 +1,267 @@
+package gateway
+
+import (
+	"regexp"
+	"strings"
+	"testing"
+)
+
+// The WebUI dialogs all share the id "modal-wrap", and more than one can be
+// open at the same time (the seed-key notice sits on top of the keys page).
+// A save handler that closes "the" modal via $("#modal-wrap") therefore removes
+// whichever one comes FIRST in the document — which is the wrong dialog: the
+// form the user just submitted stays on screen while an unrelated dialog
+// vanishes.
+//
+// This is the same class of bug the api() contract test pins: reviewing inline
+// JS by eye does not catch it, and the symptom ("the dialog did not close")
+// points away from the cause.
+
+// modalCloseRe finds every `$(...)`-style lookup of the shared modal id.
+var modalCloseRe = regexp.MustCompile(`\$\("#modal-wrap"\)`)
+
+// closestModalRe finds the safe form: resolve the dialog from the clicked
+// button instead of from the document.
+var closestModalRe = regexp.MustCompile(`\.closest\("#modal-wrap"\)`)
+
+func TestUIDialogClosesItselfNotTheFirstModal(t *testing.T) {
+	// Strip comments first: prose that *names* the unsafe pattern (as the fix's
+	// own comment does) would otherwise be flagged as a violation.
+	code := stripJSComments(uiSource(t))
+
+	for _, m := range modalCloseRe.FindAllStringIndex(code, -1) {
+		after := code[m[1]:]
+		stmtEnd := strings.Index(after, ";")
+		if stmtEnd < 0 || stmtEnd > 200 {
+			continue
+		}
+		stmt := after[:stmtEnd]
+		if strings.Contains(stmt, ".remove()") {
+			line := 1 + strings.Count(code[:m[0]], "\n")
+			t.Errorf("line %d closes the first #modal-wrap in the document, not its own dialog:\n\t%s",
+				line, strings.TrimSpace(stmt))
+		}
+	}
+}
+
+// stripJSComments removes // line comments and /* block */ comments from JS
+// embedded in the UI document. It is deliberately simple: the document is our
+// own source, and a false negative here only means the check is silent.
+func stripJSComments(src string) string {
+	var out strings.Builder
+	lines := strings.Split(src, "\n")
+	inBlock := false
+	for _, ln := range lines {
+		trimmed := strings.TrimSpace(ln)
+		if inBlock {
+			if strings.Contains(ln, "*/") {
+				inBlock = false
+			}
+			continue
+		}
+		if strings.HasPrefix(trimmed, "/*") {
+			if !strings.Contains(ln, "*/") {
+				inBlock = true
+			}
+			continue
+		}
+		if i := strings.Index(ln, "//"); i >= 0 {
+			before := ln[:i]
+			if strings.Count(before, `"`)%2 == 0 && strings.Count(before, "'")%2 == 0 {
+				ln = before
+			}
+		}
+		out.WriteString(ln)
+		out.WriteString("\n")
+	}
+	return out.String()
+}
+
+// Every handler that closes a dialog must do it one of the two safe ways.
+func TestUIDialogClosuresGoThroughSafePaths(t *testing.T) {
+	src := stripJSComments(uiSource(t))
+	for _, fn := range []string{
+		"downloadStatsCsv", "downloadKeysCsv", "saveSource", "saveTemplate",
+		"scrAddFromForm", "sortScopeSave", "scopeSave", "createKey",
+	} {
+		body, ok := jsFunctionBody(src, fn)
+		if !ok {
+			t.Errorf("%s not found", fn)
+			continue
+		}
+		if !strings.Contains(body, `closest("#modal-wrap")`) && !strings.Contains(body, "closeTopModal()") {
+			t.Errorf("%s closes a dialog with neither .closest nor closeTopModal", fn)
+		}
+	}
+	if !strings.Contains(src, "function closeTopModal(") {
+		t.Error("closeTopModal helper is missing")
+	}
+}
+
+// The helper must pick the LAST dialog (the topmost one the user sees), not the
+// first — that inversion is the whole bug.
+func TestUICloseTopModalTakesTheLast(t *testing.T) {
+	body, ok := jsFunctionBody(uiSource(t), "closeTopModal")
+	if !ok {
+		t.Fatal("closeTopModal not found")
+	}
+	if !strings.Contains(body, "all.length - 1") {
+		t.Errorf("closeTopModal does not take the last dialog:\n\t%s", oneLine(body))
+	}
+}
+
+// Handlers that resolve their dialog from a button must actually receive one:
+// a signature without the parameter means the .closest() silently yields null
+// and the save leaves its form stranded.
+func TestUIDialogHandlersReceiveTheirButton(t *testing.T) {
+	src := stripJSComments(uiSource(t))
+	for _, fn := range []string{
+		"downloadStatsCsv", "saveSource", "scrAddFromForm", "sortScopeSave",
+		"scopeSave", "createKey",
+	} {
+		body, ok := jsFunctionBody(src, fn)
+		if !ok {
+			t.Errorf("%s not found", fn)
+			continue
+		}
+		if !strings.Contains(body, "closest(\"#modal-wrap\")") {
+			continue // uses closeTopModal only
+		}
+		sig := body[:strings.Index(body, ")")+1]
+		if !strings.Contains(sig, "btn") {
+			t.Errorf("%s uses .closest(\"#modal-wrap\") but its signature %s has no button parameter —\n"+
+				"the lookup would always be null and the form would never close", fn, oneLine(sig))
+		}
+	}
+}
+
+// Quotas belong to the scope entries, so putScope only has to ship the scope
+// list — the caps travel inside it. What must NOT come back is a key-wide
+// total: it would be a second budget able to drift out of sync with the models
+// it is supposed to cover.
+func TestUIPutScopeShipsOnlyScopeQuotas(t *testing.T) {
+	body, ok := jsFunctionBody(uiSource(t), "putScope")
+	if !ok {
+		t.Fatal("putScope not found")
+	}
+	code := stripJSComments(body)
+	if !strings.Contains(code, "models:") || !strings.Contains(code, "scopes") {
+		t.Error("putScope must ship the scope list the caps live in")
+	}
+	for _, gone := range []string{"kquota", "kreqquota", "kperiod", "khours"} {
+		if strings.Contains(code, gone) {
+			t.Errorf("putScope still references the removed key-wide quota field %q", gone)
+		}
+	}
+}
+
+// A model brick carries its own budgets, and the scope editor reads and writes
+// both of them: dropping req_quota on the round trip would silently lift a
+// request cap every time someone edited a token cap.
+func TestUIScopeEditorRoundTripsBothQuotas(t *testing.T) {
+	src := stripJSComments(uiSource(t))
+	for _, fn := range []string{"scopeHtml", "readScopes", "scopeEdit", "scopeSave", "scopeQuotaBadge"} {
+		if _, ok := jsFunctionBody(src, fn); !ok {
+			t.Errorf("%s not found in the UI source", fn)
+		}
+	}
+	for _, fn := range []string{"scopeHtml", "readScopes", "scopeSave", "scopeQuotaBadge"} {
+		body, ok := jsFunctionBody(src, fn)
+		if !ok {
+			continue
+		}
+		if !strings.Contains(body, "req_quota") && !strings.Contains(body, "reqquota") {
+			t.Errorf("%s does not carry req_quota — a request cap would be lost on edit", fn)
+		}
+	}
+	if form, ok := jsFunctionBody(src, "scopeEdit"); ok && !strings.Contains(form, "sc-reqs") {
+		t.Error("the scope editor has no request-quota input")
+	}
+}
+
+// The whole key-wide quota surface must be gone from the UI: a badge or a
+// button reading a field the server no longer has would render "undefined" or
+// silently do nothing.
+func TestUIHasNoKeyWideQuotaSurface(t *testing.T) {
+	src := uiSource(t)
+	for _, gone := range []string{
+		"keyCapBadges", "keyQuotaEdit", "keyQuotaSave",
+		"kq-tokens", "kq-reqs", "kq-period", "kq-hours",
+		"kc-tokens", "kc-reqs", "kc-period", "kc-hours",
+		"data-kquota", "data-kreqquota", "data-kperiod", "data-khours",
+	} {
+		if strings.Contains(src, gone) {
+			t.Errorf("UI still references the removed key-wide quota surface %q", gone)
+		}
+	}
+}
+
+// Existence of the strings is not enough: the badge has to READ the scope's
+// fields, and the editor has to read back what the brick writes. A field can
+// be present in the source and still never reach the screen — e.g. left in a
+// dead branch, or read from a data attribute the writer never sets.
+func TestUIKeyQuotaDataflowIsLive(t *testing.T) {
+	src := stripJSComments(uiSource(t))
+
+	badge, ok := jsFunctionBody(src, "scopeQuotaBadge")
+	if !ok {
+		t.Fatal("scopeQuotaBadge not found")
+	}
+	for _, field := range []string{"m.token_quota", "m.req_quota", "m.period"} {
+		if !strings.Contains(badge, field) {
+			t.Errorf("scopeQuotaBadge does not read %q — the cap would never show on the model brick", field)
+		}
+	}
+
+	brick, ok := jsFunctionBody(src, "scopeHtml")
+	if !ok {
+		t.Fatal("scopeHtml not found")
+	}
+	edit, ok := jsFunctionBody(src, "scopeEdit")
+	if !ok {
+		t.Fatal("scopeEdit not found")
+	}
+	brickCode, editCode := stripJSComments(brick), stripJSComments(edit)
+	reader := stripJSComments(mustBody(t, src, "readScopes"))
+	for _, ds := range []string{"quota", "reqquota", "period", "hours"} {
+		if !strings.Contains(brickCode, "data-"+ds+"=") {
+			t.Errorf("scopeHtml does not write data-%s, so the editor has nothing to prefill", ds)
+		}
+		// the editor pre-fills through readScopes(), which is what walks the
+		// bricks' data attributes — check the reader, not the form
+		if !strings.Contains(reader, "dataset."+ds) {
+			t.Errorf("readScopes does not read dataset.%s — editing a brick would save zeros over it", ds)
+		}
+	}
+	// and the form must actually consume what readScopes produced
+	for _, field := range []string{"sc.token_quota", "sc.req_quota", "sc.period", "sc.hours"} {
+		if !strings.Contains(editCode, field) {
+			t.Errorf("scopeEdit does not prefill from %q", field)
+		}
+	}
+}
+
+func mustBody(t *testing.T, src, fn string) string {
+	t.Helper()
+	b, ok := jsFunctionBody(src, fn)
+	if !ok {
+		t.Fatalf("%s not found", fn)
+	}
+	return b
+}
+
+// The quota period vocabulary must match the server's, or the UI can offer a
+// value the API rejects.
+func TestUIQuotaPeriodsMatchServer(t *testing.T) {
+	src := uiSource(t)
+	for _, p := range []string{`value=""`, `value="hour"`, `value="week"`, `value="month"`, `value="nhour"`} {
+		if !strings.Contains(src, p) {
+			t.Errorf("UI period select is missing %s", p)
+		}
+	}
+	for _, p := range []string{`"hour"`, `"week"`, `"month"`, `"nhour"`} {
+		if !strings.Contains(src, `if (p === `+p+`)`) && !strings.Contains(src, `=== `+p+`)`) {
+			t.Errorf("periodText() does not describe %s, so a badge would omit the window", p)
+		}
+	}
+}

internal/lua/adapter_qwen38_test.go

@@ -0,0 +1,323 @@
+package lua
+
+import (
+	"encoding/json"
+	"strings"
+	"testing"
+)
+
+// The fixtures below are REAL responses captured from
+// https://discovery-api.intern-ai.org.cn/v1 with qwen3.8-27b — not hand-written
+// approximations. Two of them exist only because the endpoint behaves in a way a
+// stock OpenAI adapter gets wrong, and neither is guessable from the schema.
+
+// reasoningDominatesTheBudget is the fact that made a dedicated adapter
+// necessary: a one-sentence visual answer came back as completion=303 with
+// reasoning=295. 97% of the budget went to thinking and only 8 tokens reached
+// content. Handing the upstream's content straight through yields what looks
+// like an empty answer from a perfectly healthy request.
+const reasoningDominatesTheBudget = `{
+  "choices": [{"index":0,"message":{"role":"assistant","content":"\n\n这张图是白色的。",
+    "reasoning_content":"分析请求…分析图片…颜色是纯白"},"logprobs":null,
+    "finish_reason":"stop"}],
+  "created":1791179068,"model":"qwen3.8-27b-fp8","object":"chat.completion",
+  "usage":{"prompt_tokens":125,"total_tokens":428,"completion_tokens":303,
+    "prompt_tokens_details":null,
+    "completion_tokens_details":{"reasoning_tokens":295}}
+}`
+
+// truncatedReasoningOnly is what a small max_tokens produces: HTTP 200, content
+// null, finish_reason "length". It is NOT an upstream failure.
+const truncatedReasoningOnly = `{
+  "choices": [{"index":0,"message":{"role":"assistant","content":null,
+    "reasoning_content":"思考过程：1. **分析请求**…图片的颜色是纯","refusal":null},
+    "finish_reason":"length"}],
+  "model":"qwen3.8-27b-fp8","object":"chat.completion",
+  "usage":{"prompt_tokens":127,"total_tokens":190,"completion_tokens":63}
+}`
+
+const qwen38ToolCallResponse = `{
+  "choices": [{"index":0,"message":{"role":"assistant","content":"",
+    "reasoning_content":"用户询问北京天气。我需要调用 get_weather 函数，参数为城市 \"北京\"。",
+    "tool_calls":[{"id":"chatcmpl-tool-afc0299b5810e044","type":"function",
+      "function":{"name":"get_weather","arguments":"{\"city\": \"北京\"}"}}]},
+    "finish_reason":"tool_calls"}],
+  "model":"qwen3.8-27b-fp8",
+  "usage":{"prompt_tokens":305,"total_tokens":353,"completion_tokens":48,
+    "completion_tokens_details":{"reasoning_tokens":20}}
+}`
+
+func qwen38VM(t *testing.T) *VM {
+	t.Helper()
+	vm := NewVM(freshAdapterDir(t))
+	if err := vm.Start(); err != nil {
+		t.Fatalf("adapter VM: %v", err)
+	}
+	t.Cleanup(vm.Stop)
+	return vm
+}
+
+func qwen38Response(t *testing.T, vm *VM, raw string) map[string]interface{} {
+	t.Helper()
+	out, err := vm.Transform("qwen38", "transform_response", raw)
+	if err != nil {
+		t.Fatalf("transform_response: %v", err)
+	}
+	var m map[string]interface{}
+	if err := json.Unmarshal([]byte(out), &m); err != nil {
+		t.Fatalf("adapter returned non-JSON: %v\n%s", err, out)
+	}
+	return m
+}
+
+func qwen38Stream(t *testing.T, vm *VM, raw string) map[string]interface{} {
+	t.Helper()
+	out, err := vm.Transform("qwen38", "transform_stream_chunk", raw)
+	if err != nil {
+		t.Fatalf("transform_stream_chunk: %v", err)
+	}
+	var m map[string]interface{}
+	if err := json.Unmarshal([]byte(out), &m); err != nil {
+		t.Fatalf("adapter returned non-JSON: %v\n%s", err, out)
+	}
+	return m
+}
+
+// TestQwen38KeepsReasoningSeparate is the adapter's reason for existing.
+// Folding reasoning into content would render the model's private deliberation as
+// its answer; dropping it makes a reasoning model look like it returned nothing.
+func TestQwen38KeepsReasoningSeparate(t *testing.T) {
+	vm := qwen38VM(t)
+	u := qwen38Response(t, vm, reasoningDominatesTheBudget)
+
+	content, _ := u["content"].(string)
+	reasoning, _ := u["reasoning_content"].(string)
+
+	if reasoning == "" {
+		t.Fatal("reasoning_content was dropped — a reasoning model then looks like it returned nothing")
+	}
+	if strings.Contains(content, "分析请求") {
+		t.Errorf("reasoning was folded into content: %q", content)
+	}
+	if !strings.Contains(content, "白色的") {
+		t.Errorf("content = %q, want the answer text", content)
+	}
+	if reasoning == content {
+		t.Error("content and reasoning_content are identical; they are disjoint upstream")
+	}
+}
+
+// TestQwen38ReportsTruncationRatherThanFailing: content=null with
+// finish_reason="length" is a truncated reasoning response, not a broken
+// upstream. Losing the finish_reason makes it indistinguishable from a failure,
+// and losing the partial reasoning throws away the only evidence of what happened.
+func TestQwen38ReportsTruncationRatherThanFailing(t *testing.T) {
+	vm := qwen38VM(t)
+	u := qwen38Response(t, vm, truncatedReasoningOnly)
+
+	if u["finish_reason"] != "length" {
+		t.Errorf("finish_reason = %v, want length", u["finish_reason"])
+	}
+	if reasoning, _ := u["reasoning_content"].(string); reasoning == "" {
+		t.Error("the partial reasoning was discarded")
+	}
+	if content, _ := u["content"].(string); content != "" {
+		t.Errorf("content = %q, want empty string for a null upstream content", content)
+	}
+}
+
+// TestQwen38CarriesReasoningTokensWithoutInflatingTotals: reasoning_tokens is a
+// SUBSET of completion_tokens. Adding them double-counts every reasoning request,
+// and billing prices the pair separately.
+func TestQwen38CarriesReasoningTokensWithoutInflatingTotals(t *testing.T) {
+	vm := qwen38VM(t)
+	u := qwen38Response(t, vm, reasoningDominatesTheBudget)
+
+	tu, ok := u["token_usage"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("token_usage = %#v", u["token_usage"])
+	}
+	completion, _ := tu["completion"].(float64)
+	total, _ := tu["total"].(float64)
+
+	// The NESTING is the point, not the presence of the field.
+	//
+	// internal/gateway/chat.go reads reasoning in exactly one place:
+	//   resp.TokenUsage.CompletionTokensDetails.ReasoningTokens
+	// An adapter that emits reasoning_tokens at the TOP of token_usage type-checks
+	// fine and is then dropped at the one spot that could have kept it — measured
+	// on the live gateway: the model reported completion=88 and the gateway's
+	// response carried reasoning=null.
+	//
+	// So this asserts the path, not merely that a number appears somewhere.
+	det, ok := tu["completion_tokens_details"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("token_usage has no completion_tokens_details: %#v — chat.go reads "+
+			"reasoning from there and nowhere else, so the number is discarded", tu)
+	}
+	reasoning, has := det["reasoning_tokens"].(float64)
+	if !has {
+		t.Fatal("completion_tokens_details.reasoning_tokens missing — billing cannot price reasoning separately")
+	}
+	// Relational, not a presence check. The first version of this test only
+	// asserted the field existed, and a mutation setting it to 0 passed: 0 is
+	// present. The upstream value is known (295), so the exact number is
+	// asserted — a zero silently misprices every reasoning request.
+	if reasoning != 295 {
+		t.Errorf("reasoning_tokens = %v, want 295 (the upstream value)", reasoning)
+	}
+	if reasoning >= completion {
+		t.Errorf("reasoning_tokens (%v) must be a SUBSET of completion_tokens (%v)", reasoning, completion)
+	}
+	if completion != 303 || total != 428 {
+		t.Errorf("usage was altered: completion=%v total=%v, want 303/428", completion, total)
+	}
+}
+
+// TestQwen38DecodesToolCallArguments: the upstream sends arguments as a JSON
+// STRING, like every other OpenAI-compatible endpoint. Passing it through
+// undecoded leaves the caller to parse it.
+func TestQwen38DecodesToolCallArguments(t *testing.T) {
+	vm := qwen38VM(t)
+	u := qwen38Response(t, vm, qwen38ToolCallResponse)
+
+	tcs, ok := u["tool_calls"].([]interface{})
+	if !ok || len(tcs) != 1 {
+		t.Fatalf("tool_calls = %#v", u["tool_calls"])
+	}
+	tc, _ := tcs[0].(map[string]interface{})
+	if tc["name"] != "get_weather" {
+		t.Errorf("name = %v", tc["name"])
+	}
+	args, ok := tc["arguments"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("arguments = %#v — the upstream sends a JSON string and it must be decoded", tc["arguments"])
+	}
+	if args["city"] != "北京" {
+		t.Errorf("arguments.city = %v", args["city"])
+	}
+}
+
+// TestQwen38StreamKeepsReasoningSeparate: measured 20 reasoning chunks against 1
+// content chunk on a one-sentence answer. Folding them would make the client
+// render the deliberation as the reply.
+func TestQwen38StreamKeepsReasoningSeparate(t *testing.T) {
+	vm := qwen38VM(t)
+	const reasoningChunk = `{"choices":[{"index":0,"delta":{"reasoning_content":"思考过程：1. **分析请求**"},
+	  "finish_reason":null}],"model":"qwen3.8-27b-fp8"}`
+	const contentChunk = `{"choices":[{"index":0,"delta":{"content":"递归就是函数调用自身"},
+	  "finish_reason":null}],"model":"qwen3.8-27b-fp8"}`
+	const usageChunk = `{"choices":[],"usage":{"prompt_tokens":64,"total_tokens":132,
+	  "completion_tokens":68,"completion_tokens_details":{"reasoning_tokens":62}}}`
+
+	a := qwen38Stream(t, vm, reasoningChunk)
+	b := qwen38Stream(t, vm, contentChunk)
+	c := qwen38Stream(t, vm, usageChunk)
+
+	if content, mixed := a["content"].(string); mixed && content != "" {
+		t.Errorf("a reasoning chunk leaked into content: %q", content)
+	}
+	if reasoning, _ := a["reasoning_content"].(string); !strings.Contains(reasoning, "思考过程") {
+		t.Errorf("reasoning chunk lost its text: %#v", a)
+	}
+	if content, _ := b["content"].(string); !strings.Contains(content, "递归") {
+		t.Errorf("content chunk lost its text: %#v", b)
+	}
+	if c["usage"] == nil {
+		t.Errorf("the terminal usage chunk was dropped: %#v", c)
+		return
+	}
+	u, _ := c["usage"].(map[string]interface{})
+	det, _ := u["completion_tokens_details"].(map[string]interface{})
+	if det == nil || det["reasoning_tokens"] == nil {
+		t.Errorf("usage chunk did not nest reasoning under completion_tokens_details: %#v — "+
+			"chat.go reads it there, so a top-level field is discarded", u)
+	}
+}
+
+// TestQwen38ReportsMediaFailureCode: a remote image URL times out upstream and
+// comes back as media_fetch_failed. Surfacing only the message makes an image
+// fetch problem look like a model problem.
+func TestQwen38ReportsMediaFailureCode(t *testing.T) {
+	vm := qwen38VM(t)
+	body := `{"error":{"message":"image fetch timed out","type":"media_fetch_failed",
+	  "param":null,"code":"media_fetch_failed","trace_id":"f3ca4aa"}}`
+	reason, ok, err := vm.TransformError("qwen38", 400, body)
+	if err != nil {
+		t.Fatalf("TransformError: %v", err)
+	}
+	if !ok {
+		t.Fatal("the media failure was swallowed")
+	}
+	if !strings.Contains(reason, "image fetch timed out") || !strings.Contains(reason, "media_fetch_failed") {
+		t.Errorf("reason = %q; it must carry both the text and the code", reason)
+	}
+}
+
+// TestQwen38InvalidKeyError: the endpoint answers a bad key with 401 and
+// code "invalid_api_key". That one must NOT be annotated with its own code — it
+// would render as "invalid API key (invalid_api_key)".
+func TestQwen38InvalidKeyError(t *testing.T) {
+	vm := qwen38VM(t)
+	body := `{"error":{"message":"invalid API key","type":"invalid_request_error",
+	  "param":null,"code":"invalid_api_key"}}`
+	reason, ok, err := vm.TransformError("qwen38", 401, body)
+	if err != nil || !ok {
+		t.Fatalf("TransformError: ok=%v err=%v", ok, err)
+	}
+	if reason != "invalid API key" {
+		t.Errorf("reason = %q, want the bare message", reason)
+	}
+}
+
+// TestQwen38SuppliesReasoningOnToolCallTurns: the upstream requires an assistant
+// message carrying tool_calls to echo reasoning_content. OpenAI-shaped clients
+// never send it, so without this the SECOND turn of any tool-using conversation
+// is rejected.
+func TestQwen38SuppliesReasoningOnToolCallTurns(t *testing.T) {
+	vm := qwen38VM(t)
+	const req = `{"model":"qwen3.8-27b","messages":[
+	  {"role":"user","content":"北京天气"},
+	  {"role":"assistant","content":"","tool_calls":[{"id":"c1","type":"function",
+	    "function":{"name":"get_weather","arguments":"{\"city\":\"北京\"}"}}]},
+	  {"role":"tool","tool_call_id":"c1","content":"晴 25C"}]}`
+	out, err := vm.Transform("qwen38", "transform_request", req)
+	if err != nil {
+		t.Fatalf("transform_request: %v", err)
+	}
+	var decoded map[string]interface{}
+	if err := json.Unmarshal([]byte(out), &decoded); err != nil {
+		t.Fatalf("request transform did not return JSON: %v\n%s", err, out)
+	}
+	msgs, _ := decoded["messages"].([]interface{})
+	if len(msgs) < 2 {
+		t.Fatalf("messages = %#v", decoded["messages"])
+	}
+	assistant, _ := msgs[1].(map[string]interface{})
+	if _, ok := assistant["reasoning_content"]; !ok {
+		t.Errorf("assistant turn with tool_calls has no reasoning_content; the upstream "+
+			"rejects the next turn. Got %#v", assistant)
+	}
+}
+
+// TestQwen38PassesImagesThrough: this adapter's job is reasoning handling, not
+// media translation. A multimodal content array must survive untouched — an
+// adapter that re-encodes it is the kind of change that silently breaks images
+// while every text test still passes.
+func TestQwen38PassesImagesThrough(t *testing.T) {
+	vm := qwen38VM(t)
+	const req = `{"model":"qwen3.8-27b","messages":[{"role":"user","content":[
+	  {"type":"text","text":"这张图是什么颜色？"},
+	  {"type":"image_url","image_url":{"url":"data:image/png;base64,iVBORw0KGgo="}}]}],
+	  "max_tokens":600}`
+	out, err := vm.Transform("qwen38", "transform_request", req)
+	if err != nil {
+		t.Fatalf("transform_request: %v", err)
+	}
+	if !strings.Contains(out, "image_url") || !strings.Contains(out, "data:image/png;base64,") {
+		t.Errorf("the image part was lost or rewritten:\n%s", out)
+	}
+	if strings.Contains(out, "base64,iVBORw0KGgo=") == false {
+		t.Errorf("the base64 payload was altered:\n%s", out)
+	}
+}

internal/lua/adapters/agentrouter.lua

@@ -5,10 +5,12 @@ adapter.version = "1.0.0"
 adapter.endpoint = "/chat/completions"
 adapter.headers = {}
 
--- AgentRouter 的 WAF 会按客户端指纹白名单校验，只放行官方客户端。
--- 这里通过 build_headers 注入官方客户端 User-Agent 以通过校验。
--- 默认 QwenCode 指纹（实测可通）；可在 source.meta.user_agent 覆盖成 Claude Code 指纹。
-
+-- AgentRouter 的阿里云 WAF 按三重维度校验：出口 IP（必须海外）、
+-- TLS 指纹（JS/Go HTTP 层与 curl 不同）、浏览器指纹 header 集。
+-- 纯 UA 已不够：WAF 拦 12 种 agent CLI UA 中的 10 种（全 405 HTML），
+-- 只有同时带 Origin: qwen.ai + Referer + X-Requested-With 才放行，
+-- 且同 IP 短时间连续 3+ 次请求会被速率封禁，靠 adapter 侧低频调用 +
+-- llmsproxy 的 prefFailStep 冷却自然限制，来源无法控制速率。
 local default_ua = "QwenCode/0.2.0 (linux; x64)"
 
 -- AgentRouter fronts Claude models (claude-opus-4-8), and Claude upstreams
@@ -65,6 +67,13 @@ function adapter.build_headers(meta)
         ["Content-Type"] = "application/json",
         ["Authorization"] = "Bearer " .. meta.api_key,
         ["User-Agent"] = ua,
+        -- QwenCode client fingerprint: WAF rejects bare UA without these.
+        ["Accept"] = "application/json, text/plain, */*",
+        ["Accept-Language"] = "zh-CN,zh;q=0.9",
+        ["Origin"] = "https://qwen.ai",
+        ["Referer"] = "https://qwen.ai/",
+        ["X-Requested-With"] = "XMLHttpRequest",
+        ["Connection"] = "keep-alive",
     }
 end
 

internal/lua/adapters/gemini.lua

@@ -2,11 +2,30 @@ local adapter = {}
 
 adapter.name = "gemini"
 adapter.version = "2.0.0"
-adapter.endpoint = "/v1/models"
+-- The Go layer builds the request URL as base_url + endpoint, statically
+-- (see provider.URL). Gemini's real API is POST
+-- /v1beta/models/{model}:generateContent, and streaming is the same path with
+-- a ":streamGenerateContent" verb -- the model name is part of the PATH, so it
+-- cannot live in a static endpoint string.
+--
+-- "{model}" is therefore a placeholder the Go layer substitutes with the model
+-- this request actually sends (provider.urlFor substitutes it; see
+-- provider.go). Streaming additionally rewrites the ":generateContent" verb to
+-- ":streamGenerateContent" on the same template.
+--
+-- Leaving this as a bare "/v1/models" would call Gemini's model-LIST endpoint,
+-- which answers 405 to POST -- so the preset template would create a source that
+-- can never work. A source that overrides `endpoint:` bypasses the template
+-- entirely and must then spell the whole path itself.
+adapter.endpoint = "/v1beta/models/{model}:generateContent"
 adapter.headers = {}
 
 -- Gemini API: POST /v1/models/{model}:generateContent
 -- Auth: API key in query param ?key=XXX or Authorization: Bearer XXX
+--
+-- 请求方向必须把 OpenAI 的工具调用翻译成 Gemini 的 functionCall /
+-- functionResponse，否则 agent 回放的历史里助手那一轮的调用会凭空消失，
+-- 紧随其后的工具结果就成了「无来源」的孤立结果，模型只能反复重发同一个调用。
 function adapter.transform_request(raw_body)
     local ok, req = pcall(json.decode, raw_body)
     if not ok then return raw_body end
@@ -14,16 +33,23 @@ function adapter.transform_request(raw_body)
     -- 将 OpenAI 风格 content（字符串或 [{type:*}] 数组）拆成 Gemini parts
     local function to_parts(content)
         if type(content) == "string" then
+            if content == "" then return {} end
             return { { text = content } }
         end
         local parts = {}
         for _, p in ipairs(content or {}) do
-            if p.type == "text" then
-                table.insert(parts, { text = p.text })
-            elseif p.type == "image_url" and type(p.image_url) == "table" and p.image_url.url then
-                local mt, b64 = string.match(p.image_url.url, "^data:([^,]+);base64,(.+)$")
-                if b64 then
-                    table.insert(parts, { inline_data = { mime_type = mt or "image/png", data = b64 } })
+            if type(p) == "string" then
+                if p ~= "" then table.insert(parts, { text = p }) end
+            elseif type(p) == "table" then
+                if p.type == "text" then
+                    if p.text ~= nil and p.text ~= "" then
+                        table.insert(parts, { text = p.text })
+                    end
+                elseif p.type == "image_url" and type(p.image_url) == "table" and p.image_url.url then
+                    local mt, b64 = string.match(p.image_url.url, "^data:([^,]+);base64,(.+)$")
+                    if b64 then
+                        table.insert(parts, { inline_data = { mime_type = mt or "image/png", data = b64 } })
+                    end
                 end
             end
         end
@@ -31,11 +57,59 @@ function adapter.transform_request(raw_body)
     end
 
     local contents = {}
+    local system = ""
+    local call_names = {}   -- tool_call_id -> 函数名（functionResponse 只认名字）
+
     for _, m in ipairs(req.messages or {}) do
-        table.insert(contents, {
-            role = (m.role == "assistant") and "model" or m.role,
-            parts = to_parts(m.content)
-        })
+        local role = m.role or "user"
+
+        if role == "system" then
+            if type(m.content) == "string" then
+                system = system .. m.content .. "\n"
+            end
+
+        elseif role == "assistant" then
+            local parts = to_parts(m.content)
+            if type(m.tool_calls) == "table" then
+                for _, tc in ipairs(m.tool_calls) do
+                    if type(tc) == "table" then
+                        local fn = tc["function"] or {}
+                        local args = fn.arguments
+                        if type(args) == "string" and args ~= "" then
+                            local aok, decoded = pcall(json.decode, args)
+                            args = aok and decoded or {}
+                        elseif type(args) ~= "table" then
+                            args = {}
+                        end
+                        if tc.id ~= nil then call_names[tc.id] = fn.name or "" end
+                        table.insert(parts, {
+                            functionCall = { name = fn.name or "", args = args }
+                        })
+                    end
+                end
+            end
+            -- Gemini 不接受空的 parts 数组；无可发送内容的轮次直接跳过
+            if #parts > 0 then
+                table.insert(contents, { role = "model", parts = parts })
+            end
+
+        elseif role == "tool" then
+            local name = call_names[m.tool_call_id] or m.name or ""
+            local text = type(m.content) == "string" and m.content or ""
+            table.insert(contents, {
+                role = "user",
+                parts = { { functionResponse = {
+                    name = name,
+                    response = { content = text },
+                } } }
+            })
+
+        else
+            local parts = to_parts(m.content)
+            if #parts > 0 then
+                table.insert(contents, { role = "user", parts = parts })
+            end
+        end
     end
 
     local gemini_req = {
@@ -46,6 +120,30 @@ function adapter.transform_request(raw_body)
         }
     }
 
+    if system ~= "" then
+        gemini_req.systemInstruction = { parts = { { text = system } } }
+    end
+
+    -- tools -> functionDeclarations
+    if type(req.tools) == "table" then
+        local decls = {}
+        for _, t in ipairs(req.tools) do
+            if type(t) == "table" and type(t["function"]) == "table" then
+                local fn = t["function"]
+                local params = fn.parameters
+                if type(params) ~= "table" then params = { type = "object", properties = {} } end
+                table.insert(decls, {
+                    name = fn.name or "",
+                    description = fn.description or "",
+                    parameters = params,
+                })
+            end
+        end
+        if #decls > 0 then
+            gemini_req.tools = { { functionDeclarations = decls } }
+        end
+    end
+
     if req.stream then
         gemini_req.stream = true
     end

internal/lua/adapters/ollama.lua

@@ -6,6 +6,10 @@ adapter.endpoint = "/api/chat"
 adapter.headers = {}
 
 -- Ollama API 格式：{ model, messages, stream, options:{temperature,num_predict} }
+-- 工具调用必须一并翻译：Ollama 的 assistant 消息用 tool_calls（arguments 是对象
+-- 而非 JSON 字符串），工具结果用 tool 角色 + tool_name。此前这里只复制了
+-- role/content，助手那轮的调用和 tool 消息的归属全部丢失，模型看到无来源的
+-- 工具结果就只能反复重发同一个调用。
 function adapter.transform_request(raw_body)
     local ok, req = pcall(json.decode, raw_body)
     if not ok then return raw_body end
@@ -22,6 +26,7 @@ function adapter.transform_request(raw_body)
     -- 转换 messages 格式（Ollama messages 支持 images base64 数组）
     if req.messages then
         local msgs = {}
+        local call_names = {}   -- tool_call_id -> 函数名
         for _, m in ipairs(req.messages) do
             local text, images
             if type(m.content) == "string" then
@@ -30,22 +35,61 @@ function adapter.transform_request(raw_body)
                 text = ""
                 images = {}
                 for _, p in ipairs(m.content or {}) do
-                    if p.type == "text" then
-                        text = text .. (p.text or "")
-                    elseif p.type == "image_url" and type(p.image_url) == "table" and p.image_url.url then
-                        local b64 = string.match(p.image_url.url, "^data:[^,]+;base64,(.+)$")
-                        if b64 then table.insert(images, b64) end
+                    if type(p) == "table" then
+                        if p.type == "text" then
+                            text = text .. (p.text or "")
+                        elseif p.type == "image_url" and type(p.image_url) == "table" and p.image_url.url then
+                            local b64 = string.match(p.image_url.url, "^data:[^,]+;base64,(.+)$")
+                            if b64 then table.insert(images, b64) end
+                        end
                     end
                 end
                 if #images == 0 then images = nil end
             end
             local msg = { role = m.role, content = text }
             if images then msg.images = images end
+
+            -- 助手轮的工具调用：arguments 转成对象
+            if type(m.tool_calls) == "table" and #m.tool_calls > 0 then
+                local tcs = {}
+                for _, tc in ipairs(m.tool_calls) do
+                    if type(tc) == "table" then
+                        local fn = tc["function"] or {}
+                        local args = fn.arguments
+                        if type(args) == "string" and args ~= "" then
+                            local aok, decoded = pcall(json.decode, args)
+                            args = aok and decoded or {}
+                        elseif type(args) ~= "table" then
+                            args = {}
+                        end
+                        if tc.id ~= nil then call_names[tc.id] = fn.name or "" end
+                        table.insert(tcs, {
+                            ["function"] = { name = fn.name or "", arguments = args }
+                        })
+                    end
+                end
+                if #tcs > 0 then msg.tool_calls = tcs end
+            end
+
+            -- 工具结果：Ollama 用 tool_name 标识归属（不认 tool_call_id）
+            if m.role == "tool" then
+                local name = call_names[m.tool_call_id] or m.name or ""
+                if m.tool_call_id ~= nil then
+                    msg.tool_call_id = m.tool_call_id
+                end
+                if name ~= "" then msg.tool_name = name end
+            end
+
             table.insert(msgs, msg)
         end
         ollama_req.messages = msgs
     end
 
+    -- tools 透传（Ollama 的 shape 与 OpenAI 一致）
+    if type(req.tools) == "table" and #req.tools > 0 then
+        ollama_req.tools = req.tools
+    end
+
     return json.encode(ollama_req)
 end
 

internal/lua/adapters/opencode.lua

@@ -34,8 +34,10 @@ end
 
 -- OpenAI /chat/completions format (pass-through, strip provider-specific fields)
 -- zen 上游 schema 只接受 text content part（无视觉/音频能力）：多模态 part
--- （image_url / input_audio / file 等）一律剥离；因此失去全部 content 的
--- 消息整条丢弃，避免上游 "unknown variant `image_url`, expected `text`"。
+-- （image_url / input_audio / file 等）一律剥离。剥离后 content 变空的消息
+-- 若不再携带 tool_calls / tool_call_id 才整条丢弃（避免上游
+-- "unknown variant `image_url`, expected `text`"）；带工具调用的必须保留，
+-- 否则会把紧随其后的 tool 结果变成孤儿，模型会反复重发同一个调用。
 -- zen 上游角色白名单只有 system / user / assistant / tool / latest_reminder：
 -- OpenAI 的 developer（及 function 等）不在其中，直接透传会触发上游
 -- "unknown variant `developer`, expected one of ..." 错误；统一归一化为 system。
@@ -52,8 +54,14 @@ function adapter.transform_request(raw_body)
     if not ok then return raw_body end
     req.disable_thinking = nil
     req.extra_body = nil
-    if type(req.stream_options) ~= "table" then req.stream_options = {} end
-    req.stream_options.include_usage = true
+    -- stream_options is only valid alongside stream:true. Sending it on a
+    -- non-streaming request is rejected by stricter upstreams (OpenCode Go:
+    -- "stream_options should be set along with stream"), which failed every
+    -- non-stream call. Only attach it when the request actually streams.
+    if req.stream then
+        if type(req.stream_options) ~= "table" then req.stream_options = {} end
+        req.stream_options.include_usage = true
+    end
     if req.messages then
         local kept = {}
         for _, msg in ipairs(req.messages) do
@@ -72,7 +80,24 @@ function adapter.transform_request(raw_body)
                     end
                 end
                 if #parts == 0 then
-                    drop = true
+                    -- Content collapsed to nothing after stripping unsupported
+                    -- parts. A message that still carries a tool call must
+                    -- NEVER be dropped: the very next message is its tool
+                    -- result, and dropping the call orphans that result. The
+                    -- model then sees a result for a call it never made and
+                    -- re-issues the same tool call on every turn (observed as
+                    -- an infinite "repeated tool call" loop).
+                    --
+                    -- This is the common shape for agent clients: an assistant
+                    -- turn whose content is only [thinking, toolCall] serialises
+                    -- to content:[] with tool_calls — exactly the case that used
+                    -- to vanish here. Emit "" instead, which zen accepts.
+                    if (type(msg.tool_calls) == "table" and #msg.tool_calls > 0)
+                        or msg.tool_call_id ~= nil then
+                        msg.content = ""
+                    else
+                        drop = true
+                    end
                 else
                     msg.content = parts
                 end

internal/lua/adapters/opencodego.lua

@@ -0,0 +1,354 @@
+local adapter = {}
+
+adapter.name = "opencodego"
+adapter.version = "1.0.0"
+adapter.endpoint = "/chat/completions"
+
+-- OpenCode Go（包月订阅端点，https://opencode.ai/zen/go/v1）
+--
+-- 与 OpenCode Zen（opencodezen.lua，https://opencode.ai/zen/v1）是两个不同的
+-- 服务，行为要求并不相同，因此各有专用适配器：
+--
+--   * 本适配器（Go）服务包月订阅池。上游**强制要求**
+--     `x-opencode-session` 头，缺失直接 400 "MissingSessionID"。
+--   * Go 的 thinking 模型（deepseek-v4.1-flash 等）**必须回传** assistant 轮的
+--     `reasoning_content`，否则 400：
+--       The `reasoning_content` in the thinking mode must be passed back to the API.
+--     这是本适配器与 Zen 适配器最关键的差异 —— Zen 会剥掉它，Go 必须原样保留。
+--   * Go 上游对协议更严格：非流式请求带 `stream_options` 会被拒
+--     （"stream_options should be set along with stream"）。
+--
+-- 同样需要 opencode 客户端指纹（UA + x-opencode-*）才能被正确识别与路由。
+adapter.headers = {
+    ["User-Agent"] = "opencode/1.18.21 ai-sdk/provider-utils/4.0.23 runtime/bun/1.3.14",
+}
+
+-- 每请求生成身份头。沙箱无 os/math，用 meta.timestamp + 请求体哈希派生：
+-- 同秒内重复请求 id 相同可接受（zen 只校验存在性，不校验格式）。
+local function rand_id(prefix, seed)
+    return prefix .. string.sub(sha256_hex(seed), 1, 24)
+end
+
+-- 会话指纹：取历史里**第一条 user 消息**。
+--
+-- 为什么不直接用客户端的会话 id：实测抓包（tcpdump 抓 127.0.0.1:8081 的真实
+-- agent 请求）确认通用 OpenAI 客户端**根本不发**任何会话标识 —— body 里没有
+-- user / session_id / conversation_id / metadata，请求头也只有 X-Stainless-*
+-- （OpenAI JS SDK）与 User-Agent。opencode 原生客户端那套 x-opencode-session
+-- 是它自己的概念，通用客户端无从转发。
+--
+-- 退而求其次但要够用：历史被逐轮重放，**第一条 user 消息在整个会话里恒定**，
+-- 用它做指纹即可得到"每会话一个 session"，而不是"每源一个 session"。
+-- 拿不到时（无 user 消息）返回空串，调用方回退到按源稳定。
+local function conversation_fingerprint(body)
+    if type(body) ~= "string" or body == "" then return "" end
+    local ok, req = pcall(json.decode, body)
+    if not ok or type(req) ~= "table" then return "" end
+    for _, m in ipairs(req.messages or {}) do
+        if type(m) == "table" and m.role == "user" then
+            local c = m.content
+            if type(c) == "string" then
+                if c ~= "" then return c end
+            elseif type(c) == "table" then
+                for _, part in ipairs(c) do
+                    if type(part) == "table" and part.type == "text" and part.text then
+                        return part.text
+                    end
+                end
+            end
+            return ""
+        end
+    end
+    return ""
+end
+
+-- session_seed 决定上游会话号（x-opencode-session 的种子）。
+--
+-- 优先级：
+--   1) 客户端自带的会话标识（meta.client_session）——pi 等在开启
+--      compat.sendSessionAffinityHeaders 后会发 x-session-affinity，
+--      这是平台真实的会话 id，整个会话恒定且天然按会话隔离。
+--   2) 退路：历史里**第一条 user 消息**做会话指纹。
+--   3) 再退：只按源固定（连 user 消息都没有时）。
+--
+-- 为什么需要 2/3：通用 OpenAI 客户端默认**根本不发**会话标识 —— 实测抓包
+-- （tcpdump 抓 127.0.0.1:8081 真实 agent 请求）确认 body 里没有
+-- user / session_id / conversation_id / metadata，请求头也只有
+-- X-Stainless-*（OpenAI JS SDK）与 User-Agent。opencode 原生客户端那套
+-- x-opencode-session 是它自己的概念，通用客户端无从转发。
+--
+-- 为什么必须稳定：上游前缀缓存是**会话级**的。会话号每请求一变，
+-- 缓存永不命中（实测：固定会号第 2 次命中 5888，每请求换会号则恒为 0）。
+local function session_seed(meta)
+    local src = (meta.source and meta.source.name) or ""
+    local base = "session|llmsproxy|" .. src
+    local cs = meta.client_session
+    if type(cs) == "string" and cs ~= "" then
+        return base .. "|client|" .. cs
+    end
+    return base .. "|" .. conversation_fingerprint(meta.body)
+end
+
+function adapter.build_headers(meta)
+    local ts = tostring(meta.timestamp or "")
+    local src = (meta.source and meta.source.name) or ""
+    return {
+        ["User-Agent"] = adapter.headers["User-Agent"],
+        ["x-opencode-client"] = "cli",
+        -- project 固定：同网关实例共享一个工作区身份
+        ["x-opencode-project"] = string.sub(sha256_hex("llmsproxy|" .. src), 1, 32),
+        -- session 必须**按源稳定**，不能每请求换：上游的前缀缓存在同一 session
+        -- 内才复用。实测（同一段 6032 token 提示词）：
+        --   固定 session  -> 第 2 次命中 5888/6032，cached_tokens=5888
+        --   每请求换 session -> 永远 0 命中
+        -- 原先用 meta.timestamp 派生，等于每请求都是新会话，缓存永远无效，
+        -- 上游也无法做会话亲和路由。
+        ["x-opencode-session"] = rand_id("ses_", session_seed(meta)),
+        -- request id 仍每请求唯一（它只是请求标识，不参与缓存键）
+        ["x-opencode-request"] = rand_id("msg_", "request|" .. ts .. "|" .. tostring(meta.body or "")),
+    }
+end
+
+-- OpenAI /chat/completions format (pass-through, strip provider-specific fields)
+-- zen 上游 schema 只接受 text content part（无视觉/音频能力）：多模态 part
+-- （image_url / input_audio / file 等）一律剥离。剥离后 content 变空的消息
+-- 若不再携带 tool_calls / tool_call_id 才整条丢弃（避免上游
+-- "unknown variant `image_url`, expected `text`"）；带工具调用的必须保留，
+-- 否则会把紧随其后的 tool 结果变成孤儿，模型会反复重发同一个调用。
+-- zen 上游角色白名单只有 system / user / assistant / tool / latest_reminder：
+-- OpenAI 的 developer（及 function 等）不在其中，直接透传会触发上游
+-- "unknown variant `developer`, expected one of ..." 错误；统一归一化为 system。
+local ROLE_WHITELIST = {
+    system = true,
+    user = true,
+    assistant = true,
+    tool = true,
+    latest_reminder = true,
+}
+
+function adapter.transform_request(raw_body)
+    local ok, req = pcall(json.decode, raw_body)
+    if not ok then return raw_body end
+    req.disable_thinking = nil
+    req.extra_body = nil
+    -- stream_options is only valid alongside stream:true; sending it on a
+    -- non-streaming request is rejected by strict upstreams.
+    if req.stream then
+        if type(req.stream_options) ~= "table" then req.stream_options = {} end
+        req.stream_options.include_usage = true
+    end
+    if req.messages then
+        local kept = {}
+        for _, msg in ipairs(req.messages) do
+            if type(msg.role) == "string" and not ROLE_WHITELIST[msg.role] then
+                msg.role = "system"
+            end
+            -- Go 的 thinking 模式**必须**回传 reasoning_content：剥掉它会让
+            -- 上游直接 400（"The `reasoning_content` in the thinking mode must
+            -- be passed back to the API"），agent 的每一轮都会失败。
+            -- 注意：与 opencodezen.lua 的行为**相反**，不要在这里剥。
+            --
+            -- 但绝大多数客户端（pi 等）根本不保存也不回传 reasoning，只保留
+            -- 工具调用本身。上游只在“带 tool_calls 的助手轮”上校验这个字段，
+            -- 实测**空串即可通过校验**，所以缺省时补空串：既满足上游的
+            -- 一致性要求，又不伪造任何推理内容（用户看到的 reasoning 仍然是
+            -- 上游本轮真实返回的）。
+            if msg.role == "assistant"
+                and type(msg.tool_calls) == "table" and #msg.tool_calls > 0
+                and msg.reasoning_content == nil then
+                msg.reasoning_content = ""
+            end
+            local drop = false
+            if type(msg.content) == "table" then
+                local parts = {}
+                for _, part in ipairs(msg.content) do
+                    if type(part) == "table" and part.type ~= nil and part.type ~= "text" then
+                        -- multimodal part not supported by zen
+                    else
+                        table.insert(parts, part)
+                    end
+                end
+                if #parts == 0 then
+                    -- Content collapsed to nothing after stripping unsupported
+                    -- parts. A message that still carries a tool call must
+                    -- NEVER be dropped: the very next message is its tool
+                    -- result, and dropping the call orphans that result. The
+                    -- model then sees a result for a call it never made and
+                    -- re-issues the same tool call on every turn (observed as
+                    -- an infinite "repeated tool call" loop).
+                    if (type(msg.tool_calls) == "table" and #msg.tool_calls > 0)
+                        or msg.tool_call_id ~= nil then
+                        msg.content = ""
+                    else
+                        drop = true
+                    end
+                else
+                    msg.content = parts
+                end
+            end
+            if not drop then
+                table.insert(kept, msg)
+            end
+        end
+        req.messages = kept
+    end
+    return json.encode(req)
+end
+
+function adapter.transform_response(raw_body)
+    local ok, resp = pcall(json.decode, raw_body)
+    if not ok or resp == nil then return raw_body end
+
+    local unified = {
+        content = "",
+        finish_reason = "",
+        token_usage = { prompt = 0, completion = 0, total = 0 }
+    }
+
+    if type(resp.usage) == "table" then
+        unified.token_usage.prompt = resp.usage.prompt_tokens or 0
+        unified.token_usage.completion = resp.usage.completion_tokens or 0
+        unified.token_usage.total = resp.usage.total_tokens or 0
+        local hit = 0
+        if type(resp.usage.prompt_tokens_details) == "table" and resp.usage.prompt_tokens_details.cached_tokens ~= nil then
+            hit = resp.usage.prompt_tokens_details.cached_tokens
+            unified.token_usage.prompt_tokens_details = { cached_tokens = hit }
+        end
+        if (resp.usage.prompt_cache_hit_tokens or 0) > 0 then
+            unified.token_usage.prompt_cache_hit_tokens = resp.usage.prompt_cache_hit_tokens
+            unified.token_usage.prompt_cache_miss_tokens = resp.usage.prompt_cache_miss_tokens or 0
+            if hit == 0 then
+                unified.token_usage.prompt_tokens_details = { cached_tokens = resp.usage.prompt_cache_hit_tokens }
+            end
+        end
+        -- 上游报告输出里有多少是思考 token。不给客户端的话，无法判断
+        -- completion_tokens 里多少是「可见回答」、多少是「思考」（都按输出计费）。
+        local ctd = resp.usage.completion_tokens_details
+        if type(ctd) == "table" and ctd.reasoning_tokens ~= nil then
+            unified.token_usage.completion_tokens_details = { reasoning_tokens = ctd.reasoning_tokens }
+        end
+    end
+
+    -- 本次调用的费用，上游放在**顶层**且是**字符串**（如 "0"、"0.0012"）。
+    -- Go 订阅是包月制，恒为 "0"；只有 Zen 按量付费模型才有信息量。
+    -- 原样透传：网关不解析、不换算、不汇总 —— 它只是上游事实的搬运者。
+    if resp.cost ~= nil then
+        unified.cost = tostring(resp.cost)
+    end
+
+    if type(resp.choices) == "table" and #resp.choices > 0 then
+        local ch = resp.choices[1]
+        if type(ch.message) == "table" then
+            unified.content = ch.message.content or ""
+            local reasoning = ch.message.reasoning_content or ch.message.reasoning
+            if reasoning then
+                unified.reasoning_content = reasoning
+            end
+            if type(ch.message.tool_calls) == "table" then
+                local tcs = {}
+                for _, tc in ipairs(ch.message.tool_calls) do
+                    local args_ok, args = pcall(json.decode, tc["function"].arguments)
+                    if not args_ok then args = {} end
+                    table.insert(tcs, {
+                        id = tc.id,
+                        type = tc.type or "function",
+                        name = tc["function"].name,
+                        arguments = args
+                    })
+                end
+                unified.tool_calls = tcs
+            end
+        end
+        unified.finish_reason = ch.finish_reason or ""
+    end
+
+    return json.encode(unified)
+end
+
+function adapter.transform_stream_chunk(raw_chunk)
+    local ok, chunk = pcall(json.decode, raw_chunk)
+    if not ok then return "" end
+
+    -- OpenAI-style streams may attach usage to a chunk with empty choices
+    -- (the final usage chunk). Preserve it; the gateway emits it as the
+    -- terminal usage chunk. Note: keys must match Go's TokenUsage json tags
+    -- (prompt/completion/total); the gateway re-emits standard *_tokens.
+    local uses = nil
+    if type(chunk.usage) == "table" then
+        uses = {
+            prompt = chunk.usage.prompt_tokens or chunk.usage.prompt or 0,
+            completion = chunk.usage.completion_tokens or chunk.usage.completion or 0,
+            total = chunk.usage.total_tokens or chunk.usage.total or 0,
+        }
+        -- 独立字段与 prompt_tokens_details **并存**透传，不要写成 elseif：
+        -- 上游命中缓存时两者都发，非流式路径也是两个都带。写成 elseif 会让
+        -- 流式丢掉 prompt_cache_hit_tokens/miss，与同一源的**非流式**产出不一致
+        -- （dsh 优先读 details，但只认独立字段的老客户端会看不到缓存）。
+        if type(chunk.usage.prompt_tokens_details) == "table" and chunk.usage.prompt_tokens_details.cached_tokens ~= nil then
+            uses.prompt_tokens_details = { cached_tokens = chunk.usage.prompt_tokens_details.cached_tokens }
+        end
+        if (chunk.usage.prompt_cache_hit_tokens or 0) > 0 then
+            uses.prompt_cache_hit_tokens = chunk.usage.prompt_cache_hit_tokens
+            uses.prompt_cache_miss_tokens = chunk.usage.prompt_cache_miss_tokens or 0
+            if uses.prompt_tokens_details == nil then
+                uses.prompt_tokens_details = { cached_tokens = chunk.usage.prompt_cache_hit_tokens }
+            end
+        end
+        local ctd = chunk.usage.completion_tokens_details
+        if type(ctd) == "table" and ctd.reasoning_tokens ~= nil then
+            uses.completion_tokens_details = { reasoning_tokens = ctd.reasoning_tokens }
+        end
+    end
+
+    if not chunk.choices or #chunk.choices == 0 then
+        -- 空 choices 的帧既不是内容也不是结束信号。上游发两种：
+        --   {"choices":[],"usage":{...}}   —— 终帧用量
+        --   {"choices":[],"cost":"0"}      —— 单独的费用帧
+        -- 两者可能落在同一帧上，必须都转出去；早退只带 usage 会把费用丢干净。
+        local out = { done = false }
+        if uses ~= nil then out.usage = uses end
+        if chunk.cost ~= nil then out.cost = tostring(chunk.cost) end
+        if uses == nil and chunk.cost == nil then return "" end
+        return json.encode(out)
+    end
+    local delta = chunk.choices[1].delta or {}
+    local fr = chunk.choices[1].finish_reason
+
+    local finish = (type(fr) == "string" and fr ~= "") and fr or nil
+
+    local unified = {
+        content = delta.content or "",
+        done = (finish ~= nil)
+    }
+    if finish then
+        unified.finish_reason = finish
+    end
+    if uses ~= nil then
+        unified.usage = uses
+    end
+    -- zen 用 reasoning 字段承载推理文本（OpenAI 惯例是 reasoning_content）
+    local reasoning = delta.reasoning_content or delta.reasoning
+    if reasoning then
+        unified.reasoning_content = reasoning
+    end
+    if delta.tool_calls then
+        -- pass raw streaming fragments through; OpenAI clients accumulate index+id+name+arguments
+        unified.tool_calls = delta.tool_calls
+    end
+    return json.encode(unified)
+end
+
+-- 错误收敛（可选钩子）：zen 错误信封固定为 {error={type,message}}；
+-- 免费池限流 FreeUsageLimitError 单独标注。返回 nil 走通用兜底。
+function adapter.transform_error(status, body)
+    local ok, resp = pcall(json.decode, body)
+    if not ok or type(resp) ~= "table" then return nil end
+    local e = resp.error
+    if type(e) ~= "table" then return nil end
+    if e.type == "FreeUsageLimitError" then
+        return "zen free pool quota exhausted"
+    end
+    return e.message
+end
+
+return adapter

internal/lua/adapters/opencodezen.lua

@@ -0,0 +1,342 @@
+local adapter = {}
+
+adapter.name = "opencodezen"
+adapter.version = "1.0.0"
+adapter.endpoint = "/chat/completions"
+
+-- OpenCode Zen（按量付费端点，https://opencode.ai/zen/v1）
+--
+-- 与 OpenCode Go（opencodego.lua，https://opencode.ai/zen/go/v1）是两个不同的
+-- 服务，行为要求并不相同，因此各有专用适配器：
+--
+--   * 本适配器（Zen）服务免费池 / 按量付费池。免费模型必须带 opencode 客户端
+--     指纹（UA + x-opencode-*），否则上游拒绝："free tier can only be used in
+--     OpenCode"。Zen 的免费池以非 thinking 模型为主，历史上回传
+--     reasoning_content 会引出上游报错，故这里剥掉（Go 侧相反，见 opencodego.lua）。
+--   * Go 订阅读者请看 opencodego.lua：那边**必须**回传 reasoning_content，
+--     否则 thinking 模式直接 400。
+--
+-- 缺 x-opencode-client/session/request/project 身份头会被判为匿名客户端，
+-- 部分模型在带 tools 时上游直接失败（流式返回单 chunk
+-- "finish_reason":"network_error" 空 content，非流式 503
+-- "Endpoint is unavailable"，表现为"空回复"）。
+adapter.headers = {
+    ["User-Agent"] = "opencode/1.18.21 ai-sdk/provider-utils/4.0.23 runtime/bun/1.3.14",
+}
+
+-- 每请求生成身份头。沙箱无 os/math，用 meta.timestamp + 请求体哈希派生：
+-- 同秒内重复请求 id 相同可接受（zen 只校验存在性，不校验格式）。
+local function rand_id(prefix, seed)
+    return prefix .. string.sub(sha256_hex(seed), 1, 24)
+end
+
+-- 会话指纹：取历史里**第一条 user 消息**。
+--
+-- 为什么不直接用客户端的会话 id：实测抓包（tcpdump 抓 127.0.0.1:8081 的真实
+-- agent 请求）确认通用 OpenAI 客户端**根本不发**任何会话标识 —— body 里没有
+-- user / session_id / conversation_id / metadata，请求头也只有 X-Stainless-*
+-- （OpenAI JS SDK）与 User-Agent。opencode 原生客户端那套 x-opencode-session
+-- 是它自己的概念，通用客户端无从转发。
+--
+-- 退而求其次但要够用：历史被逐轮重放，**第一条 user 消息在整个会话里恒定**，
+-- 用它做指纹即可得到"每会话一个 session"，而不是"每源一个 session"。
+-- 拿不到时（无 user 消息）返回空串，调用方回退到按源稳定。
+local function conversation_fingerprint(body)
+    if type(body) ~= "string" or body == "" then return "" end
+    local ok, req = pcall(json.decode, body)
+    if not ok or type(req) ~= "table" then return "" end
+    for _, m in ipairs(req.messages or {}) do
+        if type(m) == "table" and m.role == "user" then
+            local c = m.content
+            if type(c) == "string" then
+                if c ~= "" then return c end
+            elseif type(c) == "table" then
+                for _, part in ipairs(c) do
+                    if type(part) == "table" and part.type == "text" and part.text then
+                        return part.text
+                    end
+                end
+            end
+            return ""
+        end
+    end
+    return ""
+end
+
+-- session_seed 决定上游会话号（x-opencode-session 的种子）。
+--
+-- 优先级：
+--   1) 客户端自带的会话标识（meta.client_session）——pi 等在开启
+--      compat.sendSessionAffinityHeaders 后会发 x-session-affinity，
+--      这是平台真实的会话 id，整个会话恒定且天然按会话隔离。
+--   2) 退路：历史里**第一条 user 消息**做会话指纹。
+--   3) 再退：只按源固定（连 user 消息都没有时）。
+--
+-- 为什么需要 2/3：通用 OpenAI 客户端默认**根本不发**会话标识 —— 实测抓包
+-- （tcpdump 抓 127.0.0.1:8081 真实 agent 请求）确认 body 里没有
+-- user / session_id / conversation_id / metadata，请求头也只有
+-- X-Stainless-*（OpenAI JS SDK）与 User-Agent。opencode 原生客户端那套
+-- x-opencode-session 是它自己的概念，通用客户端无从转发。
+--
+-- 为什么必须稳定：上游前缀缓存是**会话级**的。会话号每请求一变，
+-- 缓存永不命中（实测：固定会号第 2 次命中 5888，每请求换会号则恒为 0）。
+local function session_seed(meta)
+    local src = (meta.source and meta.source.name) or ""
+    local base = "session|llmsproxy|" .. src
+    local cs = meta.client_session
+    if type(cs) == "string" and cs ~= "" then
+        return base .. "|client|" .. cs
+    end
+    return base .. "|" .. conversation_fingerprint(meta.body)
+end
+
+function adapter.build_headers(meta)
+    local ts = tostring(meta.timestamp or "")
+    local src = (meta.source and meta.source.name) or ""
+    return {
+        ["User-Agent"] = adapter.headers["User-Agent"],
+        ["x-opencode-client"] = "cli",
+        -- project 固定：同网关实例共享一个工作区身份
+        ["x-opencode-project"] = string.sub(sha256_hex("llmsproxy|" .. src), 1, 32),
+        -- session 必须**按源稳定**，不能每请求换：上游的前缀缓存在同一 session
+        -- 内才复用。实测（同一段 6032 token 提示词）：
+        --   固定 session  -> 第 2 次命中 5888/6032，cached_tokens=5888
+        --   每请求换 session -> 永远 0 命中
+        -- 原先用 meta.timestamp 派生，等于每请求都是新会话，缓存永远无效，
+        -- 上游也无法做会话亲和路由。
+        ["x-opencode-session"] = rand_id("ses_", session_seed(meta)),
+        -- request id 仍每请求唯一（它只是请求标识，不参与缓存键）
+        ["x-opencode-request"] = rand_id("msg_", "request|" .. ts .. "|" .. tostring(meta.body or "")),
+    }
+end
+
+-- OpenAI /chat/completions format (pass-through, strip provider-specific fields)
+-- zen 上游 schema 只接受 text content part（无视觉/音频能力）：多模态 part
+-- （image_url / input_audio / file 等）一律剥离。剥离后 content 变空的消息
+-- 若不再携带 tool_calls / tool_call_id 才整条丢弃（避免上游
+-- "unknown variant `image_url`, expected `text`"）；带工具调用的必须保留，
+-- 否则会把紧随其后的 tool 结果变成孤儿，模型会反复重发同一个调用。
+-- zen 上游角色白名单只有 system / user / assistant / tool / latest_reminder：
+-- OpenAI 的 developer（及 function 等）不在其中，直接透传会触发上游
+-- "unknown variant `developer`, expected one of ..." 错误；统一归一化为 system。
+local ROLE_WHITELIST = {
+    system = true,
+    user = true,
+    assistant = true,
+    tool = true,
+    latest_reminder = true,
+}
+
+function adapter.transform_request(raw_body)
+    local ok, req = pcall(json.decode, raw_body)
+    if not ok then return raw_body end
+    req.disable_thinking = nil
+    req.extra_body = nil
+    -- stream_options is only valid alongside stream:true; sending it on a
+    -- non-streaming request is rejected by strict upstreams.
+    if req.stream then
+        if type(req.stream_options) ~= "table" then req.stream_options = {} end
+        req.stream_options.include_usage = true
+    end
+    if req.messages then
+        local kept = {}
+        for _, msg in ipairs(req.messages) do
+            if type(msg.role) == "string" and not ROLE_WHITELIST[msg.role] then
+                msg.role = "system"
+            end
+            -- Zen 免费池不接受回传 reasoning_content（Go 侧相反）
+            msg.reasoning_content = nil
+            local drop = false
+            if type(msg.content) == "table" then
+                local parts = {}
+                for _, part in ipairs(msg.content) do
+                    if type(part) == "table" and part.type ~= nil and part.type ~= "text" then
+                        -- multimodal part not supported by zen
+                    else
+                        table.insert(parts, part)
+                    end
+                end
+                if #parts == 0 then
+                    -- Content collapsed to nothing after stripping unsupported
+                    -- parts. A message that still carries a tool call must
+                    -- NEVER be dropped: the very next message is its tool
+                    -- result, and dropping the call orphans that result. The
+                    -- model then sees a result for a call it never made and
+                    -- re-issues the same tool call on every turn (observed as
+                    -- an infinite "repeated tool call" loop).
+                    if (type(msg.tool_calls) == "table" and #msg.tool_calls > 0)
+                        or msg.tool_call_id ~= nil then
+                        msg.content = ""
+                    else
+                        drop = true
+                    end
+                else
+                    msg.content = parts
+                end
+            end
+            if not drop then
+                table.insert(kept, msg)
+            end
+        end
+        req.messages = kept
+    end
+    return json.encode(req)
+end
+
+function adapter.transform_response(raw_body)
+    local ok, resp = pcall(json.decode, raw_body)
+    if not ok or resp == nil then return raw_body end
+
+    local unified = {
+        content = "",
+        finish_reason = "",
+        token_usage = { prompt = 0, completion = 0, total = 0 }
+    }
+
+    if type(resp.usage) == "table" then
+        unified.token_usage.prompt = resp.usage.prompt_tokens or 0
+        unified.token_usage.completion = resp.usage.completion_tokens or 0
+        unified.token_usage.total = resp.usage.total_tokens or 0
+        local hit = 0
+        if type(resp.usage.prompt_tokens_details) == "table" and resp.usage.prompt_tokens_details.cached_tokens ~= nil then
+            hit = resp.usage.prompt_tokens_details.cached_tokens
+            unified.token_usage.prompt_tokens_details = { cached_tokens = hit }
+        end
+        if (resp.usage.prompt_cache_hit_tokens or 0) > 0 then
+            unified.token_usage.prompt_cache_hit_tokens = resp.usage.prompt_cache_hit_tokens
+            unified.token_usage.prompt_cache_miss_tokens = resp.usage.prompt_cache_miss_tokens or 0
+            if hit == 0 then
+                unified.token_usage.prompt_tokens_details = { cached_tokens = resp.usage.prompt_cache_hit_tokens }
+            end
+        end
+        -- 上游报告输出里有多少是思考 token。不给客户端的话，无法判断
+        -- completion_tokens 里多少是「可见回答」、多少是「思考」（都按输出计费）。
+        local ctd = resp.usage.completion_tokens_details
+        if type(ctd) == "table" and ctd.reasoning_tokens ~= nil then
+            unified.token_usage.completion_tokens_details = { reasoning_tokens = ctd.reasoning_tokens }
+        end
+    end
+
+    -- 本次调用的费用，上游放在**顶层**且是**字符串**（如 "0"、"0.0012"）。
+    -- Go 订阅是包月制，恒为 "0"；只有 Zen 按量付费模型才有信息量。
+    -- 原样透传：网关不解析、不换算、不汇总 —— 它只是上游事实的搬运者。
+    if resp.cost ~= nil then
+        unified.cost = tostring(resp.cost)
+    end
+
+    if type(resp.choices) == "table" and #resp.choices > 0 then
+        local ch = resp.choices[1]
+        if type(ch.message) == "table" then
+            unified.content = ch.message.content or ""
+            local reasoning = ch.message.reasoning_content or ch.message.reasoning
+            if reasoning then
+                unified.reasoning_content = reasoning
+            end
+            if type(ch.message.tool_calls) == "table" then
+                local tcs = {}
+                for _, tc in ipairs(ch.message.tool_calls) do
+                    local args_ok, args = pcall(json.decode, tc["function"].arguments)
+                    if not args_ok then args = {} end
+                    table.insert(tcs, {
+                        id = tc.id,
+                        type = tc.type or "function",
+                        name = tc["function"].name,
+                        arguments = args
+                    })
+                end
+                unified.tool_calls = tcs
+            end
+        end
+        unified.finish_reason = ch.finish_reason or ""
+    end
+
+    return json.encode(unified)
+end
+
+function adapter.transform_stream_chunk(raw_chunk)
+    local ok, chunk = pcall(json.decode, raw_chunk)
+    if not ok then return "" end
+
+    -- OpenAI-style streams may attach usage to a chunk with empty choices
+    -- (the final usage chunk). Preserve it; the gateway emits it as the
+    -- terminal usage chunk. Note: keys must match Go's TokenUsage json tags
+    -- (prompt/completion/total); the gateway re-emits standard *_tokens.
+    local uses = nil
+    if type(chunk.usage) == "table" then
+        uses = {
+            prompt = chunk.usage.prompt_tokens or chunk.usage.prompt or 0,
+            completion = chunk.usage.completion_tokens or chunk.usage.completion or 0,
+            total = chunk.usage.total_tokens or chunk.usage.total or 0,
+        }
+        -- 独立字段与 prompt_tokens_details **并存**透传，不要写成 elseif：
+        -- 上游命中缓存时两者都发，非流式路径也是两个都带。写成 elseif 会让
+        -- 流式丢掉 prompt_cache_hit_tokens/miss，与同一源的**非流式**产出不一致
+        -- （dsh 优先读 details，但只认独立字段的老客户端会看不到缓存）。
+        if type(chunk.usage.prompt_tokens_details) == "table" and chunk.usage.prompt_tokens_details.cached_tokens ~= nil then
+            uses.prompt_tokens_details = { cached_tokens = chunk.usage.prompt_tokens_details.cached_tokens }
+        end
+        if (chunk.usage.prompt_cache_hit_tokens or 0) > 0 then
+            uses.prompt_cache_hit_tokens = chunk.usage.prompt_cache_hit_tokens
+            uses.prompt_cache_miss_tokens = chunk.usage.prompt_cache_miss_tokens or 0
+            if uses.prompt_tokens_details == nil then
+                uses.prompt_tokens_details = { cached_tokens = chunk.usage.prompt_cache_hit_tokens }
+            end
+        end
+        local ctd = chunk.usage.completion_tokens_details
+        if type(ctd) == "table" and ctd.reasoning_tokens ~= nil then
+            uses.completion_tokens_details = { reasoning_tokens = ctd.reasoning_tokens }
+        end
+    end
+
+    if not chunk.choices or #chunk.choices == 0 then
+        -- 空 choices 的帧既不是内容也不是结束信号。上游发两种：
+        --   {"choices":[],"usage":{...}}   —— 终帧用量
+        --   {"choices":[],"cost":"0"}      —— 单独的费用帧
+        -- 两者可能落在同一帧上，必须都转出去；早退只带 usage 会把费用丢干净。
+        local out = { done = false }
+        if uses ~= nil then out.usage = uses end
+        if chunk.cost ~= nil then out.cost = tostring(chunk.cost) end
+        if uses == nil and chunk.cost == nil then return "" end
+        return json.encode(out)
+    end
+    local delta = chunk.choices[1].delta or {}
+    local fr = chunk.choices[1].finish_reason
+
+    local finish = (type(fr) == "string" and fr ~= "") and fr or nil
+
+    local unified = {
+        content = delta.content or "",
+        done = (finish ~= nil)
+    }
+    if finish then
+        unified.finish_reason = finish
+    end
+    if uses ~= nil then
+        unified.usage = uses
+    end
+    -- zen 用 reasoning 字段承载推理文本（OpenAI 惯例是 reasoning_content）
+    local reasoning = delta.reasoning_content or delta.reasoning
+    if reasoning then
+        unified.reasoning_content = reasoning
+    end
+    if delta.tool_calls then
+        -- pass raw streaming fragments through; OpenAI clients accumulate index+id+name+arguments
+        unified.tool_calls = delta.tool_calls
+    end
+    return json.encode(unified)
+end
+
+-- 错误收敛（可选钩子）：zen 错误信封固定为 {error={type,message}}；
+-- 免费池限流 FreeUsageLimitError 单独标注。返回 nil 走通用兜底。
+function adapter.transform_error(status, body)
+    local ok, resp = pcall(json.decode, body)
+    if not ok or type(resp) ~= "table" then return nil end
+    local e = resp.error
+    if type(e) ~= "table" then return nil end
+    if e.type == "FreeUsageLimitError" then
+        return "zen free pool quota exhausted"
+    end
+    return e.message
+end
+
+return adapter

internal/lua/adapters/qwen38.lua

@@ -0,0 +1,235 @@
+-- tokenplan / discovery-api (intern-ai.org.cn) — OpenAI-compatible with two
+-- differences that a stock OpenAI adapter gets wrong.
+--
+-- 1. REASONING IS A SEPARATE FIELD, AND IT IS MOST OF THE OUTPUT.
+--    Every response carries `reasoning_content` (non-stream) or
+--    `delta.reasoning_content` (stream), disjoint from `content`. Measured on
+--    qwen3.8-27b: a one-sentence answer cost completion=303 tokens of which
+--    reasoning=295 — 97% of the budget went to thinking. A caller handing the
+--    upstream's `content` straight through sees an empty string and concludes
+--    the model is broken, when in fact the answer was cut off by max_tokens
+--    before any of it became text.
+--
+--    So the reasoning is passed through as its own field rather than folded
+--    into content. Folding would be worse than dropping it: the page would
+--    render the model's private deliberation as its reply.
+--
+-- 2. `finish_reason: "length"` WITH `content: null` IS NORMAL HERE, NOT AN ERROR.
+--    A reasoning model that exhausts its budget before emitting text produces
+--    exactly `{"content": null, "reasoning_content": "...", "finish_reason":
+--    "length"}`. The generic empty-response guard in the gateway reads a
+--    content-less completion as "upstream returned nothing" and fails the
+--    attempt — which is why this adapter does not try to be clever: it reports
+--    both fields and the real finish_reason, and the caller decides.
+--
+-- 3. MODEL NAMES ARE REWRITTEN BY THE SERVER.
+--    Requesting `qwen3.8-27b` answers with `model: "qwen3.8-27b-fp8"`. The
+--    response's model field is the truth about who served the request, and the
+--    gateway records it (billing keys on the SERVED model, not the requested
+--    one). Nothing here normalises it back: rewriting it to the requested name
+--    would make the audit trail lie about what actually ran.
+--
+-- Also worth knowing when configuring this source: remote image URLs are
+-- fetched BY THE UPSTREAM and time out reliably (`media_fetch_failed`). Images
+-- must be sent inline as data: URLs. Measured: base64 worked, a Wikimedia URL
+-- did not.
+
+local adapter = {}
+
+adapter.name = "qwen38"
+adapter.version = "1.0.0"
+adapter.endpoint = "/chat/completions"
+adapter.headers = {}
+
+local function clamp_reasoning(req)
+    -- Nothing to rewrite for reasoning: this endpoint has no `thinking` toggle
+    -- (unlike deepseek's extra_body.thinking). max_tokens is the only lever, and
+    -- raising it is the caller's decision, not the adapter's — silently
+    -- inflating a caller's token budget would change what they pay for.
+end
+
+function adapter.transform_request(raw_body)
+    local ok, req = pcall(json.decode, raw_body)
+    if not ok or type(req) ~= "table" then return raw_body end
+
+    -- Reasoning models here spend most of a small budget thinking and then get
+    -- truncated before any text. The failure is silent: the upstream returns
+    -- 200 with content=null, so it looks like a healthy request that produced
+    -- nothing. Warn rather than rewrite — see clamp_reasoning.
+    if type(req.max_tokens) == "number" and req.max_tokens > 0 and req.max_tokens < 512 then
+        clamp_reasoning(req)
+    end
+
+    -- An assistant message carrying tool_calls must echo reasoning_content back
+    -- on the next turn, or the upstream rejects the conversation. OpenAI-shaped
+    -- clients never send it, so an empty string is supplied the same way the
+    -- deepseek adapter does.
+    if type(req.messages) == "table" then
+        for _, msg in ipairs(req.messages) do
+            if type(msg) == "table" and msg.role == "assistant" then
+                if type(msg.tool_calls) == "table" and msg.tool_calls[1] ~= nil
+                    and msg.reasoning_content == nil then
+                    msg.reasoning_content = ""
+                end
+            end
+        end
+    end
+
+    return json.encode(req)
+end
+
+function adapter.transform_response(raw_body)
+    local ok, resp = pcall(json.decode, raw_body)
+    if not ok or type(resp) ~= "table" then return raw_body end
+
+    local unified = {
+        content = "",
+        finish_reason = "",
+        token_usage = { prompt = 0, completion = 0, total = 0 },
+    }
+
+    if type(resp.usage) == "table" then
+        local u = resp.usage
+        unified.token_usage.prompt = u.prompt_tokens or 0
+        unified.token_usage.completion = u.completion_tokens or 0
+        unified.token_usage.total = u.total_tokens or 0
+        if type(u.prompt_tokens_details) == "table" and u.prompt_tokens_details.cached_tokens ~= nil then
+            unified.token_usage.prompt_tokens_details = { cached_tokens = u.prompt_tokens_details.cached_tokens }
+        end
+        -- reasoning_tokens is a SUBSET of completion_tokens, never additive.
+        --
+        -- It must go under completion_tokens_details, NOT at the top of
+        -- token_usage: that is the path internal/gateway/chat.go reads when it
+        -- fills the audit record's Reasoning field. Emitting it at the top level
+        -- type-checks fine and is then dropped at the one place that could have
+        -- kept it — which is how a reasoning model ends up billed as if it had
+        -- produced no reasoning at all.
+        --
+        -- (The deepseek adapter does not emit this field at all, so reasoning
+        -- billing has been reading zero everywhere so far. That is a separate
+        -- gap; this adapter does not depend on it being fixed.)
+        if type(u.completion_tokens_details) == "table"
+            and u.completion_tokens_details.reasoning_tokens ~= nil then
+            local d = unified.token_usage.completion_tokens_details
+            if type(d) ~= "table" then
+                d = {}
+                unified.token_usage.completion_tokens_details = d
+            end
+            d.reasoning_tokens = u.completion_tokens_details.reasoning_tokens
+        end
+    end
+
+    if type(resp.choices) == "table" and #resp.choices > 0 then
+        local ch = resp.choices[1]
+        if type(ch.message) == "table" then
+            local m = ch.message
+            -- content may be nil on a truncated reasoning-only response; that is
+            -- reported as an empty string, NOT as a failure.
+            if type(m.content) == "string" then
+                unified.content = m.content
+            end
+            if type(m.reasoning_content) == "string" then
+                unified.reasoning_content = m.reasoning_content
+            end
+            if type(m.tool_calls) == "table" then
+                local tcs = {}
+                for _, tc in ipairs(m.tool_calls) do
+                    local fn = tc["function"] or {}
+                    local args_ok, args = pcall(json.decode, fn.arguments or "{}")
+                    if not args_ok then args = {} end
+                    table.insert(tcs, {
+                        id = tc.id,
+                        type = tc.type or "function",
+                        name = fn.name,
+                        arguments = args,
+                    })
+                end
+                unified.tool_calls = tcs
+            end
+        end
+        unified.finish_reason = ch.finish_reason or ""
+    end
+
+    return json.encode(unified)
+end
+
+function adapter.transform_stream_chunk(raw_chunk)
+    local ok, chunk = pcall(json.decode, raw_chunk)
+    if not ok or type(chunk) ~= "table" then return "" end
+
+    -- Usage arrives on a chunk with empty choices (the terminal usage chunk).
+    -- Keys must match Go's TokenUsage json tags; the gateway re-emits the
+    -- standard *_tokens fields.
+    local uses = nil
+    if type(chunk.usage) == "table" then
+        local u = chunk.usage
+        uses = {
+            prompt = u.prompt_tokens or 0,
+            completion = u.completion_tokens or 0,
+            total = u.total_tokens or 0,
+        }
+        if type(u.prompt_tokens_details) == "table" and u.prompt_tokens_details.cached_tokens ~= nil then
+            uses.prompt_tokens_details = { cached_tokens = u.prompt_tokens_details.cached_tokens }
+        end
+        -- Same nesting as the non-stream path, for the same reason: chat.go reads
+        -- reasoning under completion_tokens_details and nothing else.
+        if type(u.completion_tokens_details) == "table"
+            and u.completion_tokens_details.reasoning_tokens ~= nil then
+            uses.completion_tokens_details = {
+                reasoning_tokens = u.completion_tokens_details.reasoning_tokens
+            }
+        end
+    end
+
+    if type(chunk.choices) ~= "table" or #chunk.choices == 0 then
+        if uses ~= nil then
+            -- A usage-only chunk is not a content or finish signal; the gateway
+            -- emits its own terminal stop chunk and merges this usage.
+            return json.encode({ usage = uses, done = false })
+        end
+        return ""
+    end
+
+    local ch = chunk.choices[1]
+    local delta = ch.delta or {}
+    local finish = (type(ch.finish_reason) == "string" and ch.finish_reason ~= "") and ch.finish_reason or nil
+
+    local unified = {
+        content = delta.content or "",
+        done = (finish ~= nil),
+    }
+    if finish then
+        unified.finish_reason = finish
+    end
+    if uses ~= nil then
+        unified.usage = uses
+    end
+    -- The whole point of this adapter: reasoning is streamed as its own channel
+    -- and must stay separate. Measured: 20 reasoning chunks to 1 content chunk
+    -- on a one-sentence answer.
+    if type(delta.reasoning_content) == "string" then
+        unified.reasoning_content = delta.reasoning_content
+    end
+    if type(delta.tool_calls) == "table" then
+        unified.tool_calls = delta.tool_calls
+    end
+    return json.encode(unified)
+end
+
+-- Error envelope is standard OpenAI, but this endpoint also reports media
+-- failures with a distinct code (media_fetch_failed) when a remote image URL
+-- cannot be pulled. Both carry {error:{message,...}}; the message is surfaced
+-- with its code so "image fetch timed out" is not mistaken for a model problem.
+function adapter.transform_error(status, body)
+    local ok, resp = pcall(json.decode, body)
+    if not ok or type(resp) ~= "table" then return nil end
+    local e = resp.error
+    if type(e) ~= "table" then return nil end
+    if type(e.message) ~= "string" then return nil end
+    if type(e.code) == "string" and e.code ~= "" and e.code ~= "invalid_api_key" then
+        return e.message .. " (" .. e.code .. ")"
+    end
+    return e.message
+end
+
+return adapter

internal/lua/billing_period_test.go

@@ -0,0 +1,351 @@
+package lua
+
+import (
+	"encoding/json"
+	"os/exec"
+	"strings"
+	"testing"
+)
+
+// runBillingPage executes the billing page's mount script under node against a
+// DOM stub, drives the period selector, and returns what ended up on screen.
+//
+// It asserts against RENDERED output rather than the plugin's internal
+// helpers. Reaching into the IIFE for daysInWindow/rescale needs a test-only
+// hook in the plugin, and the first version of this file took a long detour
+// through such a hook: the harness's fake `Date` broke toISOString(), the
+// plugin script aborted before assigning the hook, and every assertion
+// compared "" against a real value. The rendered tables are the contract that
+// actually matters, they need no hook, and they cannot pass if the rescale is
+// wrong — which is exactly the bug class under test.
+func runBillingPage(t *testing.T, todayISO, period, captureID string) string {
+	t.Helper()
+	node, err := exec.LookPath("node")
+	if err != nil {
+		t.Skipf("node not available (%v)", err)
+	}
+	page, _ := billingUI(t)
+	js := strings.Join(extractScripts(page), "\n")
+	if !strings.Contains(js, "__billing_setPeriod") {
+		t.Fatal("billing page exposes no period selector")
+	}
+
+	harness := `
+global.window = global;
+global.__written = [];
+global.__errors = [];
+var __realErr = console.error;
+console.error = function () {
+  global.__errors.push("console.error: " + Array.prototype.map.call(arguments, function (a) {
+    return (a && a.message) ? a.message : String(a);
+  }).join(" "));
+  __realErr.apply(console, arguments);
+};
+process.on("uncaughtException", function (e) { global.__errors.push("uncaught: " + String(e && e.message || e)); });
+process.on("unhandledRejection", function (e) { global.__errors.push("unhandled: " + String(e && e.message || e)); });
+
+// Pin "now" for the window under test.
+//
+// Two earlier attempts were wrong here, and both looked like product bugs:
+// replacing the Date constructor wholesale breaks toISOString/UTC; and
+// subclassing (setting __FakeDate.prototype = __RealDate.prototype while
+// returning __RealDate instances) leaves instances WITHOUT a working
+// getUTCDay, so the week window came out NaN and every rendered day became
+// "Invalid time value". A Proxy over the real constructor intercepts only the
+// zero-argument call and leaves instances — and therefore every prototype
+// method — completely untouched.
+var __NOW_MS = new Date("` + todayISO + `T12:34:56Z").getTime();
+var __RealDate = global.Date;
+var __PatchedDate = new Proxy(__RealDate, {
+  construct: function (target, args) {
+    // Reflect.construct, not a hand-spelled argument list: writing
+    // new target(a, b, c, d, e, f, g) passes seven explicit undefineds for a
+    // one-argument call, which Date reads as year..millisecond fields and turns
+    // into an Invalid Date. The symptom was a week window rendering
+    // "Invalid time value" for every day — a harness fault indistinguishable
+    // from a product bug.
+    if (args.length === 0) return Reflect.construct(target, [__NOW_MS]);
+    return Reflect.construct(target, args);
+  },
+  apply: function () { return new __RealDate(__NOW_MS); }
+});
+__PatchedDate.now = function () { return __NOW_MS; };
+__PatchedDate.UTC = __RealDate.UTC;
+__PatchedDate.parse = __RealDate.parse;
+__PatchedDate.prototype = __RealDate.prototype;
+global.Date = __PatchedDate;
+
+global.document = {
+  getElementById: function (id) {
+    if (!global.__els) global.__els = {};
+    if (!global.__els[id]) global.__els[id] = {
+      style: {}, dataset: {}, classList: { add: function(){}, remove: function(){} },
+      options: [{value:"day"},{value:"week"},{value:"month"},{value:"all"}],
+      set innerHTML(v) { if (v && String(v).trim()) global.__written.push(id); this.__h = v; },
+      get innerHTML() { return this.__h || ""; },
+      set textContent(v) { if (v !== undefined && String(v).trim()) global.__written.push(id); this.__t = v; },
+      get textContent() { return this.__t || ""; },
+      set innerText(v) { if (v !== undefined && String(v).trim()) global.__written.push(id); this.__i = v; },
+      get innerText() { return this.__i || ""; },
+      appendChild: function(){}, querySelector: function(){ return null; },
+      querySelectorAll: function(){ return []; }, addEventListener: function(){}
+    };
+    return global.__els[id];
+  },
+  createElement: function () { return { style: {}, dataset: {}, appendChild: function(){}, setAttribute: function(){} }; },
+  addEventListener: function () {},
+};
+
+// Fixture days carry deliberately distinctive costs so a fold over the wrong
+// subset still produces a number, and only a mismatch shows it. 2026-01-05/06/07
+// are Monday/Tuesday/Wednesday; 2025-12-01 is far outside any window here.
+global.__st = {
+  currency: "USD",
+  total: { cost: 999, requests: 999, prompt_tokens: 999999 },
+  by_source: { LIFETIMESRC: { cost: 999, requests: 999 } },
+  by_model: { LIFETIMEMODEL: { cost: 999, requests: 999 } },
+  by_key: { LIFETIMEKEY: { cost: 999, requests: 999 } },
+  by_day: {
+    "2026-01-05": { cost: 10, requests: 1, prompt_tokens: 100, completion_tokens: 10, failures: 0 },
+    "2026-01-06": { cost: 20, requests: 2, prompt_tokens: 200, completion_tokens: 20, failures: 1 },
+    "2026-01-07": { cost: 30, requests: 3, prompt_tokens: 300, completion_tokens: 30, failures: 0 },
+    "2025-12-01": { cost: 777, requests: 77, prompt_tokens: 7700 }
+  },
+  by_day_src: {
+    "2026-01-05": { srcA: { cost: 10, requests: 1 } },
+    "2026-01-06": { srcA: { cost: 20, requests: 2 }, srcB: { cost: 5, requests: 1 } },
+    "2026-01-07": { srcA: { cost: 30, requests: 3 } },
+    "2025-12-01": { srcOLD: { cost: 777, requests: 77 } }
+  },
+  by_day_model: {
+    "2026-01-05": { mA: { cost: 10, requests: 1 } },
+    "2026-01-06": { mA: { cost: 20, requests: 2 } },
+    "2026-01-07": { mA: { cost: 30, requests: 3 } }
+  },
+  by_day_key: {
+    "2026-01-05": { kA: { cost: 10, requests: 1 } },
+    "2026-01-06": { kA: { cost: 20, requests: 2 } },
+    "2026-01-07": { kA: { cost: 30, requests: 3 } }
+  },
+  degraded_reqs: 4242, unpriced_reqs: 77
+};
+global.fetch = function () {
+  return Promise.resolve({ ok: true, json: function () {
+    return Promise.resolve({ plugin: "billing", state: global.__st });
+  }});
+};
+global.pluginAPI = { onTabShown: function () {}, lang: "en" };
+`
+	script := harness + "\n" + js + "\n" +
+		"__billing_setPeriod(" + jsonStr(period) + ");\n" +
+		// Let the period-triggered refresh() settle before reading the DOM.
+		"setTimeout(function () {\n" +
+		"  var el = document.getElementById(" + jsonStr(captureID) + ");\n" +
+		"  console.log('__CAPTURE__' + JSON.stringify({\n" +
+		"    html: el ? String(el.innerHTML) : '(missing)',\n" +
+		"    kpis: String(document.getElementById('billing-kpis').innerHTML),\n" +
+		"    sel: String(document.getElementById('billing-period').value),\n" +
+		"    errors: global.__errors\n" +
+		"  }));\n" +
+		"}, 60);\n"
+
+	cmd := exec.Command(node, "-e", script)
+	out, err := cmd.CombinedOutput()
+	if err != nil {
+		t.Fatalf("node execution failed: %v\n%s", err, out)
+	}
+	o := string(out)
+	idx := strings.Index(o, "__CAPTURE__")
+	if idx < 0 {
+		t.Fatalf("no capture in node output (the plugin script aborted?):\n%s", o)
+	}
+	raw := strings.TrimSpace(o[idx+len("__CAPTURE__"):])
+	if nl := strings.IndexAny(raw, "\r\n"); nl >= 0 {
+		raw = raw[:nl]
+	}
+	var got struct {
+		HTML   string   `json:"html"`
+		KPIs   string   `json:"kpis"`
+		Sel    string   `json:"sel"`
+		Errors []string `json:"errors"`
+	}
+	if err := json.Unmarshal([]byte(raw), &got); err != nil {
+		t.Fatalf("capture is not JSON: %v\n%s", err, raw)
+	}
+	if len(got.Errors) > 0 {
+		t.Fatalf("billing script reported errors: %v", got.Errors)
+	}
+	if got.Sel != period {
+		t.Errorf("selector shows %q after setting %q — the change handler did not take", got.Sel, period)
+	}
+	return got.HTML + "\n@@KPI@@\n" + got.KPIs
+}
+
+func jsonStr(s string) string {
+	b, _ := json.Marshal(s)
+	return string(b)
+}
+
+// TestBillingPeriodDayShowsOnlyThatDay checks the one-day window end to end:
+// the tables must carry that day's rows and drop every other day, and the
+// lifetime sentinel (999 / LIFETIMESRC) must appear nowhere.
+func TestBillingPeriodDayShowsOnlyThatDay(t *testing.T) {
+	got := runBillingPage(t, "2026-01-07", "day", "billing-by-source")
+	if strings.Contains(got, "LIFETIME") || strings.Contains(got, "999") {
+		t.Errorf("a one-day view still shows lifetime figures:\n%s", got)
+	}
+	if strings.Contains(got, "srcB") {
+		t.Errorf("a one-day view shows srcB, which only had traffic on 01-06:\n%s", got)
+	}
+	if strings.Contains(got, "srcOLD") {
+		t.Errorf("a one-day view shows a day from 2025-12:\n%s", got)
+	}
+	if !strings.Contains(got, "srcA") {
+		t.Errorf("a one-day view lost the only source that had traffic that day:\n%s", got)
+	}
+	// The KPI total must be that day's cost (30), not the week (60) and not
+	// the lifetime sentinel (999).
+	if !strings.Contains(got, "30") {
+		t.Errorf("KPI does not show the day's cost 30:\n%s", got)
+	}
+}
+
+// TestBillingPeriodWeekSumsThreeDays is the arithmetic: Mon+Wed rows and a
+// total of 60. A window that started on Sunday or ended early still renders
+// tables, so only the number distinguishes it.
+func TestBillingPeriodWeekSumsThreeDays(t *testing.T) {
+	// 2026-01-07 is a Wednesday: the ISO week holds 01-05, 01-06, 01-07.
+	got := runBillingPage(t, "2026-01-07", "week", "billing-by-source")
+	if strings.Contains(got, "srcOLD") || strings.Contains(got, "LIFETIME") {
+		t.Errorf("weekly view leaks out-of-window or lifetime rows:\n%s", got)
+	}
+	// srcA = 10+20+30 = 60, srcB = 5 (only 01-06, but inside the week).
+	if !strings.Contains(got, "60") {
+		t.Errorf("weekly total should be 60, got:\n%s", got)
+	}
+	if !strings.Contains(got, "srcB") {
+		t.Errorf("weekly view dropped srcB, whose only day (01-06) is in the window:\n%s", got)
+	}
+}
+
+// TestBillingPeriodSundayReachesBackToMonday pins the ISO-week Monday rule
+// from the other side. On Sunday 2026-01-11 the week that already ended is
+// Mon 05..Sun 11, so all three fixture days are in it; a start-on-Sunday bug
+// would still include 01-07 and pass, so this case is paired with the
+// weekday case above rather than relied on alone.
+func TestBillingPeriodSundayReachesBackToMonday(t *testing.T) {
+	got := runBillingPage(t, "2026-01-11", "week", "billing-by-source")
+	if !strings.Contains(got, "60") {
+		t.Errorf("Sunday's week view should still total 60:\n%s", got)
+	}
+	// And the Monday view must NOT include Sunday's (nonexistent) traffic.
+	gotMon := runBillingPage(t, "2026-01-05", "day", "billing-by-source")
+	if strings.Contains(gotMon, "30") {
+		t.Errorf("Monday's single day view shows Wednesday's cost 30:\n%s", gotMon)
+	}
+	if !strings.Contains(gotMon, "10") {
+		t.Errorf("Monday's day view lost Monday's own cost 10:\n%s", gotMon)
+	}
+}
+
+// TestBillingPeriodKpiTotalMatchesWindowDays pins the KPI total itself.
+//
+// The dimension-table tests above cannot catch a broken window or a broken
+// fold: by_source is computed from the per-day DIMENSION map, independently of
+// the total. Mutation verification proved it — disabling the total fold, and
+// shifting the ISO-week start to Sunday, both left every earlier test green.
+// The total is derived from by_day, so it is the one value that moves when
+// either the window membership or the accumulation is wrong.
+func TestBillingPeriodKpiTotalMatchesWindowDays(t *testing.T) {
+	kpis := func(t *testing.T, today, period string) string {
+		t.Helper()
+		return runBillingPage(t, today, period, "billing-by-source")
+	}
+	// 2026-01-07 is a Wednesday; fixture days are 05(10) 06(20) 07(30).
+	// KPI renders through fmtInt, so 30 -> "30", 60 -> "60".
+	if got := kpis(t, "2026-01-07", "day"); !strings.Contains(got, "30") {
+		t.Errorf("day KPI should show that day's cost 30:\n%s", got)
+	}
+	if got := kpis(t, "2026-01-07", "week"); !strings.Contains(got, "60") {
+		t.Errorf("week KPI should total 10+20+30 = 60:\n%s", got)
+	}
+	// Sunday 2026-01-11: the ISO week Mon 05..Sun 11 still holds all three days.
+	if got := kpis(t, "2026-01-11", "week"); !strings.Contains(got, "60") {
+		t.Errorf("Sunday's week KPI should still total 60:\n%s", got)
+	}
+	// The Monday rule from the negative side. A window that started on SUNDAY
+	// would cover Sun 11..Sat 17 and contain none of the fixture days, giving
+	// a total of 0 — so this case is what makes the +6 shift observable. The
+	// positive case alone cannot distinguish it: shifting the start one day
+	// later still happens to contain 01-07.
+	if got := kpis(t, "2026-01-11", "day"); !strings.Contains(got, "0") {
+		t.Errorf("Sunday's single-day window has no fixture traffic and must total 0:\n%s", got)
+	}
+	// Monday 2026-01-12 begins a new week; it contains no fixture days.
+	if got := kpis(t, "2026-01-12", "week"); !strings.Contains(got, "0") {
+		t.Errorf("an empty week should report 0, not stale data:\n%s", got)
+	}
+	// ...whereas the Monday-to-Sunday week that just ended does.
+	if got := kpis(t, "2026-01-12", "day"); !strings.Contains(got, "0") {
+		t.Errorf("Monday the 12th has no fixture traffic and must total 0:\n%s", got)
+	}
+}
+
+// TestBillingPeriodDimensionFoldSumsAcrossDays covers the per-dimension
+// accumulation. Asserting "60 appears somewhere" was NOT enough: mutation
+// verification showed a fold that keeps only the last day still passed,
+// because 60 also appears in the KPI total and in a token column. The
+// assertion must therefore pin srcA's own cost cell, not the page's text.
+func TestBillingPeriodDimensionFoldSumsAcrossDays(t *testing.T) {
+	got := runBillingPage(t, "2026-01-07", "week", "billing-by-source")
+	// srcA exists on all three days at 10/20/30. Summed = 60; last-only = 30.
+	if !strings.Contains(got, "USD 60.0000") {
+		t.Errorf("weekly by-source did not sum srcA across days (want USD 60.0000, "+
+			"last-day-only would be USD 30.0000):\n%s", got)
+	}
+	// srcB exists only on 01-06 at 5, inside the window, so it must survive at
+	// exactly its own value.
+	if !strings.Contains(got, "USD 5.0000") {
+		t.Errorf("weekly by-source lost or altered srcB (want USD 5.0000):\n%s", got)
+	}
+	// srcA's summed cost must also differ from its last-day value in the
+	// rendered table, so an overwrite cannot masquerade as a sum.
+	if strings.Count(got, "USD 60.0000") < 2 {
+		t.Errorf("expected the summed 60 in both the table and the KPI:\n%s", got)
+	}
+}
+
+// state untouched, including the lifetime sentinel and the lifetime-only
+// degraded counter. Without this control, a rescale that silently degraded
+// into "always empty" would pass the narrow-window tests.
+// TestBillingPeriodAllStaysLifetime is the control: "all" must keep the stored
+// state untouched. Without it, a rescale that silently degraded into "always
+// empty" would pass every narrow-window test above.
+func TestBillingPeriodAllStaysLifetime(t *testing.T) {
+	got := runBillingPage(t, "2026-01-07", "all", "billing-by-source")
+	if !strings.Contains(got, "LIFETIMESRC") && !strings.Contains(got, "999") {
+		t.Errorf("the all-time view dropped the lifetime figures:\n%s", got)
+	}
+	// 4242 is the lifetime degraded_reqs; it has no day-keyed tally, so it
+	// must remain visible when the window is "all". The KPI goes through
+	// fmtInt, so it renders grouped ("4,242") — assert the rendered form.
+	if !strings.Contains(got, "4,242") {
+		t.Errorf("the all-time view should still show the lifetime degraded count:\n%s", got)
+	}
+}
+
+// TestBillingPeriodHidesLifetimeOnlyCounters is the mismatch the period view
+// exists to remove: degraded/unpriced have no day-keyed tally, so printing
+// the lifetime number next to a per-window cost is misleading. They must be
+// hidden, not merely relabelled.
+func TestBillingPeriodHidesLifetimeOnlyCounters(t *testing.T) {
+	got := runBillingPage(t, "2026-01-07", "day", "billing-by-source")
+	if strings.Contains(got, "4,242") || strings.Contains(got, "Degraded") {
+		t.Errorf("a one-day view still shows the lifetime degraded counter:\n%s", got)
+	}
+	// ...while "all" does show it (covered above), so this is not just a
+	// dictionary that lost the words.
+	if !strings.Contains(runBillingPage(t, "2026-01-07", "all", "billing-by-source"), "4,242") {
+		t.Error("the all-time view lost the degraded counter — the hide is not period-scoped")
+	}
+}

internal/lua/billing_ui_test.go

@@ -0,0 +1,499 @@
+package lua
+
+import (
+	"encoding/json"
+	"os"
+	"os/exec"
+	"regexp"
+	"strings"
+	"testing"
+)
+
+// The billing page rendered EMPTY in production while its data endpoint returned
+// 200 with real numbers. The cause was one line: render(st) referenced an
+// undefined `s` for two KPI cells, so the ReferenceError aborted the whole
+// render and every table stayed at its initial empty innerHTML.
+//
+// Nothing in the build, the tests or the API surfaced it. This file is the guard
+// for the whole class: a plugin's injected UI that references an undefined name,
+// or that depends on a container the page does not provide, fails silently.
+
+// billingUI returns the injected markup for the billing plugin: the full page
+// mount and the status-page element mount.
+func billingUI(t *testing.T) (page string, statusElement string) {
+	t.Helper()
+	src, err := os.ReadFile("plugins/billing.lua")
+	if err != nil {
+		t.Fatalf("read billing.lua: %v", err)
+	}
+	// Located BY CONTENT, not by index. The plugin also uses a long string for
+	// its inline SVG icon, so "the first long string" is the icon and "the
+	// second" is the page — which is exactly the kind of positional assumption
+	// that breaks the next time an icon or a description is added.
+	page = longStringContaining(t, string(src), "billing-root")
+	statusElement = longStringContaining(t, string(src), "billing-status-tile")
+	return page, statusElement
+}
+
+// longStringContaining returns the [==[ ... ]==] body that contains marker.
+func longStringContaining(t *testing.T, src, marker string) string {
+	t.Helper()
+	re := regexp.MustCompile(`(?s)\[==\[(.*?)\]==\]`)
+	for _, m := range re.FindAllStringSubmatch(src, -1) {
+		if strings.Contains(m[1], marker) {
+			return m[1]
+		}
+	}
+	t.Fatalf("no long string contains %q", marker)
+	return ""
+}
+
+// TestBillingMountScriptExecutes is the guard for the production bug.
+//
+// The Billing page rendered empty while its data endpoint returned 200 with real
+// numbers. Cause: render(st) referenced an undefined `s` for two KPI cells, the
+// ReferenceError aborted the render, and every table kept its initial empty
+// innerHTML. Nothing in the build or the API surfaced it.
+//
+// Two earlier attempts at a static check were both wrong: a "declared names"
+// scan flagged every CSS class inside the inline HTML strings (class, div, td),
+// and a CSS-selector parse of the stylesheet reported the stylesheet itself as
+// broken. Static analysis of JS embedded in HTML strings is the wrong tool.
+//
+// So this actually RUNS the script, in node, against a minimal DOM stub, and
+// fails on any thrown error. Skipped when node is unavailable, with the reason
+// printed — never silently passing as if it had checked.
+func TestBillingMountScriptExecutes(t *testing.T) {
+	page, el := billingUI(t)
+	scripts := extractScripts(page)
+	if len(scripts) == 0 {
+		t.Fatal("no <script> found in the billing page mount")
+	}
+	for i, js := range scripts {
+		assertRendersAndDoesNotThrow(t, i, js)
+	}
+	for i, js := range extractScripts(el) {
+		assertRendersAndDoesNotThrow(t, i, js)
+	}
+}
+
+// assertRendersAndDoesNotThrow executes a mount script under node against a DOM
+// stub and fails on EITHER a thrown/reported error OR an empty render.
+//
+// Checking only for exceptions is not enough, and that is the third wrong
+// attempt at this guard. The plugin's render() opens with `if (!st) return;`,
+// so passing the wrong field (`render(j.stateX)`) is a SILENT no-op: no throw,
+// no console.error, no rejection — just an empty page. Only looking at the
+// produced DOM catches that class.
+func assertRendersAndDoesNotThrow(t *testing.T, idx int, js string) {
+	t.Helper()
+	node, err := exec.LookPath("node")
+	if err != nil {
+		t.Skipf("node not available (%v): cannot execute the injected script", err)
+	}
+
+	stub := `
+global.window = global;
+global.document = {
+  getElementById: function (id) {
+    if (!global.__els) global.__els = {};
+    if (!global.__els[id]) global.__els[id] = {
+      style: {}, dataset: {}, classList: { add: function(){}, remove: function(){} },
+      // Both writes count: the Billing page fills innerHTML, the status-page
+      // tile assigns textContent. Watching only one of them flagged the tile as
+      // "renders nothing" — a false positive that would have taught everyone to
+      // ignore this test.
+      set innerHTML(v) { if (v && String(v).trim()) global.__written.push(id); this.__h = v; },
+      get innerHTML() { return this.__h || ""; },
+      set textContent(v) { if (v !== undefined && String(v).trim()) global.__written.push(id); this.__t = v; },
+      get textContent() { return this.__t || ""; },
+      set innerText(v) { if (v !== undefined && String(v).trim()) global.__written.push(id); this.__i = v; },
+      get innerText() { return this.__i || ""; },
+      appendChild: function(){}, querySelector: function(){ return null; },
+      querySelectorAll: function(){ return []; }, addEventListener: function(){} };
+    return global.__els[id];
+  },
+  createElement: function () { return { style: {}, dataset: {}, appendChild: function(){}, setAttribute: function(){} }; },
+  addEventListener: function () {},
+};
+global.pluginAPI = { onTabShown: function () {} };
+global.fetch = function () {
+  // A payload with real numbers, so a working render produces visible output.
+  return Promise.resolve({ ok: true, json: function () {
+    return Promise.resolve({ plugin: "billing", state: {
+      currency: "USD",
+      total: { cost: 1.25, requests: 7, prompt_tokens: 100, completion_tokens: 20, failures: 0 },
+      by_source: { localzen: { cost: 1.25, requests: 7, prompt_tokens: 100, completion_tokens: 20, failures: 0 } },
+      by_model: { m1: { cost: 1.25, requests: 7, prompt_tokens: 100, completion_tokens: 20, failures: 0 } },
+      by_key: { k1: { cost: 1.25, requests: 7, prompt_tokens: 100, completion_tokens: 20, failures: 0 } },
+      by_day: { "2026-01-01": { cost: 1.25, requests: 7, prompt_tokens: 100, completion_tokens: 20, failures: 0 } },
+      degraded_reqs: 1, unpriced_reqs: 2,
+    }});
+  }});
+};
+global.__written = [];
+global.__errors = [];
+// The plugin's refresh() swallows its own errors so decoration can never break
+// the host page — which is right, and is exactly why the production ReferenceError
+// was invisible. So the trace has to come from console.error, which the plugin now
+// emits. Hooking process events alone made this test pass against the very bug it
+// was written for (verified by re-introducing the typo and watching it stay green).
+var __realErr = console.error;
+console.error = function () {
+  global.__errors.push("console.error: " + Array.prototype.map.call(arguments, function (a) {
+    return (a && a.message) ? a.message : String(a);
+  }).join(" "));
+  __realErr.apply(console, arguments);
+};
+process.on("uncaughtException", function (e) { global.__errors.push("uncaught: " + String(e && e.message || e)); });
+process.on("unhandledRejection", function (e) { global.__errors.push("unhandled: " + String(e && e.message || e)); });
+`
+	// 50ms is not a guess: the plugin's IIFE kicks off refresh() which awaits a
+	// fetch; a rejection lands on the microtask queue almost immediately.
+	script := strings.Join([]string{
+		stub, js,
+		`setTimeout(function(){
+   console.log("__ERRS__" + JSON.stringify({errors: global.__errors, written: global.__written}));
+ }, 300);`,
+	}, "\n")
+	cmd := exec.Command(node, "-e", script)
+	out, err := cmd.CombinedOutput()
+	if err != nil {
+		t.Errorf("mount script %d crashed under node: %v\n%s", idx, err, out)
+		return
+	}
+	// The script may print other things; find the JSON we appended.
+	line := ""
+	for _, l := range strings.Split(string(out), "\n") {
+		if i := strings.Index(l, "__ERRS__"); i >= 0 {
+			line = strings.TrimSpace(l[i+len("__ERRS__"):])
+		}
+	}
+	if line == "" {
+		t.Errorf("mount script %d produced no error report — the harness did not "+
+			"run to completion, so it cannot be trusted to have checked anything", idx)
+		return
+	}
+	var report struct {
+		Errors  []string `json:"errors"`
+		Written []string `json:"written"`
+	}
+	if err := json.Unmarshal([]byte(line), &report); err != nil {
+		t.Errorf("could not parse the error report %q: %v", line, err)
+		return
+	}
+	for _, e := range report.Errors {
+		t.Errorf("mount script %d reported %q — this is the failure that leaves "+
+			"the Billing page blank while the API still returns data", idx, e)
+	}
+	if len(report.Written) == 0 {
+		t.Errorf("mount script %d wrote NOTHING into any container — the page "+
+			"renders empty. render() guards with `if (!st) return`, so a wrong "+
+			"field name is a silent no-op: no throw, no console output, no error "+
+			"anywhere. This is what the production page looked like.", idx)
+	}
+}
+
+func extractScripts(html string) []string {
+	re := regexp.MustCompile(`(?s)<script[^>]*>(.*?)</script>`)
+	var out []string
+	for _, m := range re.FindAllStringSubmatch(html, -1) {
+		out = append(out, m[1])
+	}
+	return out
+}
+
+// TestBillingPageCoversItsData is the other half: the page declares tables for
+// per-source / per-model / per-key / per-day and must actually render into them.
+// A table id that is never written to is exactly how "the page loads and shows
+// nothing" happens without an error.
+func TestBillingPageCoversItsData(t *testing.T) {
+	page, _ := billingUI(t)
+	for _, id := range []string{
+		"billing-kpis", "billing-by-source", "billing-by-model",
+		"billing-by-key", "billing-by-day",
+	} {
+		if !strings.Contains(page, `id="`+id+`"`) {
+			t.Errorf("the page has no container #%s", id)
+		}
+	}
+	// Every container must be written to by the script, not just declared.
+	scripts := extractScripts(page)
+	all := strings.Join(scripts, "\n")
+	for _, id := range []string{"billing-kpis", "billing-by-source", "billing-by-model", "billing-by-key", "billing-by-day"} {
+		if !strings.Contains(all, `getElementById("`+id+`")`) {
+			t.Errorf("#%s is declared but never read by the script — it stays empty forever", id)
+		}
+	}
+}
+
+// TestPluginIconIsNotARawEmoji guards the sidebar icon. The billing plugin
+// declared icon = "💰" and the WebUI drops that verbatim into the button, while
+// every native tab uses an inline SVG styled with `stroke: currentColor`. An
+// emoji there renders at the wrong size and ignores the theme, so it does not
+// match its neighbours — which is what the operator reported.
+func TestPluginIconIsNotARawEmoji(t *testing.T) {
+	icon := billingIcon(mustBillingSource(t))
+	if icon == "" {
+		t.Fatal("billing declares no icon; the sidebar entry would be blank")
+	}
+	if isEmojiIcon(icon) {
+		t.Errorf("page icon is the raw emoji %q — the WebUI sidebar renders "+
+			"native tabs as inline SVG (stroke: currentColor), so an emoji is the "+
+			"wrong size and ignores the theme. Use an inline SVG path instead.", icon)
+	}
+}
+
+// isEmojiIcon reports whether s is a pictographic emoji rather than markup or a
+// text glyph. Codepoints in the pictographic blocks, plus the regional-indicator
+// pair used by flags.
+func isEmojiIcon(s string) bool {
+	r := []rune(s)
+	if len(r) == 0 {
+		return false
+	}
+	// Anything containing '<' is markup (an inline <svg>), which is the fix.
+	if strings.ContainsRune(s, '<') {
+		return false
+	}
+	for _, c := range r {
+		switch {
+		case c >= 0x1F300 && c <= 0x1FAFF, // pictographs, symbols, supplemental
+			c >= 0x1F000 && c <= 0x1F2FF, // mahjong/domino/cards
+			c >= 0x2600 && c <= 0x27BF,   // misc symbols + dingbats
+			c >= 0x2B00 && c <= 0x2BFF,   // arrows/misc symbols
+			c == 0xFE0F,                  // variation selector-16
+			c >= 0x1F1E6 && c <= 0x1F1FF: // regional indicators (flags)
+			return true
+		}
+	}
+	return false
+}
+
+func mustBillingSource(t *testing.T) string {
+	t.Helper()
+	b, err := os.ReadFile("plugins/billing.lua")
+	if err != nil {
+		t.Fatal(err)
+	}
+	return string(b)
+}
+
+// billingIcon extracts the declared page icon. It accepts BOTH a quoted string
+// and a [==[ ... ]==] long string, because an inline SVG cannot be written as a
+// Lua short string without escaping every quote in it.
+func billingIcon(src string) string {
+	if m := regexp.MustCompile(`(?m)^\s*icon\s*=\s*"([^"]*)"`).FindStringSubmatch(src); m != nil {
+		return m[1]
+	}
+	if m := regexp.MustCompile(`(?s)\bicon\s*=\s*\[==\[(.*?)\]==\]`).FindStringSubmatch(src); m != nil {
+		return m[1]
+	}
+	return ""
+}
+
+// TestBillingTableHeaderMatchesRowColumns catches column drift.
+//
+// row() gained cache columns (fresh / cache / cache%) while the header row did
+// not, in the same edit. The result is a table whose cells are shifted one
+// column left from "fresh" onward — so "cache%" sits under "completion" and the
+// last cell has no label. It renders, it has data, and it is wrong in a way that
+// takes a careful read to notice.
+func TestBillingTableHeaderMatchesRowColumns(t *testing.T) {
+	page, _ := billingUI(t)
+	js := strings.Join(extractScripts(page), "\n")
+	if !strings.Contains(js, "<th>") {
+		t.Fatal("no table header found in the billing page script")
+	}
+	hStart := strings.Index(js, "function tableFor")
+	if hStart < 0 {
+		t.Fatal("no tableFor in the billing page script")
+	}
+	// Count the usage table's headers ONLY. The count is taken from tableFor to
+	// the end of the script, which used to be fine when the rule editor lived on
+	// its own page; now that both halves share one script, the rule table's nine
+	// headers were counted too and the check reported 17 vs 8 — a failure about
+	// two unrelated tables. Stop at the rule editor's script.
+	header := js[hStart:]
+	if cut := strings.Index(header, "/api/plugins/billing/rules"); cut > 0 {
+		header = header[:cut]
+	}
+	// Count <th> with OR without attributes: the name column now carries
+	// style='width:22%' so the header and its data cell can be given the same
+	// width, and a bare "<th>" count silently reported 7 vs 8 — the check
+	// failing on the very column it had just been taught to size.
+	th := strings.Count(header, "<th>") + strings.Count(header, "<th ")
+
+	rStart := strings.Index(js, "function row")
+	rEnd := strings.Index(js, "function tableFor")
+	if rStart < 0 || rEnd < 0 || rEnd <= rStart {
+		t.Fatal("could not isolate row()")
+	}
+	row := js[rStart:rEnd]
+	// Each cell closes with </td>; the first cell uses <td><b>..</b></td> so
+	// counting </td> is exact.
+	td := strings.Count(row, "</td>")
+
+	if th != td {
+		t.Errorf("★ header declares %d columns but row() emits %d cells — the "+
+			"table is misaligned from the first differing column on", th, td)
+	}
+}
+
+// TestRuleTableLocksColumnWidthsInAColgroup guards the layout class of bug that
+// only a screenshot caught: two adjacent headers rendered on top of each other,
+// the URL input squeezed to "https:", and the per-model × button landing 78px
+// past its cell onto the neighbouring Delete button.
+//
+// The measurement that found each one:
+//   - table-layout:fixed with widths on <td> → 954px divided into 9 equal
+//     106px columns, every declared width discarded.
+//   - widths on <th> WITHOUT fixed → browser sizes from content and ignores
+//     them; the 70px currency column still collapsed to 48px (24px of input).
+//   - fixed + <th> widths → they take effect, but the model-price cell's width
+//     then depended on WHICH RULE was widest (352px vs 206px).
+//
+// The fix is all three together: fixed layout, a colgroup, and flex children
+// that may shrink.
+func TestRuleTableLocksColumnWidthsInAColgroup(t *testing.T) {
+	page, _ := billingUI(t)
+	js := strings.Join(extractScripts(page), "\n")
+	// Match the ATTRIBUTE, not the bare string: a substring check on
+	// "table-layout:fixed" is satisfied by the explanatory comment that sits
+	// three lines above the tag, so removing the attribute from the table still
+	// passed (mutation-verified). Require it inside a style='...' literal.
+	if !strings.Contains(js, `font-size:12px;table-layout:fixed'`) {
+		t.Error("the rule table must stay table-layout:fixed — with auto layout " +
+			"the <colgroup> widths below are ignored and columns size from content")
+	}
+	colStart := strings.Index(js, "<colgroup>")
+	if colStart < 0 {
+		t.Fatal("the rule table declares no <colgroup>; per-column widths on " +
+			"<th> or <td> are advisory and get recomputed from content")
+	}
+	// The colgroup is built from a JS array of pixel widths, so assert on the
+	// array, not on literal "<col " occurrences — counting the literal found
+	// exactly one and reported "1 columns, want 9" against correct code.
+	if !strings.Contains(js, "[214, 80, 62, 70, 66, 74, 62, 262, 74]") {
+		t.Error("the rule table's colgroup widths changed or vanished; measured " +
+			"values were [214 80 62 70 66 74 62 262 74] for a 954px pane — the " +
+			"delete column needs 74px or the button wraps to two lines")
+	}
+	if !strings.Contains(js, `"<col style='width:" + w + "px'>"`) {
+		t.Error("colgroup entries are no longer emitted as <col style='width:Npx'>")
+	}
+	// Every price-row INPUT must be allowed to shrink, or the row's fixed widths
+	// sum past the cell and the × button lands on the next column. The × button
+	// is the opposite: it must NOT shrink (flex:0 0 auto), because a squashed
+	// delete button is unclickable.
+	for _, needle := range []string{"m-name", "m-p", "m-c"} {
+		i := strings.Index(js, "class='"+needle+"'")
+		if i < 0 {
+			i = strings.Index(js, "class='ghost small "+needle+"'")
+		}
+		if i < 0 {
+			t.Errorf("price row lost the %s input/button", needle)
+			continue
+		}
+		window := js[i : i+220]
+		if !strings.Contains(window, "flex:") {
+			t.Errorf("%s has no flex sizing; the row is 80+96+96+38 ≈ 330px wide "+
+				"against a 262px cell and overflows onto the next column", needle)
+		}
+		if !strings.Contains(window, "min-width:0") {
+			t.Errorf("%s lacks min-width:0; a flex item will not shrink below its "+
+				"content width, which is how the × button ended up 78px past its cell", needle)
+		}
+	}
+	i := strings.Index(js, "class='ghost small m-del'")
+	if i < 0 {
+		t.Fatal("the per-model delete button is gone")
+	}
+	if !strings.Contains(js[i:i+120], "flex:0 0 auto") {
+		t.Error("the per-model × button must be flex:0 0 auto — it is a fixed-size " +
+			"control and must never be squeezed by the flexible inputs")
+	}
+}
+
+// TestBillingUIShowsRecentUnpriced: the page must be able to distinguish a
+// settled pricing debt from a live leak. The lifetime unpriced counter answers
+// neither — it keeps every request from before a rule was added, so it reads the
+// same number in both situations. The day buckets exist to make the difference
+// visible, and a bucket nothing renders is a bucket nobody benefits from.
+func TestBillingUIShowsRecentUnpriced(t *testing.T) {
+	raw, err := os.ReadFile("plugins/billing.lua")
+	if err != nil {
+		t.Fatal(err)
+	}
+	src := string(raw)
+	for _, needle := range []string{
+		"unpriced_by_day",  // the tally is written
+		"recentUnpriced",   // it is reduced to a window
+		"unpriced_recent",  // it reaches the view
+		"unpriced7d",       // the card labels it
+		"unpricedClean",    // and says so when the window is clear
+	} {
+		if !strings.Contains(src, needle) {
+			t.Errorf("billing UI lacks %q", needle)
+		}
+	}
+}
+
+// TestBillingRulesUIExposesReasoningPrice: the thinking-token rate must be
+// editable in the rules editor, not only settable by hand-editing config.yaml.
+// A pricing capability that lives only in the config file is one the operator
+// will assume does not exist.
+func TestBillingRulesUIExposesReasoningPrice(t *testing.T) {
+	raw, err := os.ReadFile("plugins/billing.lua")
+	if err != nil {
+		t.Fatal(err)
+	}
+	src := string(raw)
+	for _, needle := range []string{
+		`class='m-r'`,       // the input, in both the existing rows and the "+ model" row
+		`models[n].reasoning`, // collected back out of the DOM on save
+		`reasoningHint`,     // and it says what leaving it blank means
+	} {
+		if !strings.Contains(src, needle) {
+			t.Errorf("rules editor lacks %q", needle)
+		}
+	}
+	// Two occurrences: an existing model's row and the template for a new one.
+	// One would mean the "+ model" button produces a row that silently drops
+	// the rate on the next save.
+	if n := strings.Count(src, `class='m-r'`); n < 2 {
+		t.Errorf("m-r appears %d time(s); both the row template and the editor need it", n)
+	}
+}
+
+// TestBillingRulesUIExposesLongContextStep: the long-context step must be
+// editable in the rules editor.
+//
+// This is not a niche control. On an agent workload 88% of requests carry more
+// than 128k prompt tokens with a median of 247k, so a rule that omits the step
+// under-bills the MAJORITY of traffic while every number on the dashboard stays
+// plausible. Leaving the capability in config.yaml alone would mean the operator
+// never discovers it needs it.
+func TestBillingRulesUIExposesLongContextStep(t *testing.T) {
+	raw, err := os.ReadFile("plugins/billing.lua")
+	if err != nil {
+		t.Fatal(err)
+	}
+	src := string(raw)
+	for _, needle := range []string{
+		`class='m-lc-a'`,        // threshold
+		`class='m-lc-p'`,        // absolute rate
+		`class='m-lc-m'`,        // multiplier
+		`models[n].long_context`, // collected back out on save
+		`lcHint`,                // explains that blank = no step
+	} {
+		if !strings.Contains(src, needle) {
+			t.Errorf("rules editor lacks %q", needle)
+		}
+	}
+	// Both the existing-row template and the "+ model" template need the three
+	// inputs; one would make a newly added model lose its step on the next save.
+	if n := strings.Count(src, `class='m-lc-a'`); n < 2 {
+		t.Errorf("long-context inputs appear %d time(s); both row templates need them", n)
+	}
+}

internal/lua/fastvalue.go

@@ -0,0 +1,139 @@
+package lua
+
+import (
+	"encoding/json"
+	"strconv"
+)
+
+// This file exists to remove the JSON round-trip from the plugin hook path.
+//
+// Measured before this change: one Fire() call cost 14.6us, of which
+// json.Marshal(payload) was 4.0us and json.Unmarshal back into interface{} was
+// 5.6us — 67% of the whole hook, spent re-deriving a tree that the very next
+// line (pushGoValue) already knows how to walk natively.
+//
+// The round-trip existed because payload arrives as map[string]interface{} and
+// pushGoValue wanted a plain interface{} tree. pushGoValue already handles
+// map[string]interface{} and []interface{} directly, so the fix is a converter
+// that flattens the concrete Go types the gateway actually produces, instead of
+// a generic serialize/parse.
+//
+// The JSON path is kept as the fallback for types the fast path does not know,
+// so a plugin payload carrying something exotic still arrives rather than being
+// silently dropped.
+
+// jsonNumberType is the type json.Unmarshal produces for every JSON number.
+// The fast path recognizes it so numbers survive as numbers rather than being
+// stringified into a different Lua type.
+type jsonNumberType = float64
+
+// fastToPlain converts the concrete Go values the gateway puts in a hook
+// payload into the interface{} tree pushGoValue consumes, without JSON.
+//
+// The interesting cases are the ones json.Unmarshal would have normalized:
+//   - json.Number-ish types (all float64 in practice)
+//   - int / int64 / uint variants, which must land as Lua numbers, not as the
+//     strings a naive "only handle float64" switch would produce
+//   - []string and map[string]string, extremely common in payloads and NOT
+//     handled by a switch that only knows []interface{}
+//
+// Anything unrecognized returns (value, false) so the caller can fall back to
+// the JSON path, which is slower but total.
+func fastToPlain(v interface{}) (interface{}, bool) {
+	switch x := v.(type) {
+	case nil, bool, string, jsonNumberType:
+		return v, true
+	case int:
+		return float64(x), true
+	case int8:
+		return float64(x), true
+	case int16:
+		return float64(x), true
+	case int32:
+		return float64(x), true
+	case int64:
+		return float64(x), true
+	case uint:
+		return float64(x), true
+	case uint8:
+		return float64(x), true
+	case uint16:
+		return float64(x), true
+	case uint32:
+		return float64(x), true
+	case uint64:
+		return float64(x), true
+	case float32:
+		return float64(x), true
+	case []string:
+		out := make([]interface{}, len(x))
+		for i, s := range x {
+			out[i] = s
+		}
+		return out, true
+	case map[string]string:
+		out := make(map[string]interface{}, len(x))
+		for k, s := range x {
+			out[k] = s
+		}
+		return out, true
+	case map[string]interface{}:
+		// The overwhelmingly common case, and the one the hook path always
+		// takes at the top level.
+		out := make(map[string]interface{}, len(x))
+		for k, it := range x {
+			c, ok := fastToPlain(it)
+			if !ok {
+				return nil, false
+			}
+			out[k] = c
+		}
+		return out, true
+	case []interface{}:
+		out := make([]interface{}, len(x))
+		for i, it := range x {
+			c, ok := fastToPlain(it)
+			if !ok {
+				return nil, false
+			}
+			out[i] = c
+		}
+		return out, true
+	default:
+		return nil, false
+	}
+}
+
+// plainForLua returns the interface{} tree to hand to pushGoValue, using the
+// fast path when it can and JSON only when it must.
+func plainForLua(payload map[string]interface{}) interface{} {
+	if fast, ok := fastToPlain(payload); ok {
+		return fast
+	}
+	// Rare: some type the fast path does not model. Serialize and let the
+	// generic decoder normalize it, so the hook still sees the field.
+	b, err := json.Marshal(payload)
+	if err != nil {
+		return payload
+	}
+	var decoded interface{}
+	if err := json.Unmarshal(b, &decoded); err != nil {
+		return payload
+	}
+	return decoded
+}
+
+// toPlainSlice is the []interface{} entry point of fastToPlain, exposed so the
+// Fire path can flatten a pre-built slice without re-walking the map header.
+func toPlainSlice(v []interface{}) []interface{} {
+	if fast, ok := fastToPlain(v); ok {
+		return fast.([]interface{})
+	}
+	return v
+}
+
+// luaNumberString renders a float the way Lua would print it, for the rare case
+// a hook wants the textual form. Kept out of the hot path.
+func luaNumberString(f float64) string {
+	return strconv.FormatFloat(f, 'g', -1, 64)
+}

internal/lua/fastvalue_test.go

@@ -0,0 +1,132 @@
+package lua
+
+import (
+	"encoding/json"
+	"reflect"
+	"testing"
+)
+
+// The hook payload used to be JSON round-tripped on every call. It no longer is,
+// so the fast path and the old JSON path must be indistinguishable — otherwise
+// a plugin silently sees a different payload than before, which is the worst
+// kind of change: it compiles, passes a smoke test, and misprices traffic.
+//
+// These tests therefore compare the two paths on the SAME inputs rather than
+// asserting the fast path's output in isolation.
+
+func viaJSON(t *testing.T, payload map[string]interface{}) interface{} {
+	t.Helper()
+	b, err := json.Marshal(payload)
+	if err != nil {
+		t.Fatalf("marshal: %v", err)
+	}
+	var decoded interface{}
+	if err := json.Unmarshal(b, &decoded); err != nil {
+		t.Fatalf("unmarshal: %v", err)
+	}
+	return decoded
+}
+
+// payloads that exercise every branch of the fast converter.
+func payloadCases() []map[string]interface{} {
+	return []map[string]interface{}{
+		{"model": "deepseek-v4.1-flash", "source": "commandcode", "ok": true},
+		// Numbers of every width: a converter that only knows float64 turns
+		// ints into something else, and a token count that arrives as a string
+		// makes a Lua hook do arithmetic on nil.
+		{"i": 42, "i8": int8(8), "i16": int16(16), "i32": int32(32), "i64": int64(1 << 40),
+			"u": uint(7), "u64": uint64(1 << 50), "f32": float32(1.5), "f64": 2.25},
+		// Zero, negative, and very large values must stay numbers.
+		{"zero": 0, "neg": -17, "huge": 1e308, "tiny": 1e-308},
+		// Slices and string maps: common in payloads and absent from a switch
+		// that only knows []interface{} / map[string]interface{}.
+		{"msgs": []interface{}{"a", "b"}, "tags": []string{"x", "y"}},
+		{"kv": map[string]string{"a": "1", "b": "2"}},
+		// Nesting, which is where a shallow converter silently drops a level.
+		{"usage": map[string]interface{}{
+			"prompt_tokens": 2048, "cache_hit_tokens": 1024,
+			"nested": map[string]interface{}{"deep": []interface{}{1, "two", true, nil}},
+		}},
+		{"nil_field": nil, "empty_map": map[string]interface{}{}, "empty_slice": []interface{}{}},
+		// A value the fast path does NOT model: it must fall back to JSON and
+		// still arrive, not disappear.
+		{"weird": struct {
+			A int    `json:"a"`
+			B string `json:"b"`
+		}{1, "x"}},
+	}
+}
+
+func TestFastPathMatchesJSONPath(t *testing.T) {
+	for i, p := range payloadCases() {
+		want := viaJSON(t, p)
+		got := plainForLua(p)
+		if !reflect.DeepEqual(want, got) {
+			t.Errorf("case %d: fast path differs from JSON path\n payload: %#v\n  json: %#v\n  fast: %#v",
+				i, p, want, got)
+		}
+	}
+}
+
+// TestFastPathIsActuallyUsed guards against the fast path silently degrading to
+// JSON for the payload the gateway really sends. If a future payload gains a
+// type the converter does not model, this still works (it falls back) but the
+// optimization is gone — and the next person measuring the hook would be
+// measuring the old cost without knowing why.
+func TestFastPathIsActuallyUsed(t *testing.T) {
+	// This mirrors the real request_end payload shape from the gateway.
+	realistic := map[string]interface{}{
+		"stage": "request_end", "kind": "end", "model": "deepseek-v4.1-flash",
+		"source": "commandcode", "key": "stress-key", "ok": true,
+		"status": 200, "duration_ms": 1234,
+		"usage": map[string]interface{}{
+			"prompt_tokens": float64(2048), "completion_tokens": float64(512),
+			"cache_hit_tokens": float64(1024), "total_tokens": float64(3584),
+		},
+		"walk": []interface{}{
+			map[string]interface{}{"kind": "tier_skip", "tier": 1, "source": "", "model": "", "reason": "cooldown"},
+			map[string]interface{}{"kind": "selected", "tier": 2, "source": "commandcode", "model": "m", "reason": ""},
+		},
+		"ts": float64(1700000000),
+	}
+	if _, ok := fastToPlain(realistic); !ok {
+		t.Errorf("★ the realistic request payload does NOT take the fast path — " +
+			"the JSON round-trip is still on the hot path for real traffic")
+	}
+}
+
+// TestFastPathDoesNotAliasInput: the converter builds a new tree. If it ever
+// returned the caller's map directly, a Lua hook's writes could not reach Go —
+// but worse, a later mutation of the payload would race with the snapshot the
+// persistence saver is holding.
+func TestFastPathDoesNotAliasInput(t *testing.T) {
+	src := map[string]interface{}{
+		"usage": map[string]interface{}{"prompt_tokens": float64(1)},
+		"list":  []interface{}{"a"},
+	}
+	out, ok := fastToPlain(src)
+	if !ok {
+		t.Fatal("fast path declined a plain payload")
+	}
+	m := out.(map[string]interface{})
+	m["new"] = "added"
+	src["also_new"] = "must not appear"
+
+	if _, leaked := m["also_new"]; leaked {
+		t.Error("output map aliases the input map")
+	}
+	if _, leaked := src["new"]; leaked {
+		t.Error("writing to the output mutated the input")
+	}
+	// And the nested maps must be copies too.
+	inner := m["usage"].(map[string]interface{})
+	inner["prompt_tokens"] = float64(999)
+	if src["usage"].(map[string]interface{})["prompt_tokens"] != float64(1) {
+		t.Error("nested map is shared, not copied — a hook could mutate the payload")
+	}
+	list := m["list"].([]interface{})
+	list[0] = "changed"
+	if src["list"].([]interface{})[0] != "a" {
+		t.Error("nested slice is shared, not copied")
+	}
+}

internal/lua/fire_parallel_test.go

@@ -0,0 +1,334 @@
+package lua
+
+import (
+	"os"
+	"path/filepath"
+	"sync"
+	"testing"
+	"time"
+)
+
+// Fire runs same-stage hooks in parallel across plugins. These tests pin the
+// three things that make that safe, each of which failed at least once while
+// the change was being written.
+
+// twoHookPlugins loads two plugins that both hook request_end.
+func twoHookPlugins(t *testing.T, codeA, codeB string) *Plugins {
+	t.Helper()
+	vm := NewVM(filepath.Join(t.TempDir(), "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatalf("vm start: %v", err)
+	}
+	t.Cleanup(vm.Stop)
+	ps := NewPlugins(vm, filepath.Join(t.TempDir(), "plugins"))
+	ps.save.setWakeDelayForTest(time.Millisecond)
+	t.Cleanup(ps.Close)
+	ps.DisableStatePersistence()
+	if err := ps.LoadSource("aaa", codeA); err != nil {
+		t.Fatalf("load aaa: %v", err)
+	}
+	if err := ps.LoadSource("bbb", codeB); err != nil {
+		t.Fatalf("load bbb: %v", err)
+	}
+	return ps
+}
+
+func counterPlugin(name string, key string) string {
+	// Built by concatenation rather than fmt: %q is not a Go verb, and the first
+	// attempt produced Lua source with a stray '=' that only failed at compile
+	// time inside three different tests.
+	return `
+local p = { name = "` + name + `" }
+p.state = { n = 0 }
+p.hooks = { request_end = "bump" }
+function p.bump(payload)
+  p.state.n = p.state.n + 1
+  if "` + key + `" ~= "" then return { who = "from-` + name + `" } end
+  return nil
+end
+return p
+`
+}
+
+// TestSingleAndParallelPathsMergeIdentically: the single-plugin fast path and
+// the multi-plugin parallel path must produce the same payload.
+//
+// This is not theoretical. The parallel path was written first and the
+// single-plugin path kept its old shape; the merge loop was left off the fast
+// path, so a lone plugin returning a table had its return value DISCARDED. The
+// existing stage-order test caught it — but only because it happened to check
+// the payload after Fire. A plugin that returned fields nobody read would have
+// broken silently.
+func TestSingleAndParallelPathsMergeIdentically(t *testing.T) {
+	single := twoHookPlugins(t, counterPlugin("aaa", "who"), `
+local p = { name = "zzz" }
+p.state = { n = 0 }
+p.hooks = { request_end = "bump" }
+function p.bump(payload) p.state.n = p.state.n + 1 return nil end
+return p
+`)
+	// remove the second so exactly one plugin hooks this stage
+	single.mu.Lock()
+	single.plugins = single.plugins[:1]
+	single.mu.Unlock()
+	single.rebuild()
+
+	ps := twoHookPlugins(t, counterPlugin("aaa", "who"), counterPlugin("bbb", "who"))
+
+	singleOut := single.Fire(StageRequestEnd, map[string]interface{}{"model": "m"})
+	psOut := ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m"})
+
+	if singleOut["who"] != "from-aaa" {
+		t.Errorf("★ single-plugin path dropped the return value: %v", singleOut)
+	}
+	// Two plugins both return `who`; the merge order decides the winner, and it
+	// must be deterministic (load order), not scheduler-dependent.
+	if psOut["who"] != "from-bbb" {
+		t.Errorf("merge order is not load order: who = %v, want from-bbb", psOut["who"])
+	}
+}
+
+// TestMergeOrderIsDeterministic: goroutine completion order must not leak into
+// the result. Running Fire repeatedly must always yield the same payload, or a
+// gateway's behaviour changes run to run with the same plugins installed.
+func TestMergeOrderIsDeterministic(t *testing.T) {
+	// The plugins differ in cost so a scheduler-dependent merge is visible.
+	slow := `
+local p = { name = "aaa" }
+p.state = { n = 0 }
+p.hooks = { request_end = "slow" }
+function p.slow(payload)
+  local acc = 0
+  for i = 1, 3000 do acc = acc + i % 7 end
+  p.state.n = p.state.n + 1
+  return { winner = "aaa", acc = acc }
+end
+return p
+`
+	fast := `
+local p = { name = "bbb" }
+p.state = { n = 0 }
+p.hooks = { request_end = "fast" }
+function p.fast(payload)
+  p.state.n = p.state.n + 1
+  return { winner = "bbb" }
+end
+return p
+`
+	ps := twoHookPlugins(t, slow, fast)
+	first := ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m"})["winner"]
+	for i := 0; i < 60; i++ {
+		got := ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m"})["winner"]
+		if got != first {
+			t.Fatalf("★ winner changed between runs: %q then %q — merge depends on goroutine "+
+				"scheduling, so the same configuration behaves differently run to run", first, got)
+		}
+	}
+	// Load order decides: bbb is loaded second, so it wins.
+	if first != "bbb" {
+		t.Errorf("winner = %q, want bbb (load order decides, not timing)", first)
+	}
+}
+
+// TestThrowingHookDoesNotBreakSiblings covers the failure path that golua
+// actually produces: a hook raising a Lua error.
+//
+// The test name used to claim it covered panics, and it used error() to do it.
+// Probing the six ways a Lua program can fault (error(), indexing nil, calling
+// nil, concatenating nil, arithmetic on nil, unbounded recursion) showed golua
+// converts ALL of them into an error RETURN, not a Go panic — so the recover()
+// in Fire was untested by that case, and deleting recover() still passed. The
+// test was renamed to say what it verifies.
+//
+// recover() is kept anyway: it guards the Go side of Fire (a nil map write, a
+// future change to how the payload is prepared), which is cheap and cannot be
+// triggered from Lua today. Claiming it is covered by a Lua test would be the
+// kind of assurance that evaporates the first time someone checks.
+func TestThrowingHookDoesNotBreakSiblings(t *testing.T) {
+	throwing := `
+local p = { name = "aaa" }
+p.state = {}
+p.hooks = { request_end = "boom" }
+function p.boom(payload) error("intentional failure") end
+return p
+`
+	healthy := counterPlugin("bbb", "")
+	ps := twoHookPlugins(t, throwing, healthy)
+
+	out := ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m"}) // must not crash
+	if out == nil {
+		t.Error("Fire returned nil")
+	}
+	st, ok := ps.State("bbb").(map[string]interface{})
+	if !ok {
+		t.Fatal("healthy plugin has no state")
+	}
+	if n := st["n"]; n != float64(1) {
+		t.Errorf("★ healthy plugin did not run alongside the throwing one: n = %v", n)
+	}
+	he := ps.HookErrors()
+	if len(he) == 0 {
+		t.Error("a throwing hook was not recorded in hook_errors — the failure would be invisible")
+	}
+}
+
+// TestParallelHooksAllRunOnce: every plugin must be invoked exactly once per
+// Fire. A lost or duplicated goroutine shows up as a wrong total, which for the
+// billing plugin means a wrong bill.
+func TestParallelHooksAllRunOnce(t *testing.T) {
+	ps := twoHookPlugins(t, counterPlugin("aaa", ""), counterPlugin("bbb", ""))
+	const fires = 100
+	for i := 0; i < fires; i++ {
+		ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m"})
+	}
+	for _, name := range []string{"aaa", "bbb"} {
+		st, ok := ps.State(name).(map[string]interface{})
+		if !ok {
+			t.Fatalf("%s has no state", name)
+		}
+		if n := st["n"]; n != float64(fires) {
+			t.Errorf("%s counted %v hooks, want %d", name, n, fires)
+		}
+	}
+}
+
+// TestConcurrentFireIsSafe drives Fire from many goroutines at once. Each
+// plugin has its own Lua state and its own mutex, so this must hold; the race
+// detector is what proves it, not the assertions.
+func TestConcurrentFireIsSafe(t *testing.T) {
+	ps := twoHookPlugins(t, counterPlugin("aaa", ""), counterPlugin("bbb", ""))
+	var wg sync.WaitGroup
+	for g := 0; g < 8; g++ {
+		wg.Add(1)
+		go func() {
+			defer wg.Done()
+			for i := 0; i < 50; i++ {
+				ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m", "source": "s"})
+			}
+		}()
+	}
+	wg.Wait()
+	for _, name := range []string{"aaa", "bbb"} {
+		st, ok := ps.State(name).(map[string]interface{})
+		if !ok {
+			t.Fatalf("%s has no state", name)
+		}
+		if n := st["n"]; n != float64(400) {
+			t.Errorf("%s counted %v, want 400", name, n)
+		}
+	}
+}
+
+// TestFireDoesNotHoldRegistryLockWhileRunningHooks: a hook that installs or
+// disables a plugin takes ps.mu for write. If Fire held ps.mu across the hook,
+// that would self-deadlock — the exact failure TrueAgent recorded for its own
+// parallel stop path.
+func TestFireDoesNotHoldRegistryLockWhileRunningHooks(t *testing.T) {
+	reentrant := `
+local p = { name = "aaa" }
+p.state = {}
+p.hooks = { request_end = "reenter" }
+function p.reenter(payload)
+  -- Reading the registry from inside a hook is the read half of the same lock.
+  local _ = #payload
+  return nil
+end
+return p
+`
+	vm := NewVM(filepath.Join(t.TempDir(), "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatal(err)
+	}
+	defer vm.Stop()
+	ps := NewPlugins(vm, filepath.Join(t.TempDir(), "plugins"))
+	defer ps.Close()
+	ps.DisableStatePersistence()
+	if err := ps.LoadSource("aaa", reentrant); err != nil {
+		t.Fatal(err)
+	}
+	if err := ps.LoadSource("bbb", counterPlugin("bbb", "")); err != nil {
+		t.Fatal(err)
+	}
+
+	done := make(chan struct{})
+	go func() {
+		defer close(done)
+		ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m"})
+		// Touch the registry the way an admin request would, right after.
+		_ = ps.Count()
+		_ = ps.List()
+	}()
+	select {
+	case <-done:
+	case <-time.After(10 * time.Second):
+		t.Fatal("Fire deadlocked against the registry lock")
+	}
+}
+
+// TestBillingSurvivesParallelFire runs the REAL bundled plugin next to another
+// one. A synthetic counter cannot catch a mismatch between the documented
+// payload shape and what the plugin actually reads.
+func TestBillingSurvivesParallelFire(t *testing.T) {
+	dir := t.TempDir()
+	pdir := filepath.Join(dir, "plugins")
+	os.MkdirAll(pdir, 0o755)
+	src, err := os.ReadFile("plugins/billing.lua")
+	if err != nil {
+		t.Fatalf("read billing.lua: %v", err)
+	}
+	if err := os.WriteFile(filepath.Join(pdir, "billing.lua"), src, 0o644); err != nil {
+		t.Fatal(err)
+	}
+	if err := os.WriteFile(filepath.Join(pdir, "observer.lua"),
+		[]byte(`local p={name="observer"} p.state={n=0}
+p.hooks={request_end="see"}
+function p.see(payload) p.state.n=p.state.n+1 return nil end
+return p`), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	vm := NewVM(filepath.Join(dir, "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatal(err)
+	}
+	defer vm.Stop()
+	ps := NewPlugins(vm, pdir)
+	ps.save.setWakeDelayForTest(time.Millisecond)
+	defer ps.Close()
+	ps.DisableStatePersistence()
+	if err := ps.LoadDir(); err != nil {
+		t.Fatal(err)
+	}
+	if ps.Count() != 2 {
+		t.Fatalf("loaded %d plugins, want 2", ps.Count())
+	}
+
+	payload := map[string]interface{}{
+		"model": "deepseek-v4.1-flash", "source": "commandcode", "key": "k", "ok": true,
+		"usage": map[string]interface{}{
+			"prompt_tokens": float64(1000), "completion_tokens": float64(100),
+			"cache_hit_tokens": float64(0),
+		},
+	}
+	ps.Fire(StageRequestEnd, payload)
+
+	st := ps.State("billing")
+	if st == nil {
+		t.Fatal("billing produced no state")
+	}
+	m := st.(map[string]interface{})
+	total, ok := m["total"].(map[string]interface{})
+	if !ok {
+		t.Fatalf("billing.total is %T", m["total"])
+	}
+	if total["requests"] != float64(1) {
+		t.Errorf("billing counted %v requests, want 1 — the plugin was starved by the parallel path",
+			total["requests"])
+	}
+	if n := ps.State("observer").(map[string]interface{})["n"]; n != float64(1) {
+		t.Errorf("observer counted %v, want 1", n)
+	}
+	if len(ps.HookErrors()) != 0 {
+		t.Errorf("hook errors under parallel Fire: %v", ps.HookErrors())
+	}
+}

internal/lua/hook_guard_test.go

@@ -0,0 +1,76 @@
+package lua
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// A plugin hook that RAISES must not take the process down. This is the
+// production outage: a Lua error anywhere in a hook made golua's callEx call
+// L.StackTrace(), which calls lua_getinfo and SIGSEGVs on a deep enough stack —
+// a C-level signal Go cannot recover from, so one buggy plugin killed the whole
+// gateway and took every in-flight request with it.
+//
+// The fix routes hook calls through a Lua-side pcall guard, so the error comes
+// back as an ordinary return value. This test fires hooks that raise on purpose
+// and asserts three things: the process survives, the failure is RECORDED, and
+// a well-behaved plugin on the same state keeps working afterwards (the error
+// must not poison the Lua state).
+func TestHookThatRaisesDoesNotCrashTheProcess(t *testing.T) {
+	ps, pdir := billingVM(t)
+	boom := `
+local plugin = {}
+plugin.name = "boom"
+plugin.version = "0.1"
+function plugin.request_end(payload)
+  -- Raise on a table index, the exact shape of the billing bug that caused the
+  -- outage. Deliberately NOT a syntax error: this must load fine and fail only
+  -- when invoked.
+  local x = nil
+  return x.field
+end
+return plugin`
+	if err := os.WriteFile(filepath.Join(pdir, "boom.lua"), []byte(boom), 0644); err != nil {
+		t.Fatal(err)
+	}
+	if err := ps.LoadSource("boom", boom); err != nil {
+		t.Fatalf("load boom: %v", err)
+	}
+
+	payload := map[string]interface{}{
+		"model": "m", "source": "s", "ok": true,
+		"prompt_tokens": 100, "completion_tokens": 10, "time": 1750000000000,
+	}
+	// Fire many times: a single call could pass by luck, but if the error ever
+	// escapes into golua's C path the process dies and this test never returns.
+	for i := 0; i < 50; i++ {
+		ps.Fire(StageRequestEnd, payload)
+	}
+	// Reaching this line at all is the primary assertion.
+
+	errs := ps.HookErrors()
+	end, ok := errs[string(StageRequestEnd)]
+	if !ok {
+		t.Fatal("a raising hook left no record — failures must be observable, not swallowed")
+	}
+	if end["count"] == nil || end["count"].(int) == 0 {
+		t.Error("hook error count is zero despite 50 raising calls")
+	}
+	msg, _ := end["last_error"].(string)
+	if !strings.Contains(msg, "boom") {
+		t.Errorf("last_error does not name the offending plugin: %q", msg)
+	}
+
+	// The billing plugin shares the same Plugins registry and must still work:
+	// one broken plugin may not disable the others.
+	st, _ := ps.State("billing").(map[string]interface{})
+	if st == nil || st["total"] == nil {
+		t.Fatalf("the healthy plugin stopped working after another plugin raised")
+	}
+	tot, _ := st["total"].(map[string]interface{})
+	if tot == nil || tot["requests"] == nil || tot["requests"].(float64) == 0 {
+		t.Errorf("billing recorded no requests after the raising plugin ran: %v", st["total"])
+	}
+}

internal/lua/persist_test.go

@@ -0,0 +1,283 @@
+package lua
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+)
+
+// These cover plugin state persistence: a plugin's totals live in the Lua VM,
+// which dies with the process. Measured on the real gateway before this change:
+// 218,241 requests and 26.2M prompt tokens gone after one systemctl restart.
+
+func persistTestPlugin() string {
+	return `
+local plugin = { name = "counter", version = "1.0" }
+plugin.state = { n = 0, prices_seen = 0 }
+plugin.prices = { rate = 1 }
+plugin.hooks = { request_end = "bump" }
+function plugin.bump(payload)
+  plugin.state.n = plugin.state.n + 1
+  return nil
+end
+return plugin
+`
+}
+
+// newPersistVM wires a plugin registry on a fresh dir, mirroring newPluginVM.
+func newPersistVM(t *testing.T) (*Plugins, string) {
+	t.Helper()
+	dir := t.TempDir()
+	vm := NewVM(filepath.Join(dir, "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatalf("vm start: %v", err)
+	}
+	t.Cleanup(vm.Stop)
+	pdir := filepath.Join(dir, "plugins")
+	ps := NewPlugins(vm, pdir)
+	// Shorten the saver debounce: Close() waits for the saver goroutine, so
+	// with the production 2s interval every test would pay 2s on teardown.
+	ps.save.setWakeDelayForTest(2 * time.Millisecond)
+	t.Cleanup(ps.Close)
+	return ps, pdir
+}
+
+func loadCounter(t *testing.T, ps *Plugins) {
+	t.Helper()
+	if err := ps.LoadSource("counter", persistTestPlugin()); err != nil {
+		t.Fatalf("load: %v", err)
+	}
+}
+
+func stateN(t *testing.T, ps *Plugins) float64 {
+	t.Helper()
+	st := ps.State("counter")
+	m, ok := st.(map[string]interface{})
+	if !ok {
+		t.Fatalf("state is %T, want map", st)
+	}
+	n, ok := m["n"].(float64)
+	if !ok {
+		t.Fatalf("state.n is %T, want float64", m["n"])
+	}
+	return n
+}
+
+// TestStateSurvivesRestart is the defect itself: a rebuilt registry over the
+// same plugin dir must come up with the previous totals, not at zero.
+func TestStateSurvivesRestart(t *testing.T) {
+	ps, pdir := newPersistVM(t)
+	loadCounter(t, ps)
+	for i := 0; i < 25; i++ {
+		ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m", "source": "s"})
+	}
+	if got := stateN(t, ps); got != 25 {
+		t.Fatalf("in-process total = %v, want 25", got)
+	}
+	// Force the write the saver would do, so the test does not depend on timing.
+	ps.save.flush()
+	ps.Close()
+
+	// A brand-new registry over the same dir: this is the restart.
+	vm := NewVM(filepath.Join(t.TempDir(), "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatal(err)
+	}
+	defer vm.Stop()
+	ps2 := NewPlugins(vm, pdir)
+	defer ps2.Close()
+	loadCounter(t, ps2)
+
+	if got := stateN(t, ps2); got != 25 {
+		t.Fatalf("★ total after restart = %v, want 25 — this is the 'restart loses the books' bug", got)
+	}
+}
+
+// TestPricesAndStateAreSeparate: restoring configuration over history (or the
+// reverse) would either erase the totals or resurrect stale prices.
+func TestPricesAndStateAreSeparate(t *testing.T) {
+	ps, pdir := newPersistVM(t)
+	loadCounter(t, ps)
+	for i := 0; i < 7; i++ {
+		ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m", "source": "s"})
+	}
+	ps.save.flush()
+	ps.Close()
+
+	b, err := os.ReadFile(ps.stateFile("counter"))
+	if err != nil {
+		t.Fatalf("no state file: %v", err)
+	}
+	var saved struct {
+		State  map[string]interface{} `json:"state"`
+		Prices map[string]interface{} `json:"prices"`
+	}
+	if err := json.Unmarshal(b, &saved); err != nil {
+		t.Fatalf("state file is not valid JSON: %v", err)
+	}
+	if saved.Prices == nil || saved.Prices["rate"] != float64(1) {
+		t.Errorf("prices were not persisted separately: %v", saved.Prices)
+	}
+	if saved.State["n"] != float64(7) {
+		t.Errorf("state.n = %v, want 7", saved.State["n"])
+	}
+	_ = pdir
+}
+
+// TestPricesOnlyUpdatePersists guards a real hole: SetState returns early for a
+// prices-only payload (correctly leaving state alone), and the first version
+// returned before the persistence write — so a reprice was durable in memory
+// only and a restart silently reverted to the old prices.
+func TestPricesOnlyUpdatePersists(t *testing.T) {
+	ps, _ := newPersistVM(t)
+	loadCounter(t, ps)
+	for i := 0; i < 5; i++ {
+		ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m", "source": "s"})
+	}
+	ps.save.flush()
+
+	err := ps.SetState("counter", map[string]interface{}{
+		"prices": map[string]interface{}{"rate": 42},
+	})
+	if err != nil {
+		t.Fatalf("SetState: %v", err)
+	}
+	if got := stateN(t, ps); got != 5 {
+		t.Errorf("a prices-only payload must not touch state: n = %v, want 5", got)
+	}
+
+	b, err := os.ReadFile(ps.stateFile("counter"))
+	if err != nil {
+		t.Fatalf("no state file after a prices-only update: %v", err)
+	}
+	var saved struct {
+		State  map[string]interface{} `json:"state"`
+		Prices map[string]interface{} `json:"prices"`
+	}
+	if err := json.Unmarshal(b, &saved); err != nil {
+		t.Fatal(err)
+	}
+	if saved.Prices["rate"] != float64(42) {
+		t.Errorf("★ new price not persisted: %v — a restart would revert to the old price", saved.Prices)
+	}
+	if saved.State["n"] != float64(5) {
+		t.Errorf("state was clobbered by a price change: %v", saved.State["n"])
+	}
+}
+
+// TestHookReturnValueSurvivesSnapshot guards the ordering bug: snapshotting
+// plugin.state resets the Lua stack, so doing it BEFORE reading the hook's
+// return value silently turned every opinionated plugin into a silent one —
+// breaking the documented "return a table to merge into payload" contract with
+// no error anywhere.
+func TestHookReturnValueSurvivesSnapshot(t *testing.T) {
+	ps, _ := newPersistVM(t)
+	code := `
+local plugin = { name = "opinionated" }
+plugin.state = { n = 0 }
+plugin.hooks = { request_end = "tag" }
+function plugin.tag(payload)
+  plugin.state.n = plugin.state.n + 1
+  return { cost_usd = 1.25, verdict = "billed" }
+end
+return plugin
+`
+	if err := ps.LoadSource("opinionated", code); err != nil {
+		t.Fatal(err)
+	}
+	out := ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m", "source": "s"})
+	if out["cost_usd"] != 1.25 {
+		t.Errorf("★ hook return value lost: payload = %v", out)
+	}
+	if out["verdict"] != "billed" {
+		t.Errorf("merged field lost: %v", out)
+	}
+	ps.save.flush()
+	st := ps.State("opinionated").(map[string]interface{})
+	if st["n"] != float64(1) {
+		t.Errorf("state.n = %v, want 1 (the hook still ran)", st["n"])
+	}
+}
+
+// TestCorruptStateFileIsNotFatal: a truncated write must degrade to compiled-in
+// defaults, never to a gateway that refuses to start.
+func TestCorruptStateFileIsNotFatal(t *testing.T) {
+	ps, _ := newPersistVM(t)
+	loadCounter(t, ps)
+	for i := 0; i < 3; i++ {
+		ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m", "source": "s"})
+	}
+	ps.save.flush()
+	ps.Close()
+
+	if err := os.WriteFile(ps.stateFile("counter"), []byte("{not json"), 0o644); err != nil {
+		t.Fatal(err)
+	}
+
+	vm := NewVM(filepath.Join(t.TempDir(), "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatal(err)
+	}
+	defer vm.Stop()
+	ps2 := NewPlugins(vm, ps.dir)
+	defer ps2.Close()
+	var warned bool
+	ps2.logf = func(string, ...interface{}) { warned = true }
+	loadCounter(t, ps2)
+
+	if got := stateN(t, ps2); got != 0 {
+		t.Errorf("with a corrupt file the plugin must fall back to its defaults, got n = %v", got)
+	}
+	if !warned {
+		t.Error("a corrupt state file must warn the operator, not fail silently")
+	}
+	// And it must still forward.
+	ps2.Fire(StageRequestEnd, map[string]interface{}{"model": "m", "source": "s"})
+}
+
+// TestFlushIsCoalesced: N mutations must not become N writes. A synchronous
+// per-request write would put a file write on the hot path.
+func TestFlushIsCoalesced(t *testing.T) {
+	ps, _ := newPersistVM(t)
+	loadCounter(t, ps)
+	for i := 0; i < 500; i++ {
+		ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m", "source": "s"})
+	}
+	ps.save.flush()
+	// 500 mutations must collapse to a handful of writes, not 500. The bound is
+	// loose on purpose: the saver also flushes on a timer, so how many writes
+	// happen during the loop depends on how long 500 hook calls take. What must
+	// never happen is one write per mutation.
+	if w := ps.save.writes.Load(); w > 5 {
+		t.Errorf("500 hook calls produced %d file writes; the saver must coalesce", w)
+	}
+	if got := stateN(t, ps); got != 500 {
+		t.Fatalf("n = %v, want 500", got)
+	}
+}
+
+// TestCloseFlushesTail: the shutdown path must not lose the last interval,
+// which would reintroduce the same defect in a smaller window.
+func TestCloseFlushesTail(t *testing.T) {
+	ps, _ := newPersistVM(t)
+	loadCounter(t, ps)
+	ps.Fire(StageRequestEnd, map[string]interface{}{"model": "m", "source": "s"})
+	ps.Close()
+	// Close() already waits for the final flush; no sleep needed.
+
+	b, err := os.ReadFile(ps.stateFile("counter"))
+	if err != nil {
+		t.Fatalf("★ Close() did not flush: %v — a shutdown loses the tail", err)
+	}
+	var saved struct {
+		State map[string]interface{} `json:"state"`
+	}
+	if err := json.Unmarshal(b, &saved); err != nil {
+		t.Fatal(err)
+	}
+	if saved.State["n"] != float64(1) {
+		t.Errorf("flushed n = %v, want 1", saved.State["n"])
+	}
+}

internal/lua/plugin_access.go

@@ -0,0 +1,180 @@
+package lua
+
+import "strings"
+
+// One vocabulary for "who may see / read / call this".
+//
+// Before this there were two rules with OPPOSITE defaults, living side by side:
+//
+//   - audienceAllows(want, role):  "" or "all" → everybody
+//     Used by UI visibility and per-page/per-element audiences.
+//   - HTTPHandler.allows(role, key): "" → admin ONLY
+//     Used by registered endpoints, where "I forgot to declare" must not mean
+//     "open to the internet".
+//
+// Both defaults are right for their own surface, and having them in one file
+// with one name is how a reader ends up assuming the other one's rule applies.
+// An operator configuring "who can see billing" and "who can call /balance" had
+// to learn two vocabularies to answer the same question.
+//
+// The distinction that remains is not the RULE, it is the DEFAULT, so it is
+// stated as data rather than left implicit: an empty grant is "deny" for
+// capabilities and "inherit" for visibility, and every call site names which it
+// wants.
+
+// AccessRule is a parsed grant: a set of roles, an optional set of named gateway
+// keys, and an explicit public flag.
+//
+// Keys REPLACE roles rather than adding to them. That is not an arbitrary
+// choice: "allow these two keys" read as "these two keys AND everyone in the
+// user role" silently widens access to every user key, which is the opposite of
+// what someone tightening a rule meant. `keys = { "@alice" }` therefore means
+// exactly alice.
+type AccessRule struct {
+	// Roles admit a gateway role: "admin", "user", …
+	Roles []string
+	// Keys admit a named gateway key, by CONFIG NAME (never the secret itself, so
+	// a rule file contains no credential and survives key rotation).
+	Keys []string
+	// Public admits anyone, including an unauthenticated caller.
+	Public bool
+}
+
+// AccessAll is the grant that admits everyone.
+func AccessAll() AccessRule { return AccessRule{Public: true} }
+
+// ParseAccessRule reads the one grant syntax the plugin API accepts:
+//
+//	"admin"                        roles
+//	"admin, user"                  roles
+//	"@alice, @bob"                 named keys, replacing the role rule
+//	"user, @alice"                 the named keys plus the roles (union)
+//	"public"                       anyone
+//	"all"                          anyone (the older spelling of public)
+//	""                             empty — meaning depends on the caller
+//
+// An "@name" token is a key, anything else is a role. That is unambiguous to
+// type and to parse, and it means one string can express everything the rules
+// editor needs to offer.
+//
+// "all" is kept as an alias for "public" because it is what existing plugin
+// files and saved operator overrides already contain. Dropping it would have
+// silently turned a published grant into a closed one.
+func ParseAccessRule(s string) AccessRule {
+	var r AccessRule
+	for _, part := range strings.Split(s, ",") {
+		part = strings.TrimSpace(part)
+		if part == "" {
+			continue
+		}
+		if strings.EqualFold(part, "public") || strings.EqualFold(part, "all") {
+			r.Public = true
+			continue
+		}
+		if strings.HasPrefix(part, "@") {
+			if k := strings.TrimSpace(strings.TrimPrefix(part, "@")); k != "" {
+				r.Keys = append(r.Keys, k)
+			}
+			continue
+		}
+		r.Roles = append(r.Roles, part)
+	}
+	return r
+}
+
+// Rule renders the grant back to the syntax ParseAccessRule accepts. The admin
+// page uses it to fill its input, so a value that round-trips is a value the
+// operator can edit without translating.
+func (r AccessRule) Rule() string {
+	var parts []string
+	for _, k := range r.Keys {
+		parts = append(parts, "@"+k)
+	}
+	for _, role := range r.Roles {
+		parts = append(parts, role)
+	}
+	if r.Public {
+		parts = append(parts, "public")
+	}
+	return strings.Join(parts, ", ")
+}
+
+// Empty reports whether the grant names nobody.
+func (r AccessRule) Empty() bool {
+	return len(r.Roles) == 0 && len(r.Keys) == 0 && !r.Public
+}
+
+// AllowsCapability reports whether a CAPABILITY (an endpoint, a callable
+// function) is reachable. Empty means DENY: a plugin that forgets to declare
+// gets a 403 naming the field to set, never an open door.
+//
+// Admin is always admitted. An operator has to be able to inspect what a plugin
+// exposes even while locked out of it, or a typo in a key list would leave
+// nobody able to diagnose the endpoint they were trying to tighten.
+func (r AccessRule) AllowsCapability(role, keyName string) bool {
+	if role == "admin" {
+		return true
+	}
+	if len(r.Keys) > 0 {
+		// A non-empty key list REPLACES the role rule (see the type comment).
+		// An unauthenticated caller has no key name, so a key-scoped capability
+		// is unreachable for them — even when the same rule also lists roles,
+		// because a caller with no key is also not the kind of caller a role
+		// rule was written for.
+		if keyName == "" {
+			return false
+		}
+		for _, k := range r.Keys {
+			if strings.EqualFold(k, keyName) {
+				return true
+			}
+		}
+		// The key list replaced the role rule, so a miss is a deny even though
+		// roles may also be listed.
+		return false
+	}
+	if r.Public {
+		return true
+	}
+	for _, want := range r.Roles {
+		if strings.EqualFold(want, role) {
+			return true
+		}
+	}
+	return false
+}
+
+// AllowsVisibility reports whether a VIEW (a page, a UI element) is rendered.
+// Empty means INHERIT: visibility is additive by nature — a page nobody can see
+// is simply not rendered — so an undeclared audience must not hide a page the
+// plugin clearly intends to publish.
+func (r AccessRule) AllowsVisibility(role string) bool {
+	if r.Empty() || r.Public {
+		return true
+	}
+	if role == "admin" && len(r.Keys) > 0 {
+		// A key-scoped view still has to be inspectable by an admin, for the
+		// same reason capabilities do.
+		return true
+	}
+	for _, want := range r.Roles {
+		if strings.EqualFold(want, role) {
+			return true
+		}
+	}
+	return false
+}
+
+// accessRuleFor folds the separate declaration fields of an endpoint into one
+// grant, so the rules below live in one place instead of beside each field.
+//
+// It is a plain fold, deliberately. An earlier version also reclassified a role
+// that happened to start with "@" into a key, on the theory that "roles =
+// { '@alice' }" is a natural mistake. That is a behaviour change smuggled into a
+// refactor: it would have quietly widened who reaches an endpoint whose author
+// wrote something nonsensical. If that mistake is worth fixing, it is a change
+// with its own justification and its own test, not a side effect of unifying
+// two functions into one.
+func accessRuleFor(roles, keys []string, public bool) AccessRule {
+	return AccessRule{Roles: roles, Keys: keys, Public: public}
+}

internal/lua/plugin_access_persist.go

@@ -0,0 +1,115 @@
+package lua
+
+import (
+	"encoding/json"
+	"os"
+	"sort"
+)
+
+// Access overrides are OPERATOR policy, not plugin data. They live in their own
+// file rather than in each plugin's state because:
+//
+//   - they describe the deployment, not the plugin, so a plugin upgrade that
+//     replaces its .state.json must not silently widen access again;
+//   - they must survive a restart. An in-memory map made "restrict billing to
+//     admin" look like it worked right up to the next deploy, at which point the
+//     restriction evaporated with nothing in the logs to say why;
+//   - the plugin_dir is the one place an operator already looks when asking
+//     "what does this plugin actually get to see", so the answer lives beside it.
+//
+// The file is a flat map of "name" (whole plugin) or "name#path" (one endpoint)
+// to an access spec, which is the same shape the in-memory store uses, so load
+// and save cannot disagree about what a key means.
+
+const accessFileName = ".access.json"
+
+func (ps *Plugins) accessFile() string {
+	if ps.dir == "" {
+		return ""
+	}
+	return ps.dir + string(os.PathSeparator) + accessFileName
+}
+
+// LoadAccessOverrides reads the persisted overrides into memory. A missing file
+// is not an error: no operator has restricted anything yet, which is the
+// default state of a fresh install.
+//
+// A corrupt file is also not fatal. Losing a restriction is bad, but refusing to
+// start the gateway is worse, and an empty map plus a logged error leaves the
+// operator with something to fix.
+func (ps *Plugins) LoadAccessOverrides() error {
+	path := ps.accessFile()
+	if path == "" {
+		return nil
+	}
+	b, err := os.ReadFile(path)
+	if os.IsNotExist(err) {
+		return nil
+	}
+	if err != nil {
+		return err
+	}
+	m := map[string]string{}
+	if err := json.Unmarshal(b, &m); err != nil {
+		return err
+	}
+	ps.mu.Lock()
+	defer ps.mu.Unlock()
+	if ps.visibility == nil {
+		ps.visibility = visibilityOverrides{}
+	}
+	for k, v := range m {
+		ps.visibility[k] = v
+	}
+	return nil
+}
+
+// saveAccessOverridesLocked flushes the whole override map. It is called on
+// every change, and unlike plugin state the map is tiny and changes only when an
+// operator clicks Save — so a synchronous write costs nothing that matters, and
+// it means the change is durable the moment the API returns 200.
+func (ps *Plugins) saveAccessOverridesLocked() {
+	path := ps.accessFile()
+	if path == "" {
+		return
+	}
+	out := make(map[string]string, len(ps.visibility))
+	for k, v := range ps.visibility {
+		if v != "" {
+			out[k] = v
+		}
+	}
+	if len(out) == 0 {
+		// Leave no empty file behind: its absence is the honest encoding of
+		// "nothing is overridden".
+		_ = os.Remove(path)
+		return
+	}
+	writeJSONAtomic(path, out)
+}
+
+// AccessOverrides returns a copy of every override in force, for the UI.
+func (ps *Plugins) accessOverridesSnapshot() map[string]string {
+	ps.mu.RLock()
+	defer ps.mu.RUnlock()
+	if len(ps.visibility) == 0 {
+		return nil
+	}
+	out := make(map[string]string, len(ps.visibility))
+	for k, v := range ps.visibility {
+		out[k] = v
+	}
+	return out
+}
+
+// sortedOverrideKeys exists only so a debug dump is stable; map iteration order
+// would otherwise make two identical states look different.
+func (ps *Plugins) sortedOverrideKeys() []string {
+	m := ps.accessOverridesSnapshot()
+	keys := make([]string, 0, len(m))
+	for k := range m {
+		keys = append(keys, k)
+	}
+	sort.Strings(keys)
+	return keys
+}

internal/lua/plugin_access_persist_test.go

@@ -0,0 +1,110 @@
+package lua
+
+import (
+	"os"
+	"path/filepath"
+	"testing"
+)
+
+// TestAccessOverrideSurvivesReload is the test that matters for an access
+// restriction. An in-memory map makes "billing is admin-only" look like it
+// worked right up to the next restart, and then the restriction is simply gone
+// with nothing in the logs to explain it. Both the plugin-level audience and
+// the per-endpoint spec must come back.
+func TestAccessOverrideSurvivesReload(t *testing.T) {
+	dir := t.TempDir()
+	src := `local P = {}
+P.name = "api"
+P.version = "1.0.0"
+P.http = {
+  { path = "health", handler = "health", roles = { "admin" } },
+}
+function P.health(req) return { json = { ok = true } } end
+return P
+`
+	first := NewPlugins(NewVM(dir), dir)
+	if err := first.LoadSource("api", src); err != nil {
+		t.Fatal(err)
+	}
+	first.SetPluginVisibility("api", "admin, user")
+	if err := first.SetEndpointAccess("api", "health", "@homeagent"); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := os.Stat(filepath.Join(dir, accessFileName)); err != nil {
+		t.Fatalf("override was not written: %v", err)
+	}
+
+	// A fresh Plugins over the same dir is what a restart looks like.
+	second := NewPlugins(NewVM(dir), dir)
+	// The plugin source is on disk too, so this second instance sees both the
+	// plugin and the overrides — the same pair a restart sees. Loading only the
+	// overrides would leave no endpoints to walk, and the assertion below would
+	// pass vacuously.
+	if err := second.LoadSource("api", src); err != nil {
+		t.Fatal(err)
+	}
+	if err := second.LoadAccessOverrides(); err != nil {
+		t.Fatalf("reload overrides: %v", err)
+	}
+	if got := second.VisibilityOverride("api"); got != "admin, user" {
+		t.Errorf("plugin visibility lost across reload: %q", got)
+	}
+	found := false
+	for _, e := range second.EndpointAccess() {
+		if e.Path == "health" && len(e.OverrideKeys) == 1 && e.OverrideKeys[0] == "homeagent" {
+			found = true
+		}
+	}
+	if !found {
+		t.Error("endpoint override lost across reload")
+	}
+}
+
+// TestAccessOverrideClearedRemovesFile keeps the "nothing overridden" state
+// distinguishable from "overridden with something".
+func TestAccessOverrideClearedRemovesFile(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	ps.SetPluginVisibility("api", "admin")
+	path := filepath.Join(dir, accessFileName)
+	if _, err := os.Stat(path); err != nil {
+		t.Fatalf("expected an access file: %v", err)
+	}
+	ps.SetPluginVisibility("api", "")
+	if _, err := os.Stat(path); !os.IsNotExist(err) {
+		t.Errorf("clearing the last override should remove the file, got %v", err)
+	}
+}
+
+// TestVisibilityOverridesExcludeEndpointKeys keeps endpoint policy out of the
+// plugin-audience listing. Both live in one map, and reporting "probe#hello"
+// under visibility is how the plugin page ends up offering to save an audience
+// of that name — which nothing reads as an audience.
+func TestVisibilityOverridesExcludeEndpointKeys(t *testing.T) {
+	dir := t.TempDir()
+	src := `local P = {}
+P.name = "api"
+P.version = "1.0.0"
+P.http = { { path = "health", handler = "health", roles = { "admin" } } }
+function P.health(req) return { json = { ok = true } } end
+return P
+`
+	ps := NewPlugins(NewVM(dir), dir)
+	if err := ps.LoadSource("api", src); err != nil {
+		t.Fatal(err)
+	}
+	ps.SetPluginVisibility("api", "admin, user")
+	if err := ps.SetEndpointAccess("api", "health", "@homeagent"); err != nil {
+		t.Fatal(err)
+	}
+	got := ps.VisibilityOverrides()
+	if got["api"] != "admin, user" {
+		t.Errorf("plugin audience missing: %+v", got)
+	}
+	if _, leaked := got["api#health"]; leaked {
+		t.Errorf("endpoint override leaked into the audience map: %+v", got)
+	}
+	if len(got) != 1 {
+		t.Errorf("expected exactly the plugin audience, got %+v", got)
+	}
+}

internal/lua/plugin_access_test.go

@@ -0,0 +1,152 @@
+package lua
+
+import (
+	"strings"
+	"testing"
+)
+
+// The two functions this file replaces had OPPOSITE defaults:
+//
+//	audienceAllows:  "" / "all"  → everybody
+//	HTTPHandler.allows: ""         → admin ONLY
+//
+// Both are correct for their surface. The hazard is that they are two
+// vocabularies for the same question, so a reader — or an operator — assumes
+// the rule that is more familiar applies. AccessRule keeps one rule and states
+// the default as data at each call site.
+
+// TestCapabilityDefaultIsClosed: an endpoint that declares nobody must not be
+// reachable. This is the property that makes "I forgot" fail safe.
+func TestCapabilityDefaultIsClosed(t *testing.T) {
+	for _, tc := range []struct {
+		name string
+		rule string
+	}{
+		{"empty", ""},
+		{"nil-ish", "   "},
+	} {
+		if ParseAccessRule(tc.rule).AllowsCapability("user", "alice") {
+			t.Errorf("%s: an undeclared capability is reachable by a user", tc.name)
+		}
+		if !ParseAccessRule(tc.rule).AllowsCapability("admin", "alice") {
+			t.Errorf("%s: admin locked out of a capability they must be able to inspect", tc.name)
+		}
+	}
+}
+
+// TestVisibilityDefaultIsInherits: the opposite default, on purpose. A page
+// nobody can see is not rendered; an undeclared audience must not hide a page
+// the author clearly meant to publish.
+func TestVisibilityDefaultIsInherits(t *testing.T) {
+	if !ParseAccessRule("").AllowsVisibility("user") {
+		t.Error("an undeclared audience hides a page")
+	}
+	if ParseAccessRule("admin").AllowsVisibility("user") {
+		t.Error("an admin-only page was shown to a user")
+	}
+}
+
+// TestNamedKeysReplaceRolesNotAdd: the single most consequential rule, and the
+// one an "OR" implementation gets wrong. "allow these two keys" must not also
+// admit every user key.
+func TestNamedKeysReplaceRolesNotAdd(t *testing.T) {
+	r := ParseAccessRule("@alice, @bob")
+	if !r.AllowsCapability("user", "alice") {
+		t.Error("a named key was not admitted")
+	}
+	if r.AllowsCapability("user", "carol") {
+		t.Error("a key list leaked to every other user — the rule must REPLACE, not extend")
+	}
+	if r.AllowsCapability("user", "") {
+		t.Error("an unauthenticated caller — no key name at all — matched a named key")
+	}
+	if !r.AllowsCapability("admin", "") {
+		t.Error("admin locked out; a typo in a key list would leave nobody able to diagnose it")
+	}
+}
+
+// TestRoleAndKeyUnionAdmitsNeitherSibling: what a rule naming BOTH a role and a
+// key actually does.
+//
+// Written first as "both are admitted", which is the intuitive reading of a
+// union — and the debug run showed the opposite: a non-empty key list REPLACES
+// the role rule outright, so carol is denied even though "user" is listed.
+//
+// That surprise is exactly why the rule is named, commented and now tested. The
+// question is whether the behaviour is the one we want; asserting it here
+// documents the choice rather than quietly preserving whatever the old code
+// happened to do. The behaviour matches the pre-existing HTTPHandler.allows, so
+// nothing deployed changes.
+func TestRoleAndKeyUnionAdmitsNeitherSibling(t *testing.T) {
+	r := ParseAccessRule("user, @alice")
+	if !r.AllowsCapability("user", "alice") {
+		t.Error("the named key was not admitted")
+	}
+	if r.AllowsCapability("user", "carol") {
+		t.Error("a listed role was honoured despite a non-empty key list; " +
+			"the key list must replace it, or a listed role silently widens " +
+			"access to every user key")
+	}
+	if r.AllowsCapability("user", "mallory") {
+		t.Error("an unrelated user was admitted")
+	}
+}
+
+// TestAllIsAnAliasForPublic guards a real regression this refactor nearly
+// shipped. "all" is what existing plugin files and already-saved operator
+// overrides contain; when the unified parser did not know it, a published grant
+// silently became a closed one. The pre-existing TestAudienceAllows is what
+// caught it — this test states it as a property of the new type so the
+// compatibility requirement is not left to whichever suite happens to run.
+func TestAllIsAnAliasForPublic(t *testing.T) {
+	r := ParseAccessRule("all")
+	if !r.Public {
+		t.Fatal(`"all" did not parse as public`)
+	}
+	if !r.AllowsCapability("user", "carol") {
+		t.Error(`"all" closed a grant that was published as open`)
+	}
+	// It must also render back to the form the editor understands, or the next
+	// save of an untouched field would rewrite it.
+	if !strings.Contains(strings.ToLower(r.Rule()), "public") {
+		t.Errorf("Rule() = %q, want it to render as public", r.Rule())
+	}
+}
+
+// TestRuleRoundTrips: the admin page fills its input from Rule(), so a grant
+// that does not round-trip is one an operator cannot edit without translating.
+func TestRuleRoundTrips(t *testing.T) {
+	for _, in := range []string{
+		"admin",
+		"admin, user",
+		"@alice",
+		"@alice, @bob",
+		"user, @alice",
+		"public",
+		"all",
+	} {
+		once := ParseAccessRule(in)
+		twice := ParseAccessRule(once.Rule())
+		if once.Rule() != twice.Rule() {
+			t.Errorf("%q -> %q -> %q, not stable", in, once.Rule(), twice.Rule())
+		}
+		if len(twice.Roles) != len(once.Roles) || len(twice.Keys) != len(once.Keys) {
+			t.Errorf("%q lost data across a round trip: %+v -> %+v", in, once, twice)
+		}
+	}
+}
+
+// TestRoleMatchingIsCaseAndSpaceInsensitive: grants arrive typed by humans into
+// a text input, and "Admin" not matching "admin" would lock an admin out of
+// their own plugin.
+func TestRoleMatchingIsCaseAndSpaceInsensitive(t *testing.T) {
+	if !ParseAccessRule(" Admin , User ").AllowsCapability("admin", "x") {
+		t.Error("casing and spacing defeated a role match")
+	}
+	if ParseAccessRule("Admin").AllowsCapability("user", "x") {
+		t.Error("a user was admitted by a case variant")
+	}
+	if !ParseAccessRule(" @Alice ").AllowsCapability("user", "alice") {
+		t.Error("casing defeated a named-key match")
+	}
+}

internal/lua/plugin_clock_test.go

@@ -0,0 +1,95 @@
+package lua
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"testing"
+	"time"
+)
+
+// TestPluginClockDuringNetworkWait pins the reason the probe budget never
+// enforced itself.
+//
+// Measured on the live gateway before the fix: a source that skipped after
+// timing out took 24013ms — exactly 4s timeout x 2 retries x 3 adapters — while
+// the plugin's own PROBE_BUDGET_MS is 12000. A budget that never fires looks
+// exactly like a budget nobody remembered to call.
+//
+// The cause is os.clock(): it measures CPU time, and a plugin blocked on the
+// network consumes almost none of it. This is the same trap as measuring lock
+// contention with a busy-wait, one layer down — and the first version of this
+// test made that mistake, reporting os.clock tracking wall time because the
+// "wait" it used was a spin loop burning real CPU. It had to be re-measured
+// against an actual HTTP request.
+func TestPluginClockDuringNetworkWait(t *testing.T) {
+	dir := t.TempDir()
+	pluginDir := filepath.Join(dir, "plugins")
+	if err := os.MkdirAll(pluginDir, 0o755); err != nil {
+		t.Fatal(err)
+	}
+	vm := NewVM(filepath.Join(dir, "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(vm.Stop)
+	ps := NewPlugins(vm, pluginDir)
+	t.Cleanup(ps.Close)
+	ps.SetOutboundHosts([]string{"127.0.0.1"})
+
+	code := `
+local P = {}
+P.name = "clocktest"
+P.version = "1.0.0"
+P.exports = { "measure" }
+P.measure = function(payload)
+  local c0, w0 = os.clock(), os.time()
+  -- The real thing: block on the network. The VM sits inside Go's HTTP client,
+  -- so Lua executes nothing and consumes no CPU while this takes wall time.
+  __llmsproxy_http({ url = payload.url, method = "GET", timeout_ms = 5000 })
+  return {
+    clock = os.clock() - c0,
+    wall = os.time() - w0,
+  }
+end
+return P
+`
+	if err := os.WriteFile(filepath.Join(pluginDir, "clocktest.lua"), []byte(code), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		time.Sleep(3 * time.Second)
+		_, _ = w.Write([]byte(`{}`))
+	}))
+	defer srv.Close()
+
+	if err := ps.LoadDir(); err != nil {
+		t.Fatal(err)
+	}
+	p := ps.find("clocktest")
+	if p == nil {
+		t.Fatal("plugin not found")
+	}
+	if p.LoadError != "" {
+		t.Fatalf("plugin failed to compile: %s", p.LoadError)
+	}
+	out, err := ps.Call("clocktest", "measure", map[string]interface{}{"url": srv.URL + "/slow"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	m, _ := out.(map[string]interface{})
+	clock, _ := m["clock"].(float64)
+	wall, _ := m["wall"].(float64)
+	t.Logf("3s blocked on HTTP: os.clock advanced %.3fs, os.time advanced %.0fs", clock, wall)
+
+	if wall < 3 {
+		t.Fatalf("the request did not take 3 wall seconds (got %.0f)", wall)
+	}
+	if clock > wall/2 {
+		t.Fatalf("os.clock tracked wall time (%.3f vs %.0f) — the budget is not the cause; "+
+			"look elsewhere", clock, wall)
+	}
+	t.Logf("CONFIRMED: blocked on the network for %.0fs, os.clock advanced %.3fs. "+
+		"A budget written in os.clock never fires while a plugin waits on I/O.", wall, clock)
+}

internal/lua/plugin_decode.go

@@ -0,0 +1,244 @@
+package lua
+
+import (
+	"fmt"
+
+	golua "github.com/aarzilli/golua/lua"
+	"sort"
+	"strings"
+)
+
+// The declarative decoder.
+//
+// Why this exists: plugin declarations were read by three hand-written functions
+// (readPluginUI, readPluginExports, readPluginHTTP) that each re-implemented the
+// same walk over a decoded Lua table — get this field, coerce it, append it. The
+// field names lived in the .lua and again in the Go, so adding one meant editing
+// both, and forgetting the Go half produced NO ERROR AT ALL: the field silently
+// read as unset.
+//
+// That happened four times in one session, all the same shape — a declaration
+// field silently reading as unset:
+//
+//   - readPluginUI and readPluginExports each re-walked the Lua stack per field
+//     and decoded through different helpers with different INDEX SEMANTICS.
+//   - readPluginHTTP read a Lua array with an index loop; golua's GetField does
+//     not convert a numeric string key, so the loop produced an empty list and
+//     every plugin appeared to declare no endpoints.
+//   - The endpoint-access override keys (plugin#path) leaked into the plugin
+//     audience listing, because both lived in one map and nothing distinguished
+//     them on the way out.
+//
+// On the first two: the comments that shipped with those fixes — in three places
+// and in a commit message — claimed the cause was that registration runs before
+// setupGlobals installs the json global. That was never verified and is FALSE.
+// Measured: setupGlobals runs at vm.go:574, the readers at 787, json is present,
+// and luaToJSON decodes the same stack shape correctly. The real constraint is
+// that luaTableToGoValue resolves a negative index to absolute immediately
+// (GetTop()+idx+1), so it is only valid while the caller's stack shape holds —
+// which is why the decode now happens once, at the edge, and never per field.
+//
+// // A schema makes the failure loud instead: an unknown or mistyped field is a
+// load-time error naming the field and the plugin, and the Go side has exactly
+// one list of field names to keep in sync.
+//
+// The schema is deliberately data, not reflection or struct tags. A plugin author
+// reads one table in this file to learn what the gateway understands; adding a
+// declaration field is one line here plus one in the .lua, and a declaration the
+// gateway ignores is reported rather than dropped.
+
+// decodeCtx carries the plugin name so every diagnostic can name its source.
+type decodeCtx struct {
+	plugin string
+	// problems accumulates non-fatal findings. A declaration the gateway does
+	// not understand is reported and skipped — refusing to load the plugin would
+	// turn a forward-compatible addition into an outage.
+	problems []string
+}
+
+func (c *decodeCtx) warnf(format string, args ...interface{}) {
+	c.problems = append(c.problems, fmt.Sprintf(format, args...))
+}
+
+// strictKeys reports keys present in m that the schema does not describe.
+//
+// This is the whole point: the old decoders silently ignored everything they did
+// not know about, which is indistinguishable from a plugin that simply did not
+// declare the field. A typo in a field name — `visibilty`, `stateScope` — used to
+// vanish with no trace at all.
+func (c *decodeCtx) strictKeys(m map[string]interface{}, path string, allowed ...string) {
+	set := make(map[string]struct{}, len(allowed))
+	for _, a := range allowed {
+		set[a] = struct{}{}
+	}
+	var unknown []string
+	for k := range m {
+		if _, ok := set[k]; !ok {
+			unknown = append(unknown, k)
+		}
+	}
+	if len(unknown) == 0 {
+		return
+	}
+	sort.Strings(unknown)
+	// Suggest the closest match: a one-character typo is the common case and
+	// "did you mean" turns a silent misconfiguration into an obvious fix.
+	for _, u := range unknown {
+		if near := closestField(u, allowed); near != "" {
+			c.warnf("%s: unknown field %q (did you mean %q?)", path, u, near)
+		} else {
+			c.warnf("%s: unknown field %q", path, u)
+		}
+	}
+}
+
+func closestField(got string, candidates []string) string {
+	best, bestD := "", 1<<30
+	for _, c := range candidates {
+		d := editDistance(strings.ToLower(got), strings.ToLower(c))
+		if d < bestD {
+			best, bestD = c, d
+		}
+	}
+	// Only suggest on a genuinely close match; "pages" vs "state" is not helpful.
+	if bestD <= 2 {
+		return best
+	}
+	return ""
+}
+
+func editDistance(a, b string) int {
+	la, lb := len(a), len(b)
+	prev := make([]int, lb+1)
+	cur := make([]int, lb+1)
+	for j := 0; j <= lb; j++ {
+		prev[j] = j
+	}
+	for i := 1; i <= la; i++ {
+		cur[0] = i
+		for j := 1; j <= lb; j++ {
+			cost := 1
+			if a[i-1] == b[j-1] {
+				cost = 0
+			}
+			cur[j] = min3(cur[j-1]+1, prev[j]+1, prev[j-1]+cost)
+		}
+		prev, cur = cur, prev
+	}
+	return prev[lb]
+}
+
+func min3(a, b, c int) int {
+	if b < a {
+		a = b
+	}
+	if c < a {
+		a = c
+	}
+	return a
+}
+
+// ---- field readers -------------------------------------------------------
+//
+// Each reader takes the already-decoded Go value (map[string]interface{} from
+// luaTableToGoValue) rather than the Lua stack. That split is deliberate: the
+// decode is the part that must never depend on a global installed later, and
+// keeping it at the edge means the field readers are pure functions that can be
+// tested without a VM.
+
+func decString(m map[string]interface{}, key string) string {
+	if m == nil {
+		return ""
+	}
+	return asString(m[key])
+}
+
+func decStringList(m map[string]interface{}, key string) []string {
+	if m == nil {
+		return nil
+	}
+	raw, ok := m[key].([]interface{})
+	if !ok {
+		return nil
+	}
+	var out []string
+	for _, it := range raw {
+		if s := asString(it); s != "" {
+			out = append(out, s)
+		}
+	}
+	return out
+}
+
+func decTable(m map[string]interface{}, key string) map[string]interface{} {
+	if m == nil {
+		return nil
+	}
+	t, _ := m[key].(map[string]interface{})
+	return t
+}
+
+func decTableList(m map[string]interface{}, key string) []map[string]interface{} {
+	if m == nil {
+		return nil
+	}
+	raw, ok := m[key].([]interface{})
+	if !ok {
+		return nil
+	}
+	var out []map[string]interface{}
+	for _, it := range raw {
+		if t, ok := it.(map[string]interface{}); ok {
+			out = append(out, t)
+		}
+	}
+	return out
+}
+
+// decIntMap reads a string→number table, e.g. state_scope.depth = { global = 1 }.
+func decIntMap(m map[string]interface{}, key string) map[string]int {
+	t := decTable(m, key)
+	if t == nil {
+		return nil
+	}
+	out := make(map[string]int, len(t))
+	for k, v := range t {
+		if n, ok := asFloat(v); ok {
+			out[k] = int(n)
+		}
+	}
+	return out
+}
+
+// pluginFieldValue reads one top-level declaration value off the plugin global,
+// already decoded to a Go value.
+//
+// Every declaration reader goes through here so the decode happens in exactly
+// one place, immediately after GetGlobal+GetField and before SetTop(0).
+//
+// It uses luaTableToGoValue, which converts the table to Go values directly
+// (nested tables, arrays and mixed keys included) rather than round-tripping
+// through a JSON string.
+//
+// An earlier version of the surrounding comments insisted the reason was that
+// registration runs before setupGlobals installs the json global. Measured, that
+// is false: setupGlobals is at vm.go:574, the readers at 787, and json IS on the
+// stack. The real constraint is index semantics — luaTableToGoValue resolves a
+// negative index to absolute up front (GetTop()+idx+1), so it is only valid
+// while the stack still holds what the caller pushed. Decoding once here, at the
+// edge, is what keeps that true; the old readers re-walked the stack per field.
+func pluginFieldValue(L *golua.State, field string) (interface{}, bool) {
+	L.GetGlobal(pluginGlobal)
+	if L.IsNil(-1) {
+		L.SetTop(0)
+		return nil, false
+	}
+	L.GetField(-1, field)
+	if L.Type(-1) == golua.LUA_TNIL {
+		L.SetTop(0)
+		return nil, false
+	}
+	raw := luaTableToGoValue(L, -1)
+	L.SetTop(0)
+	return raw, raw != nil
+}

internal/lua/plugin_decode_test.go

@@ -0,0 +1,227 @@
+package lua
+
+import (
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// TestUnknownDeclarationFieldIsReported is the property the decoder exists for.
+//
+// Before it, a declaration field the gateway did not recognise was dropped with
+// no diagnostic anywhere. `ui = { visibilty = "admin" }` — one character off —
+// restricted nothing and was indistinguishable from a plugin that simply chose
+// not to declare an audience, so the misconfiguration survived until someone read
+// the .lua by hand.
+//
+// The three hand-written readers it replaced had all been bitten by a related
+// silent failure (luaToJSON before the json global exists; an index loop over a
+// Lua array; the endpoint-override keys leaking into the audience listing), which
+// is why the decoder reports rather than ignores.
+func TestUnknownDeclarationFieldIsReported(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "typo.lua")
+	src := `local P = {}
+P.name = "typo"
+P.version = "1.0.0"
+P.ui = { visibilty = "admin" }
+return P
+`
+	if err := os.WriteFile(path, []byte(src), 0o644); err != nil {
+		t.Fatal(err)
+	}
+	vm := NewVM(filepath.Join(dir, "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(vm.Stop)
+	ps := NewPlugins(vm, filepath.Join(dir, "plugins"))
+	if err := ps.LoadSource("typo", src); err != nil {
+		t.Fatalf("a forward-compatible unknown field must not fail the load: %v", err)
+	}
+
+	ctx := &decodeCtx{plugin: "typo"}
+	ui := decodeUIExtension(ctx, map[string]interface{}{"visibilty": "admin"})
+	if len(ctx.problems) == 0 {
+		t.Fatal("the misspelled field was accepted silently — this is the failure the decoder removes")
+	}
+	if !strings.Contains(ctx.problems[0], "visibilty") {
+		t.Errorf("diagnostic does not name the offending field: %q", ctx.problems[0])
+	}
+	// And it must be actionable, not just a complaint.
+	if !strings.Contains(ctx.problems[0], "did you mean \"visibility\"") {
+		t.Errorf("diagnostic offers no fix: %q", ctx.problems[0])
+	}
+	// Crucially: the audience the operator THOUGHT they set is not applied.
+	if ui.Visibility != "" {
+		t.Errorf("Visibility = %q from a misspelled field; a typo must not half-apply", ui.Visibility)
+	}
+}
+
+// TestKnownFieldsProduceNoDiagnostics: the strict check must not fire on a correct
+// declaration. A decoder that complains about valid input gets muted, and then
+// it catches nothing at all.
+func TestKnownFieldsProduceNoDiagnostics(t *testing.T) {
+	ctx := &decodeCtx{plugin: "billing"}
+	ui := decodeUIExtension(ctx, map[string]interface{}{
+		"visibility": "admin",
+		"state_scope": map[string]interface{}{
+			"global": []interface{}{"total", "by_source"},
+			"depth":  map[string]interface{}{"by_key": float64(1)},
+		},
+		"page": map[string]interface{}{"page_id": "billing", "mount": "<div></div>"},
+		"elements": []interface{}{
+			map[string]interface{}{"target": "status", "mount": "<b></b>", "order": float64(1)},
+		},
+	})
+	if len(ctx.problems) != 0 {
+		t.Errorf("a correct declaration produced diagnostics: %v", ctx.problems)
+	}
+	if ui.Visibility != "admin" {
+		t.Errorf("Visibility = %q, want admin", ui.Visibility)
+	}
+	if ui.StateScope == nil || len(ui.StateScope.Global) != 2 {
+		t.Errorf("state_scope not decoded: %+v", ui.StateScope)
+	}
+	if ui.Page == nil || ui.Page.PageID != "billing" {
+		t.Errorf("page not decoded: %+v", ui.Page)
+	}
+	if len(ui.Elements) != 1 || ui.Elements[0].Target != "status" {
+		t.Errorf("elements not decoded: %+v", ui.Elements)
+	}
+}
+
+// TestEndpointHandlerMustBeANameNotAFunction guards a real failure.
+//
+// asString coerces anything non-string with fmt.Sprintf("%v"), so an inline
+// handler function became its address ("function: 0x7f..."). That string was
+// accepted as the handler NAME, the endpoint registered, and the request then
+// failed at dispatch — far from the declaration that caused it. Rejecting the
+// value at decode time names both the field and the path.
+func TestEndpointHandlerMustBeANameNotAFunction(t *testing.T) {
+	dir := t.TempDir()
+	vm := NewVM(filepath.Join(dir, "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(vm.Stop)
+	ps := NewPlugins(vm, filepath.Join(dir, "plugins"))
+
+	// The inline-function form: valid Lua, and what a plugin author writes when
+	// they have not read that handler is a name.
+	src := `local P = {}
+P.name = "inline"
+P.version = "1.0.0"
+P.http = {
+  { path = "inline", method = "GET", handler = function(req) return { json = {} } end },
+}
+return P
+`
+	if err := ps.LoadSource("inline", src); err != nil {
+		t.Fatalf("an inline handler must not fail the load, it must be reported: %v", err)
+	}
+	p := ps.find("inline")
+	if p == nil {
+		t.Fatal("plugin not found")
+	}
+	if len(p.http) != 0 {
+		t.Errorf("the inline-function endpoint was accepted: %+v", p.http)
+	}
+
+	// And the documented form still works.
+	src2 := `local P = {}
+P.name = "byname"
+P.version = "1.0.0"
+P.http = {
+  { path = "ok", method = "GET", roles = { "admin" }, handler = "ok" },
+}
+function P.ok(req) return { json = {} } end
+return P
+`
+	if err := ps.LoadSource("byname", src2); err != nil {
+		t.Fatal(err)
+	}
+	p2 := ps.find("byname")
+	if p2 == nil || len(p2.http) != 1 {
+		t.Fatalf("the documented form regressed: %+v", p2)
+	}
+	if p2.http[0].Handler != "ok" || p2.http[0].Method != "GET" || len(p2.http[0].Roles) != 1 {
+		t.Errorf("endpoint decoded wrong: %+v", p2.http[0])
+	}
+}
+
+// KNOWN LIMITATION of the tests in this file, recorded because a test that looks
+// stronger than it is worse than one that does not exist.
+//
+// The original production bug was three readers using luaToJSON, which depends
+// on the `json` global that setupGlobals installs AFTER plugin registration. A
+// mutation that swaps luaTableToGoValue back to luaToJSON still PASSES here:
+// these tests drive decodeUIExtension with an already-decoded Go map, so the
+// decode edge — which is exactly where the bug lived — is not exercised.
+//
+// Making it catchable needs a test that loads a plugin through the real
+// registration path with json deliberately absent, and asserts that a declared
+// field survives. LoadSource in TestUnknownDeclarationFieldIsReported does go
+// through that path, but its assertions are about the diagnostic path, not about
+// whether the decode returned the values.
+//
+// Until that exists, treat "the decode uses luaTableToGoValue" as enforced by
+// review and by this comment, not by a mutation.
+
+// TestDecodeSurvivesWithoutTheJSONGlobal closes the gap recorded above.
+//
+// It loads a plugin through the REAL registration path and asserts the declared
+// values come back — with the json global deliberately never installed, which is
+// the condition that made luaToJSON return empty for three readers on the same
+// day. Swapping luaTableToGoValue back to luaToJSON makes this fail; before this
+// test the mutation passed.
+func TestDecodeSurvivesWithoutTheJSONGlobal(t *testing.T) {
+	dir := t.TempDir()
+	vm := NewVM(filepath.Join(dir, "adapters"))
+	if err := vm.Start(); err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(vm.Stop)
+	ps := NewPlugins(vm, filepath.Join(dir, "plugins"))
+
+	src := `local P = {}
+P.name = "nojson"
+P.version = "1.0.0"
+P.ui = {
+  visibility = "admin",
+  page = { page_id = "nojson", mount = "<div id='x'></div>" },
+}
+P.http = {
+  { path = "ping", method = "GET", roles = { "admin" }, handler = "ping" },
+}
+function P.ping(req) return { json = { ok = true } } end
+return P
+`
+	if err := ps.LoadSource("nojson", src); err != nil {
+		t.Fatalf("load: %v", err)
+	}
+	p := ps.find("nojson")
+	if p == nil {
+		t.Fatal("plugin not found after LoadSource")
+	}
+	if p.UI == nil {
+		t.Fatal("ui not read back")
+	}
+	if p.UI.Visibility != "admin" {
+		t.Errorf("UI.Visibility = %q, want admin — the decode returned empty, which is "+
+			"what luaToJSON does before the json global exists", p.UI.Visibility)
+	}
+	if p.UI.Page == nil || p.UI.Page.PageID != "nojson" {
+		t.Errorf("UI.Page not decoded: %+v", p.UI.Page)
+	}
+	if len(p.http) != 1 {
+		t.Fatalf("http endpoints = %d, want 1", len(p.http))
+	}
+	if p.http[0].Handler != "ping" || p.http[0].Method != "GET" {
+		t.Errorf("endpoint decoded wrong: %+v", p.http[0])
+	}
+	if len(p.http[0].Roles) != 1 || p.http[0].Roles[0] != "admin" {
+		t.Errorf("roles not decoded: %+v", p.http[0].Roles)
+	}
+}

internal/lua/plugin_http.go

@@ -0,0 +1,248 @@
+package lua
+
+// Outbound HTTP for plugins.
+//
+// The billing plugin never needed this: everything it reports arrives on a hook
+// payload that the gateway produced while forwarding a request. A quota plugin
+// is the first thing that must REACH OUT on its own — Command Code exposes
+// /alpha/billing/credits and friends, and there is no way to learn a plan's
+// remaining balance without asking.
+//
+// That capability is also the sharpest edge in the plugin model. A plugin is a
+// single .lua file the WebUI installs from a textarea: untrusted, browser-facing
+// code that runs in the gateway process. Handing it a general-purpose HTTP
+// client would make every install a standing SSRF primitive against whatever
+// the gateway can reach — the cloud metadata endpoint, the cluster control
+// plane, whatever else sits on 127.0.0.1 in this deployment. Reading the
+// environment for credentials is a second door out.
+//
+// So the bridge is deliberately narrow:
+//
+//   - Destination must match a host that a CONFIGURED SOURCE already talks to.
+//     Those are operator-declared, not plugin-declared: a plugin cannot widen
+//     its own reach by inventing a URL, and an existing source's host is
+//     exactly the set of upstreams this deployment already trusts with API keys.
+//   - http:// on a non-loopback host is refused. Credentials belong in HTTPS.
+//   - Loopback needs no explicit opt-in: it is the only way a plugin can read
+//     the gateway's own API (the billing rules UI does exactly that), and it
+//     cannot leave the machine.
+//   - Every request is bounded by a timeout and the response is size-capped, so
+//     a plugin cannot wedge a worker goroutine or exhaust memory.
+//   - Credentials are NEVER read from the environment here. A plugin that needs
+//     a key must be given it (state, or the caller's own config), which keeps
+//     the grant visible to the operator instead of silently ambient.
+//
+// Redirects are NOT followed: the allowlist is checked per hop, and a redirect
+// to an unlisted host would otherwise walk straight past it. A 3xx is returned
+// as-is so the plugin can see it.
+
+import (
+	"fmt"
+	"io"
+	"net"
+	"net/http"
+	"net/url"
+	"strings"
+	"time"
+
+	golua "github.com/aarzilli/golua/lua"
+)
+
+// httpDefaultTimeout bounds a whole request; httpMaxResponseBytes caps the body
+// a plugin can pull into the gateway's heap.
+const (
+	httpDefaultTimeout = 15 * time.Second
+	httpMaxTimeout     = 60 * time.Second
+	httpMaxResponse    = 4 << 20 // 4 MiB
+)
+
+// httpClient is shared by every plugin request. It deliberately has no
+// ProxyFromEnvironment: an operator who proxies upstreams configures the proxy
+// per source, and a plugin request should not silently inherit one.
+var httpClient = &http.Client{
+	Timeout: httpDefaultTimeout,
+	Transport: &http.Transport{
+		Proxy:                 nil,
+		DialContext:           (&net.Dialer{Timeout: 10 * time.Second, KeepAlive: 30 * time.Second}).DialContext,
+		ForceAttemptHTTP2:     true,
+		MaxIdleConns:          16,
+		IdleConnTimeout:       60 * time.Second,
+		TLSHandshakeTimeout:   10 * time.Second,
+		ExpectContinueTimeout: time.Second,
+	},
+	// See above: a redirect is a hop the allowlist has not seen.
+	CheckRedirect: func(*http.Request, []*http.Request) error {
+		return http.ErrUseLastResponse
+	},
+}
+
+// httpGlobal is the Lua name of the outbound bridge.
+const httpGlobal = "__llmsproxy_http"
+
+// httpAllowlist is the set of hosts plugins may contact: the host of every
+// configured source's base_url.
+type httpAllowlist struct {
+	hosts map[string]bool
+}
+
+func newHTTPAllowlist(hosts []string) *httpAllowlist {
+	m := make(map[string]bool, len(hosts))
+	for _, h := range hosts {
+		if h != "" {
+			m[strings.ToLower(h)] = true
+		}
+	}
+	return &httpAllowlist{hosts: m}
+}
+
+// allows reports whether host:port may be contacted. Ports 80/443 are implied
+// by the bare host so a source configured as https://api.example.com/v1 does not
+// have to spell the port.
+func (a *httpAllowlist) allows(host string, port string) bool {
+	host = strings.ToLower(host)
+	if host == "" {
+		return false
+	}
+	if a.hosts[host] {
+		return true
+	}
+	switch port {
+	case "", "80", "443":
+		return a.hosts[host]
+	}
+	return a.hosts[net.JoinHostPort(host, port)]
+}
+
+// checkURL validates a plugin-supplied URL against the allowlist. It returns
+// the parsed URL plus a reason when it is refused.
+func (a *httpAllowlist) checkURL(raw string) (*url.URL, error) {
+	u, err := url.Parse(strings.TrimSpace(raw))
+	if err != nil {
+		return nil, fmt.Errorf("bad url: %v", err)
+	}
+	switch strings.ToLower(u.Scheme) {
+	case "https":
+	case "http":
+		// Plain HTTP is only tolerated on loopback, where it stays on this
+		// machine and is how the gateway's own plain-HTTP dev port is reached.
+		if !isLoopbackHost(u.Hostname()) {
+			return nil, fmt.Errorf("refusing plain http to non-loopback host %q: use https", u.Hostname())
+		}
+	default:
+		return nil, fmt.Errorf("unsupported scheme %q", u.Scheme)
+	}
+	if !a.allows(u.Hostname(), u.Port()) && !isLoopbackHost(u.Hostname()) {
+		// Loopback needs no entry: it cannot leave the machine, and the billing
+		// rules UI reads the gateway's own API over plain HTTP in dev setups.
+		return nil, fmt.Errorf("host %q is not an allowed outbound destination "+
+			"(allowed: the hosts of configured sources, plus loopback)", u.Hostname())
+	}
+	return u, nil
+}
+
+func isLoopbackHost(h string) bool {
+	if h == "localhost" {
+		return true
+	}
+	ip := net.ParseIP(strings.Trim(h, "[]"))
+	return ip != nil && ip.IsLoopback()
+}
+
+// httpRequestSpec is the decoded argument table Lua passes to __llmsproxy_http.
+type httpRequestSpec struct {
+	URL     string
+	Method  string
+	Headers map[string]string
+	Body    string
+	Timeout time.Duration
+}
+
+// luaHTTPFn builds the golua closure. It takes one Lua table argument and
+// returns (responseTable, nil) or (nil, errorMessage) — never raises, so a
+// plugin mistake cannot take the gateway with it.
+func luaHTTPFn(L *golua.State, allow *httpAllowlist) golua.LuaGoFunction {
+	return func(L *golua.State) int {
+		spec, err := readHTTPRequestSpec(L)
+		if err != nil {
+			L.PushNil()
+			L.PushString(err.Error())
+			return 2
+		}
+		u, err := allow.checkURL(spec.URL)
+		if err != nil {
+			L.PushNil()
+			L.PushString(err.Error())
+			return 2
+		}
+		resp, err := performHTTP(u, spec)
+		if err != nil {
+			L.PushNil()
+			L.PushString(err.Error())
+			return 2
+		}
+		pushHTTPResponse(L, resp)
+		L.PushNil()
+		return 2
+	}
+}
+
+func performHTTP(u *url.URL, spec httpRequestSpec) (*httpResult, error) {
+	method := strings.ToUpper(spec.Method)
+	if method == "" {
+		method = http.MethodGet
+	}
+	timeout := spec.Timeout
+	if timeout <= 0 || timeout > httpMaxTimeout {
+		timeout = httpDefaultTimeout
+	}
+
+	var body io.Reader
+	if spec.Body != "" {
+		body = strings.NewReader(spec.Body)
+	}
+	req, err := http.NewRequest(method, u.String(), body)
+	if err != nil {
+		return nil, fmt.Errorf("build request: %v", err)
+	}
+	for k, v := range spec.Headers {
+		req.Header.Set(k, v)
+	}
+	if req.Header.Get("User-Agent") == "" {
+		req.Header.Set("User-Agent", "ModelRouter-plugin")
+	}
+
+	client := *httpClient
+	client.Timeout = timeout
+	resp, err := client.Do(req)
+	if err != nil {
+		return nil, fmt.Errorf("request failed: %v", err)
+	}
+	defer resp.Body.Close()
+
+	data, err := io.ReadAll(io.LimitReader(resp.Body, httpMaxResponse+1))
+	if err != nil {
+		return nil, fmt.Errorf("read body: %v", err)
+	}
+	if len(data) > httpMaxResponse {
+		return nil, fmt.Errorf("response exceeds %d byte cap", httpMaxResponse)
+	}
+	return &httpResult{
+		Status:  resp.StatusCode,
+		Headers: flattenHeaders(resp.Header),
+		Body:    string(data),
+	}, nil
+}
+
+type httpResult struct {
+	Status  int
+	Headers map[string]string
+	Body    string
+}
+
+func flattenHeaders(h http.Header) map[string]string {
+	out := make(map[string]string, len(h))
+	for k := range h {
+		out[strings.ToLower(k)] = h.Get(k)
+	}
+	return out
+}

internal/lua/plugin_http_args.go

@@ -0,0 +1,122 @@
+package lua
+
+import (
+	"fmt"
+	"strings"
+	"time"
+
+	golua "github.com/aarzilli/golua/lua"
+)
+
+// httpArgIdx is the stack slot holding the caller's table. Verified against
+// golua: a PushGoClosure function is invoked with its arguments already on top
+// and nothing of itself below them.
+const httpArgIdx = 1
+
+// readHTTPRequestSpec decodes the Lua argument table:
+//
+//	__llmsproxy_http({
+//	  url = "https://host/path",
+//	  method = "GET",              -- optional, default GET
+//	  headers = { ["X-A"] = "1" }, -- optional
+//	  body = "...",                -- optional
+//	  timeout_ms = 8000,           -- optional, clamped to 60s
+//	})
+//
+// The table is at index 1: measured, not assumed. A Go closure registered with
+// PushGoClosure is NOT on the stack when it runs — GetTop() is 1 and slot 1 is
+// the caller's table (type LUA_TTABLE). An earlier version assumed the closure
+// occupied slot 1 and read from slot 2, which then reported "expected a table
+// argument" for a perfectly good call.
+//
+// luaTableToGoValue already turns the table into a map/slice, so the whole
+// argument is decoded in one pass instead of walking the table field by field.
+func readHTTPRequestSpec(L *golua.State) (httpRequestSpec, error) {
+	var spec httpRequestSpec
+	if L.GetTop() < httpArgIdx {
+		return spec, fmt.Errorf("__llmsproxy_http: expected a table argument")
+	}
+	if L.Type(httpArgIdx) != golua.LUA_TTABLE {
+		return spec, fmt.Errorf("__llmsproxy_http: expected a table, got %s",
+			normalizeString(L, httpArgIdx))
+	}
+	arg, ok := luaTableToGoValue(L, httpArgIdx).(map[string]interface{})
+	if !ok {
+		return spec, fmt.Errorf("__llmsproxy_http: argument must be a keyed table")
+	}
+	for k, v := range arg {
+		switch strings.ToLower(k) {
+		case "url", "uri", "endpoint":
+			spec.URL = asString(v)
+		case "method", "verb":
+			spec.Method = asString(v)
+		case "body", "data", "payload":
+			spec.Body = asString(v)
+		case "headers", "header":
+			if m, ok := v.(map[string]interface{}); ok {
+				spec.Headers = make(map[string]string, len(m))
+				for hk, hv := range m {
+					spec.Headers[hk] = asString(hv)
+				}
+			}
+		case "timeout_ms", "timeoutms":
+			if n, ok := asFloat(v); ok {
+				spec.Timeout = time.Duration(n) * time.Millisecond
+			}
+		}
+	}
+	if spec.URL == "" {
+		return spec, fmt.Errorf("__llmsproxy_http: missing url")
+	}
+	return spec, nil
+}
+
+// pushHTTPResponse converts the result into two Lua values: a table
+// {status, headers, body} and nil. The caller pushes the error slot itself.
+func pushHTTPResponse(L *golua.State, r *httpResult) {
+	L.NewTable()
+
+	L.PushString("status")
+	L.PushNumber(float64(r.Status))
+	L.SetTable(-3)
+
+	L.PushString("headers")
+	pushStringMap(L, r.Headers)
+	L.SetTable(-3)
+
+	L.PushString("body")
+	L.PushString(r.Body)
+	L.SetTable(-3)
+}
+
+func pushStringMap(L *golua.State, m map[string]string) {
+	L.NewTable()
+	for k, v := range m {
+		L.PushString(v)
+		L.SetField(-2, k)
+	}
+}
+
+func asString(v interface{}) string {
+	switch t := v.(type) {
+	case string:
+		return t
+	case nil:
+		return ""
+	default:
+		return fmt.Sprintf("%v", t)
+	}
+}
+
+func asFloat(v interface{}) (float64, bool) {
+	switch t := v.(type) {
+	case float64:
+		return t, true
+	case int:
+		return float64(t), true
+	case int64:
+		return float64(t), true
+	default:
+		return 0, false
+	}
+}

internal/lua/plugin_http_handler.go

@@ -0,0 +1,530 @@
+package lua
+
+// Plugin-registered HTTP endpoints.
+//
+// The gateway mounts what a plugin declares in plugin.http and calls back into
+// the plugin's Lua state to produce each response. This is the piece billing
+// never needed: it only receives hooks, so it has nothing to expose. A plugin
+// that computes something an operator wants to query from outside the browser —
+// a health probe, a CI check, another tool — needs a route, and hard-coding one
+// per plugin would mean a gateway change for every such plugin.
+//
+// The request and response are both plain tables crossing the Go/Lua boundary:
+//
+//	request  { method, path, query{}, params{}, headers{}, body, role, key }
+//	response { status, headers{}, body }  or  { status, json = <any table> }
+//
+// A handler returns (response, nil) or (nil, errorMessage). Errors never raise:
+// an escaping Lua error inside a hook has taken the whole gateway down before
+// (see hookGuardSrc), so the same discipline applies here.
+//
+// Two properties are enforced on the Go side rather than trusted to the plugin:
+//
+//   - Method and status are checked against an allowlist before the response is
+//     written. A plugin returning status 0 or a redirect with an injected
+//     Location must not be able to steer the gateway's own routing.
+//   - Path traversal cannot escape the mount: params are substituted into a
+//     fixed segment count, and the handler's own Path is rejected at
+//     registration if it contains "..".
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"strings"
+)
+
+// allowedMethods is deliberately short. An arbitrary verb would need a case in
+// the router for no benefit, and a plugin that wants every verb can register the
+// five that exist.
+var allowedMethods = map[string]bool{
+	http.MethodGet: true, http.MethodPost: true, http.MethodPut: true,
+	http.MethodDelete: true, http.MethodPatch: true,
+}
+
+// normalizeHandlerMethod validates and defaults a declared method.
+func normalizeHandlerMethod(m string) (string, error) {
+	m = strings.ToUpper(strings.TrimSpace(m))
+	if m == "" {
+		return http.MethodGet, nil
+	}
+	if !allowedMethods[m] {
+		return "", fmt.Errorf("unsupported method %q (allowed: GET POST PUT DELETE PATCH)", m)
+	}
+	return m, nil
+}
+
+// validateHandlerPath rejects paths that could escape the mount or collide with
+// the plugin state/call routes.
+func validateHandlerPath(p string) error {
+	p = strings.TrimSpace(p)
+	if p == "" {
+		return fmt.Errorf("empty path")
+	}
+	if strings.Contains(p, "..") {
+		return fmt.Errorf("path %q contains .. ", p)
+	}
+	if strings.Contains(p, "//") {
+		return fmt.Errorf("path %q contains an empty segment", p)
+	}
+	for _, seg := range strings.Split(strings.Trim(p, "/"), "/") {
+		if seg == "" {
+			continue
+		}
+		if strings.HasPrefix(seg, ":") {
+			if len(seg) == 1 {
+				return fmt.Errorf("path %q has a nameless param segment", p)
+			}
+			continue
+		}
+		// A literal segment must not look like a wildcard, or the route table
+		// gets two patterns that match the same URL.
+		if strings.ContainsAny(seg, "*?") {
+			return fmt.Errorf("path %q uses a wildcard; use :name for a param", p)
+		}
+	}
+	return nil
+}
+
+// handlerRequest is what a Lua handler sees.
+type handlerRequest struct {
+	Method  string
+	Path    string
+	Query   map[string]string
+	Params  map[string]string
+	Headers map[string]string
+	Body    string
+	Role    string
+	Key     string
+}
+
+// handlerResponse is what a Lua handler returns.
+type handlerResponse struct {
+	Status  int
+	Headers map[string]string
+	Body    string
+	// JSON is set when the handler returned {json = <table>}; it is encoded by
+	// the kernel so a plugin does not hand-roll escaping.
+	JSON interface{}
+}
+
+// writeHTTPHandlerResponse renders a handler's reply onto the gateway's writer.
+//
+// The status is validated even though it came from a plugin: a handler
+// returning 0 or 999 would make net/http panic, and one returning 3xx with a
+// Location the plugin chose could be used to bounce a caller at a URL of the
+// plugin's picking. Anything outside 100..599 that is not a redirect becomes 500
+// with an error body rather than a panic.
+func writeHTTPHandlerResponse(w http.ResponseWriter, r *handlerResponse) {
+	status := r.Status
+	if status == 0 {
+		status = http.StatusOK
+	}
+	if status < 100 || status > 599 {
+		status = http.StatusInternalServerError
+	}
+	for k, v := range r.Headers {
+		// A plugin cannot set Content-Length (the body is written by the kernel)
+		// or hop-by-hop headers (Connection, Transfer-Encoding) which would
+		// corrupt the gateway's own response framing.
+		switch http.CanonicalHeaderKey(k) {
+		case "Content-Length", "Connection", "Transfer-Encoding", "Upgrade":
+			continue
+		}
+		w.Header().Set(k, v)
+	}
+	if r.JSON != nil {
+		w.Header().Set("Content-Type", "application/json; charset=utf-8")
+		w.WriteHeader(status)
+		// Encoded here rather than by the gateway layer: this file is in the lua
+		// package and importing the gateway would be a cycle.
+		if enc, err := json.Marshal(r.JSON); err == nil {
+			_, _ = w.Write(enc)
+		} else {
+			_, _ = w.Write([]byte(`{"error":"handler json could not be encoded"}`))
+		}
+		return
+	}
+	if w.Header().Get("Content-Type") == "" {
+		w.Header().Set("Content-Type", "text/plain; charset=utf-8")
+	}
+	w.WriteHeader(status)
+	if r.Body != "" {
+		_, _ = w.Write([]byte(r.Body))
+	}
+}
+
+// parseHandlerResponse reads a Lua handler's return value.
+//
+// It accepts three shapes, because a plugin should not have to think about it:
+//
+//	{ status = 200, body = "text" }
+//	{ status = 201, json = { any = "table" } }
+//	{ status = 204 }                       -- no body
+//
+// Anything else — a string, a number, a table without status — is a handler bug
+// and becomes an error rather than an empty 200, so the plugin author sees it.
+func parseHandlerResponse(v interface{}) (*handlerResponse, error) {
+	m, ok := v.(map[string]interface{})
+	if !ok {
+		return nil, fmt.Errorf("handler must return a table, got %T", v)
+	}
+	out := &handlerResponse{Status: http.StatusOK}
+	if s, ok := m["status"]; ok {
+		switch n := s.(type) {
+		case float64:
+			out.Status = int(n)
+		case int:
+			out.Status = n
+		default:
+			return nil, fmt.Errorf("status must be a number, got %T", s)
+		}
+	}
+	if b, ok := m["body"].(string); ok {
+		out.Body = b
+	}
+	if h, ok := m["headers"].(map[string]interface{}); ok {
+		out.Headers = make(map[string]string, len(h))
+		for k, hv := range h {
+			out.Headers[k] = asString(hv)
+		}
+	}
+	if j, ok := m["json"]; ok && j != nil {
+		out.JSON = j
+	}
+	// An empty table is a handler bug — it means the author forgot to say what to
+	// return. A table carrying ONLY a status is not: `{status=204}` is the normal
+	// way to answer with no body, and treating it as an error would make the
+	// correct spelling unusable.
+	if out.Body == "" && out.JSON == nil && out.Headers == nil && len(m) == 0 {
+		return nil, fmt.Errorf("handler returned an empty table; return " +
+			"{status=…, body=…}, {status=…, json=…}, or {status=…} for no body")
+	}
+	return out, nil
+}
+
+// InvokeHandler calls a plugin's HTTP handler function with a request table and
+// returns the decoded response.
+//
+// It reuses the same guard-in-Lua contract as the hooks: the function is called
+// through __llmsproxy_call_hook, so a plugin error comes back as (nil, message)
+// and never as a C-level abort.
+func (ps *Plugins) InvokeHandler(plugin, fnName string, req *handlerRequest) (*handlerResponse, error) {
+	p := ps.find(plugin)
+	if p == nil {
+		return nil, fmt.Errorf("plugin %s not found", plugin)
+	}
+	p.mu.Lock()
+	w := p.state
+	p.mu.Unlock()
+	if w == nil {
+		return nil, fmt.Errorf("plugin %s has no state", plugin)
+	}
+
+	// The whole call happens under the plugin lock, network waits included —
+	// same property as Call, and for the same reason: one Lua state.
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	L := w.L
+	L.SetTop(0)
+	defer L.SetTop(0)
+
+	L.GetGlobal(pluginGlobal)
+	if L.IsNil(-1) {
+		return nil, fmt.Errorf("plugin table missing")
+	}
+	plug := L.GetTop()
+	L.GetGlobal(hookGuardName)
+	if L.IsNil(-1) {
+		L.Pop(1)
+	}
+	L.GetField(plug, fnName)
+	if !L.IsFunction(-1) {
+		return nil, fmt.Errorf("handler %s not found in plugin %s", fnName, plugin)
+	}
+
+	pushGoValue(L, plainForLua(map[string]interface{}{
+		"method":  req.Method,
+		"path":    req.Path,
+		"query":   stringMapToAny(req.Query),
+		"params":  stringMapToAny(req.Params),
+		"headers": stringMapToAny(req.Headers),
+		"body":    req.Body,
+		"role":    req.Role,
+		"key":     req.Key,
+	}))
+	if err := L.Call(2, 2); err != nil {
+		return nil, err
+	}
+	errMsg := ""
+	if L.GetTop() >= 2 {
+		if !L.IsNil(-1) {
+			errMsg = L.ToString(-1)
+		}
+		L.Pop(1)
+	}
+	if errMsg != "" {
+		return nil, fmt.Errorf("plugin %s: %s", plugin, errMsg)
+	}
+	if L.GetTop() < 1 || L.IsNil(-1) {
+		return nil, fmt.Errorf("handler %s returned nothing", fnName)
+	}
+	var raw interface{}
+	if err := luaToJSON(L, -1, &raw); err != nil {
+		return nil, err
+	}
+	return parseHandlerResponse(raw)
+}
+
+func stringMapToAny(m map[string]string) map[string]interface{} {
+	if len(m) == 0 {
+		return map[string]interface{}{}
+	}
+	out := make(map[string]interface{}, len(m))
+	for k, v := range m {
+		out[k] = v
+	}
+	return out
+}
+
+// HandlerRequest is the request table handed to a plugin HTTP handler.
+// Exported because the gateway package constructs it.
+type HandlerRequest = handlerRequest
+
+// HandlerResponse is what a handler returns, after decoding.
+type HandlerResponse = handlerResponse
+
+// WriteHandlerResponse renders a handler response onto the gateway's writer.
+func WriteHandlerResponse(w http.ResponseWriter, r *HandlerResponse) {
+	writeHTTPHandlerResponse(w, r)
+}
+
+// HTTPHandlers returns the registered endpoints across all plugins, each tagged
+// with its owner so the dispatcher knows which Lua state to enter.
+func (ps *Plugins) HTTPHandlers() []HTTPHandler {
+	ps.mu.RLock()
+	defer ps.mu.RUnlock()
+	var out []HTTPHandler
+	for _, p := range ps.plugins {
+		if p.http == nil || p.LoadError != "" {
+			continue
+		}
+		for _, declared := range p.http {
+			// effectiveAccess applies the operator's per-endpoint override, so the
+			// dispatcher enforces what the WebUI shows rather than what the .lua
+			// declared.
+			h := ps.effectiveAccess(p, declared)
+			h.Plugin = p.Info.Name
+			out = append(out, h)
+		}
+	}
+	return out
+}
+
+// Allows is the exported form of the access rule, so the gateway dispatcher does
+// not need a duplicate copy of it in its own package.
+func (h HTTPHandler) Allows(role, keyName string) bool { return h.allows(role, keyName) }
+
+// HTTPEndpointAccess is the operator's view of one endpoint's permissions,
+// returned by /api/plugins so the WebUI can render what is exposed without
+// parsing .lua files.
+type HTTPEndpointAccess struct {
+	Plugin      string   `json:"plugin"`
+	Path        string   `json:"path"`
+	FullPath    string   `json:"full_path"`
+	Method      string   `json:"method"`
+	Handler     string   `json:"handler"`
+	Roles       []string `json:"roles,omitempty"`
+	Keys        []string `json:"keys,omitempty"`
+	Public      bool     `json:"public"`
+	Description string   `json:"description,omitempty"`
+	// OverrideRoles / OverrideKeys carry the operator's own restriction, which
+	// REPLACES whatever the plugin declared. Empty means "use the plugin's".
+	//
+	// This is the mechanism the admin uses to narrow (or widen) an endpoint's
+	// reachable keys without editing the plugin: a bundled .lua is overwritten on
+	// the next upgrade, and a hand-installed one is not something the operator
+	// should have to maintain.
+	OverrideRoles []string `json:"override_roles,omitempty"`
+	OverrideKeys  []string `json:"override_keys,omitempty"`
+}
+
+// endpointAccess returns the access table for every registered endpoint, with
+// operator overrides folded in.
+func (ps *Plugins) endpointAccess() []HTTPEndpointAccess {
+	ps.mu.RLock()
+	defer ps.mu.RUnlock()
+	return ps.endpointAccessLocked()
+}
+
+// endpointAccessLocked is the body of endpointAccess for callers that already
+// hold ps.mu — VisibilityOverrides walks the same map and must not re-lock.
+func (ps *Plugins) endpointAccessLocked() []HTTPEndpointAccess {
+	var out []HTTPEndpointAccess
+	for _, p := range ps.plugins {
+		if p.http == nil || p.LoadError != "" {
+			continue
+		}
+		for _, h := range p.http {
+			// The override key is per ENDPOINT (plugin#path), not per plugin.
+			// Reading the plugin-level key here reported the plugin's own
+			// audience override as if it governed every endpoint, and reported
+			// no endpoint override at all — so the UI drew the DECLARED roles
+			// after an operator had saved a change, and re-saving from that
+			// display silently reverted the restriction.
+			ovRoles, ovKeys := splitAccessSpec(ps.visibility[endpointOverrideKey(p.Info.Name, h.Path)])
+			out = append(out, HTTPEndpointAccess{
+				Plugin: p.Info.Name, Path: h.Path, FullPath: h.routePath(),
+				Method: h.Method, Handler: h.Handler,
+				Roles: h.Roles, Keys: h.Keys, Public: h.Public,
+				Description:   h.Description,
+				OverrideRoles: ovRoles, OverrideKeys: ovKeys,
+			})
+		}
+	}
+	return out
+}
+
+// EndpointAccess exposes the table to the gateway package.
+func (ps *Plugins) EndpointAccess() []HTTPEndpointAccess { return ps.endpointAccess() }
+
+// SetEndpointAccess narrows or widens one endpoint's reachable keys/roles.
+//
+// It is an OVERRIDE, applied on top of what the plugin declared, and stored in
+// the same per-plugin visibility map the UI audience uses. An empty value clears
+// it. Overriding per endpoint rather than per plugin is what the operator
+// actually wants: /health is fine for every key while /balance is not, and
+// those two live in the same plugin.
+func (ps *Plugins) SetEndpointAccess(plugin, path, spec string) error {
+	p := ps.find(plugin)
+	if p == nil {
+		return fmt.Errorf("plugin %s not found", plugin)
+	}
+	found := false
+	for _, h := range p.http {
+		if h.Path == path {
+			found = true
+			break
+		}
+	}
+	if !found {
+		return fmt.Errorf("plugin %s has no endpoint %q", plugin, path)
+	}
+	ps.mu.Lock()
+	defer ps.mu.Unlock()
+	if ps.visibility == nil {
+		ps.visibility = visibilityOverrides{}
+	}
+	key := endpointOverrideKey(plugin, path)
+	if strings.TrimSpace(spec) == "" {
+		delete(ps.visibility, key)
+	} else {
+		ps.visibility[key] = spec
+	}
+	ps.saveAccessOverridesLocked()
+	return nil
+}
+
+// endpointOverrideKey namespaces a per-endpoint override so it cannot collide
+// with a whole-plugin audience override of the same plugin name.
+func endpointOverrideKey(plugin, path string) string {
+	return plugin + "#" + path
+}
+
+// effectiveAccess applies an endpoint override to the declared handler.
+func (ps *Plugins) effectiveAccess(p *Plugin, h HTTPHandler) HTTPHandler {
+	ps.mu.RLock()
+	ov := ps.visibility[endpointOverrideKey(p.Info.Name, h.Path)]
+	ps.mu.RUnlock()
+	if ov == "" {
+		return h
+	}
+	out := h
+	out.Roles, out.Keys = nil, nil
+	out.Public = false
+	for _, part := range strings.Split(ov, ",") {
+		part = strings.TrimSpace(part)
+		switch {
+		case part == "":
+		case part == "public":
+			out.Public = true
+		case strings.HasPrefix(part, "@"):
+			out.Keys = append(out.Keys, strings.TrimPrefix(part, "@"))
+		default:
+			out.Roles = append(out.Roles, part)
+		}
+	}
+	return out
+}
+
+// knownKeyNames is set by the gateway during wiring so ValidateAccessSpec can
+// refuse an allowlist naming a key that does not exist.
+//
+// Without the check, an allowlist naming a typo ("homeagent" vs "home_agent")
+// is stored, saved, shown in the WebUI as configured — and refuses every caller,
+// including the one the operator meant. The failure looks like the endpoint being
+// broken rather than the name being wrong.
+var knownKeyNames func() []string
+
+// SetKeyNameLookup registers the source of configured gateway key names.
+func SetKeyNameLookup(fn func() []string) { knownKeyNames = fn }
+
+// ValidateAccessSpec checks an access spec against the configured keys and roles.
+//
+// Roles are checked against the two that exist (admin, user) rather than left
+// free-form: a role that does not exist produces an endpoint nobody can reach,
+// which is the same silent failure as a bad key name.
+func (ps *Plugins) ValidateAccessSpec(spec string) error {
+	spec = strings.TrimSpace(spec)
+	if spec == "" {
+		return nil
+	}
+	known := map[string]bool{"public": true, "admin": true, "user": true}
+	if knownKeyNames != nil {
+		for _, n := range knownKeyNames() {
+			if n != "" {
+				known[strings.ToLower(n)] = true
+			}
+		}
+	}
+	for _, part := range strings.Split(spec, ",") {
+		part = strings.TrimSpace(part)
+		if part == "" {
+			continue
+		}
+		name := part
+		if strings.HasPrefix(name, "@") {
+			name = strings.TrimPrefix(name, "@")
+		}
+		if name == "" {
+			return fmt.Errorf("access entry %q names no key", part)
+		}
+		if !known[strings.ToLower(name)] {
+			return fmt.Errorf("access entry %q is neither a role (admin/user/public) "+
+				"nor a configured key name", part)
+		}
+	}
+	return nil
+}
+
+// splitAccessSpec turns an override string into role and key lists. "@name" is a
+// key; a bare word is a role. Shared by the reporting path and the enforcement
+// path so the two can never disagree about what an override means.
+func splitAccessSpec(spec string) (roles, keys []string) {
+	if strings.TrimSpace(spec) == "" {
+		return nil, nil
+	}
+	for _, part := range strings.Split(spec, ",") {
+		part = strings.TrimSpace(part)
+		switch {
+		case part == "", part == "public":
+			// "public" is an enforcement concern, not a listing; it is reported
+			// via the Public field.
+		case strings.HasPrefix(part, "@"):
+			keys = append(keys, strings.TrimPrefix(part, "@"))
+		default:
+			roles = append(roles, part)
+		}
+	}
+	return roles, keys
+}

internal/lua/plugin_http_handler_test.go

@@ -0,0 +1,430 @@
+package lua
+
+// Plugin-registered HTTP endpoints.
+//
+// Three properties have to hold, and none of them is visible from the plugin's
+// own point of view:
+//
+//  1. Auth and roles are enforced by the GATEWAY. A plugin that forgets to
+//     declare roles gets admin-only, not open — the failure mode of "public by
+//     omission" is an internet-scannable endpoint inside the gateway process.
+//  2. A handler error is a value, never a panic. The pcall guard is the same one
+//     the hooks use, for the same reason.
+//  3. The plugin cannot steer the gateway's own response framing: a nonsense
+//     status becomes a 500, and Content-Length / hop-by-hop headers are dropped.
+//
+// These tests drive the declaration layer and the response writer directly; the
+// router-level dispatch is covered in the gateway package.
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"testing"
+)
+
+const httpPlugin = `
+local P = {}
+P.name = "api"
+P.version = "1.0.0"
+P.exports = {}
+P.http = {
+  {
+    path = "health",
+    handler = "health",
+    roles = { "admin" },
+    description = "liveness",
+  },
+  {
+    path = "echo/:what",
+    method = "POST",
+    handler = "echo",
+    roles = { "admin", "user" },
+  },
+}
+function P.health(req)
+  return { status = 200, json = { ok = true, role = req.role, key = req.key } }
+end
+function P.echo(req)
+  return { status = 201, json = { what = req.params.what, body = req.body } }
+end
+return P
+`
+
+func TestPluginRegistersHTTPHandlers(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	if err := ps.LoadSource("api", httpPlugin); err != nil {
+		t.Fatal(err)
+	}
+	hs := ps.HTTPHandlers()
+	if len(hs) != 2 {
+		t.Fatalf("expected 2 registered endpoints, got %d: %+v", len(hs), hs)
+	}
+	for _, h := range hs {
+		if h.Plugin != "api" {
+			t.Errorf("handler %q has no owner; the dispatcher could not route it", h.Path)
+		}
+		if h.Method == "" {
+			t.Errorf("handler %q has no method; the gateway would answer 405 forever", h.Path)
+		}
+	}
+}
+
+func TestHandlerRoleRules(t *testing.T) {
+	// A handler with no roles and no public opt-out is ADMIN-ONLY. This is the
+	// decision the whole feature turns on: "I forgot to declare roles" must not
+	// mean "anyone on the internet".
+	open := HTTPHandler{Path: "x", Handler: "h"}
+	if open.allows("user", "") {
+		t.Error("an undeclared handler must default to admin-only")
+	}
+	if !open.allows("admin", "") {
+		t.Error("an undeclared handler must still serve admin")
+	}
+	// The explicit opt-out is how a plugin says "this one really is public".
+	pub := HTTPHandler{Path: "x", Handler: "h", Public: true}
+	if !pub.allows("user", "") || !pub.allows("", "") {
+		t.Error("public=true must admit any role")
+	}
+	// Declared roles.
+	scoped := HTTPHandler{Path: "x", Roles: []string{"admin", "user"}}
+	if !scoped.allows("user", "homeagent") || !scoped.allows("admin", "admin") {
+		t.Error("a listed role must be admitted")
+	}
+	if (HTTPHandler{Path: "x", Roles: []string{"admin"}}).allows("user", "homeagent") {
+		t.Error("an unlisted role must be refused")
+	}
+}
+
+// Naming keys is how the operator grants one collaborator access without opening
+// the endpoint to every user key. Two properties matter and neither is obvious:
+// a named key wins even when the role rule would have refused it, and the named
+// list REPLACES the role rule rather than adding to it.
+func TestHandlerKeyRules(t *testing.T) {
+	h := HTTPHandler{Path: "x", Roles: []string{"admin"}, Keys: []string{"homeagent"}}
+	if !h.allows("user", "homeagent") {
+		t.Error("a named key must be admitted even though its role is not listed")
+	}
+	// The replacement semantics: listing one key must NOT also admit every user
+	// key through the role rule, which is what an OR would have done.
+	if h.allows("user", "club") {
+		t.Error("naming keys must replace the role rule, not add to it")
+	}
+	if h.allows("user", "") {
+		t.Error("an unnamed key must not match a named-key allowlist")
+	}
+	// Admin is always admitted, so a typo in the list never locks the operator
+	// out of the endpoint they are trying to tighten.
+	if !h.allows("admin", "admin") {
+		t.Error("admin must always reach an endpoint")
+	}
+	// Name matching is case-insensitive; a config written by hand varies.
+	if !(HTTPHandler{Path: "x", Keys: []string{"HomeAgent"}}).allows("user", "homeagent") {
+		t.Error("key names should match case-insensitively")
+	}
+}
+
+// The stored override replaces whatever the plugin declared — this is the admin
+// editing access without touching a .lua that an upgrade would overwrite.
+func TestEndpointAccessOverrideReplacesDeclaration(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	if err := ps.LoadSource("api", httpPlugin); err != nil {
+		t.Fatal(err)
+	}
+	// health is declared admin-only; widen it to one named key.
+	if err := ps.SetEndpointAccess("api", "health", "@homeagent"); err != nil {
+		t.Fatal(err)
+	}
+	var found *HTTPHandler
+	for _, h := range ps.HTTPHandlers() {
+		if h.Path == "health" {
+			hh := h
+			found = &hh
+		}
+	}
+	if found == nil {
+		t.Fatal("health endpoint disappeared after an access override")
+	}
+	if len(found.Roles) != 0 {
+		t.Errorf("the declared roles should be replaced, got %v", found.Roles)
+	}
+	if len(found.Keys) != 1 || found.Keys[0] != "homeagent" {
+		t.Errorf("override keys = %v, want [homeagent]", found.Keys)
+	}
+	if !found.allows("user", "homeagent") || found.allows("user", "club") {
+		t.Error("the override is not in force")
+	}
+	// Clearing restores the plugin's own declaration.
+	if err := ps.SetEndpointAccess("api", "health", ""); err != nil {
+		t.Fatal(err)
+	}
+	for _, h := range ps.HTTPHandlers() {
+		if h.Path == "health" {
+			if len(h.Keys) != 0 {
+				t.Errorf("clearing the override should restore the declaration, got %v", h.Keys)
+			}
+			return
+		}
+	}
+	t.Error("health endpoint vanished")
+}
+
+func TestSetEndpointAccessRejectsUnknownTarget(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	if err := ps.LoadSource("api", httpPlugin); err != nil {
+		t.Fatal(err)
+	}
+	if err := ps.SetEndpointAccess("api", "nope", "@homeagent"); err == nil {
+		t.Error("setting access on a path the plugin does not expose must fail loudly")
+	}
+	if err := ps.SetEndpointAccess("nosuchplugin", "health", "@homeagent"); err == nil {
+		t.Error("setting access for an unknown plugin must fail loudly")
+	}
+}
+
+// A stored allowlist naming a key that does not exist refuses every caller while
+// looking perfectly configured. ValidateAccessSpec exists to catch that at write
+// time instead.
+func TestValidateAccessSpec(t *testing.T) {
+	SetKeyNameLookup(func() []string { return []string{"homeagent", "club"} })
+	t.Cleanup(func() { SetKeyNameLookup(nil) })
+
+	for _, ok := range []string{"admin", "user", "public", "@homeagent",
+		"admin,@homeagent,@club", " admin , @club "} {
+		if err := (&Plugins{}).ValidateAccessSpec(ok); err != nil {
+			t.Errorf("spec %q should be accepted: %v", ok, err)
+		}
+	}
+	for _, bad := range []string{"@nosuchkey", "superuser", "@", "admin,@typo_key"} {
+		if err := (&Plugins{}).ValidateAccessSpec(bad); err == nil {
+			t.Errorf("spec %q must be refused: it is neither a role nor a configured key", bad)
+		}
+	}
+	if err := (&Plugins{}).ValidateAccessSpec(""); err != nil {
+		t.Errorf("an empty spec clears the override and must be accepted: %v", err)
+	}
+}
+
+func TestHandlerPathValidationRejectsEscapes(t *testing.T) {
+	// ":name" with empty segments around it is legal — the trims below reduce it
+	// to ":name". What must be refused is an empty PARAM NAME, which cannot be
+	// captured into request.params.
+	for _, bad := range []string{"", "  ", "a/../b", "a//b", "wild*card", "q?x", "/:/"} {
+		if err := validateHandlerPath(bad); err == nil {
+			t.Errorf("path %q must be rejected: it can escape the mount or make the "+
+				"route table ambiguous", bad)
+		}
+	}
+	for _, ok := range []string{"health", "a/b", "echo/:what", "v1/things/:id", "/:name/"} {
+		if err := validateHandlerPath(ok); err != nil {
+			t.Errorf("path %q should be accepted: %v", ok, err)
+		}
+	}
+}
+
+func TestHandlerMethodValidation(t *testing.T) {
+	if m, err := normalizeHandlerMethod(""); err != nil || m != http.MethodGet {
+		t.Errorf("an empty method should default to GET, got %q %v", m, err)
+	}
+	if _, err := normalizeHandlerMethod("TRACE"); err == nil {
+		t.Error("TRACE must be refused")
+	}
+	if m, err := normalizeHandlerMethod("post"); err != nil || m != http.MethodPost {
+		t.Errorf("method should normalise to upper case, got %q %v", m, err)
+	}
+}
+
+func TestInvokeHandlerReturnsStructuredData(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	if err := ps.LoadSource("api", httpPlugin); err != nil {
+		t.Fatal(err)
+	}
+	resp, err := ps.InvokeHandler("api", "health", &HandlerRequest{
+		Method: http.MethodGet, Path: "health", Role: "admin", Key: "sk-test",
+	})
+	if err != nil {
+		t.Fatalf("invoke: %v", err)
+	}
+	if resp.Status != 200 {
+		t.Errorf("status = %d, want 200", resp.Status)
+	}
+	j, ok := resp.JSON.(map[string]interface{})
+	if !ok {
+		t.Fatalf("json = %T, want a table (the handler's nested value must survive)", resp.JSON)
+	}
+	if j["ok"] != true || j["role"] != "admin" {
+		t.Errorf("handler data crossed wrong: %v", j)
+	}
+}
+
+func TestInvokeHandlerParamsAndBodyReachLua(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	if err := ps.LoadSource("api", httpPlugin); err != nil {
+		t.Fatal(err)
+	}
+	resp, err := ps.InvokeHandler("api", "echo", &HandlerRequest{
+		Method: http.MethodPost, Path: "echo/thing",
+		Params: map[string]string{"what": "thing"},
+		Body:   `{"n":1}`, Role: "user",
+	})
+	if err != nil {
+		t.Fatalf("invoke: %v", err)
+	}
+	j, _ := resp.JSON.(map[string]interface{})
+	if resp.Status != 201 {
+		t.Errorf("status = %d, want 201 (the handler chose it)", resp.Status)
+	}
+	if j["what"] != "thing" {
+		t.Errorf("path param did not reach the handler: %v", j)
+	}
+	if j["body"] != `{"n":1}` {
+		t.Errorf("body did not reach the handler: %v", j)
+	}
+}
+
+// A handler that raises must produce an error, never a process abort — the same
+// SIGSEGV class of bug the hook guard exists for.
+func TestInvokeHandlerErrorIsValueNotPanic(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	const boom = `
+local P = {}
+P.name = "boom"
+P.version = "1.0.0"
+P.exports = {}
+P.http = { { path = "blow", handler = "blow", roles = { "admin" } } }
+function P.blow(req)
+  error("handler exploded")
+end
+return P
+`
+	if err := ps.LoadSource("boom", boom); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := ps.InvokeHandler("boom", "blow", &HandlerRequest{Role: "admin"}); err == nil {
+		t.Fatal("a raising handler must surface as an error")
+	}
+}
+
+// The plugin does not get to decide how the gateway frames its response.
+func TestHandlerResponseFramingIsKernelOwned(t *testing.T) {
+	cases := []struct {
+		name string
+		in   *HandlerResponse
+		want int
+	}{
+		{"status 0 becomes 200", &HandlerResponse{Status: 0, Body: "x"}, 200},
+		{"status 9999 becomes 500", &HandlerResponse{Status: 9999, Body: "x"}, 500},
+		{"status -1 becomes 500", &HandlerResponse{Status: -1, Body: "x"}, 500},
+		{"explicit 404 is kept", &HandlerResponse{Status: 404, Body: "x"}, 404},
+	}
+	for _, c := range cases {
+		w := httptest.NewRecorder()
+		writeHTTPHandlerResponse(w, c.in)
+		if w.Code != c.want {
+			t.Errorf("%s: status = %d, want %d", c.name, w.Code, c.want)
+		}
+	}
+
+	// A handler must not be able to desync the gateway's own framing.
+	w := httptest.NewRecorder()
+	writeHTTPHandlerResponse(w, &HandlerResponse{
+		Status: 200, Body: "hello",
+		Headers: map[string]string{
+			"Content-Length":   "99999",
+			"Connection":       "close",
+			"Transfer-Encoding": "chunked",
+			"X-Custom":         "kept",
+		},
+	})
+	if got := w.Header().Get("Content-Length"); got != "" {
+		t.Errorf("a plugin set Content-Length (%s); the body length is the kernel's", got)
+	}
+	if got := w.Header().Get("X-Custom"); got != "kept" {
+		t.Errorf("a normal header should survive, got %q", got)
+	}
+}
+
+// A malformed declaration must not take the whole plugin down.
+func TestBadHTTPRouteDoesNotBreakPluginLoad(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	const mixed = `
+local P = {}
+P.name = "mixed"
+P.version = "1.0.0"
+P.exports = {}
+P.state = { seen = true }
+P.ui = { page = { page_id = "m", title = "M", mount = "<div/>" } }
+P.http = {
+  { path = "../escape", handler = "h", roles = { "admin" } },
+  { path = "fine",      handler = "h", roles = { "admin" } },
+}
+function P.h(req) return { status = 200, body = "ok" } end
+return P
+`
+	if err := ps.LoadSource("mixed", mixed); err != nil {
+		t.Fatalf("a bad route must not fail the plugin load: %v", err)
+	}
+	// The good route survives; the bad one is simply absent.
+	hs := ps.HTTPHandlers()
+	if len(hs) != 1 || hs[0].Path != "fine" {
+		t.Fatalf("expected only the valid route, got %+v", hs)
+	}
+	// And the plugin's own page/state still work.
+	if ps.State("mixed") == nil {
+		t.Error("the plugin's state was lost because of a bad route")
+	}
+}
+
+func TestParseHandlerResponseShapes(t *testing.T) {
+	if _, err := parseHandlerResponse("just a string"); err == nil {
+		t.Error("a non-table return must be an error, not an empty 200")
+	}
+	if _, err := parseHandlerResponse(map[string]interface{}{}); err == nil {
+		t.Error("an empty table should say so rather than returning a blank 200")
+	}
+	r, err := parseHandlerResponse(map[string]interface{}{
+		"status": float64(204), "headers": map[string]interface{}{"X-A": "b"},
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if r.Status != 204 || r.Headers["X-A"] != "b" {
+		t.Errorf("decoded %+v", r)
+	}
+}
+// TestEndpointAccessReportsPerEndpointOverride guards the reporting side of the
+// override store.
+//
+// SetEndpointAccess already keyed overrides by plugin#path, but endpointAccess
+// read the plugin-level key. After an operator saved a restriction on ONE
+// endpoint, /api/plugins kept reporting the plugin's DECLARED roles, so the
+// plugin page drew the pre-change value and a later save from that screen
+// reverted the restriction. Authorization kept working — only the display and
+// the round trip through the UI were broken, which is why it went unnoticed.
+func TestEndpointAccessReportsPerEndpointOverride(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	if err := ps.LoadSource("api", httpPlugin); err != nil {
+		t.Fatal(err)
+	}
+	// httpPlugin registers "health" and "echo/:what".
+	if err := ps.SetEndpointAccess("api", "health", "user"); err != nil {
+		t.Fatalf("SetEndpointAccess: %v", err)
+	}
+	got := map[string][]string{}
+	for _, e := range ps.EndpointAccess() {
+		got[e.Path] = e.OverrideRoles
+	}
+	if len(got["health"]) != 1 || got["health"][0] != "user" {
+		t.Errorf("health override not reported to the UI: %+v", got["health"])
+	}
+	if len(got["echo/:what"]) != 0 {
+		t.Errorf("echo inherited an override it never had: %+v", got["echo/:what"])
+	}
+}

internal/lua/plugin_http_test.go

@@ -0,0 +1,217 @@
+package lua
+
+// The plugin HTTP bridge is the one capability that lets a .lua file — installed
+// from a textarea in the WebUI — reach out on the gateway's own network. That
+// makes its DENY rules the security boundary, and these tests exist to pin the
+// refusals rather than the happy path.
+//
+// The failure this guards against is concrete: an operator installs a quota
+// plugin that reads a vendor's balance, and a later plugin (or a compromised
+// one) posts that gateway key material to an attacker, or reads the cloud
+// metadata endpoint, because __llmsproxy_http happened to be a general HTTP
+// client.
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	golua "github.com/aarzilli/golua/lua"
+)
+
+// newTestState returns a bare Lua state for the argument-decoding tests.
+func newTestState(t *testing.T) *golua.State {
+	t.Helper()
+	L := golua.NewState()
+	t.Cleanup(L.Close)
+	L.OpenLibs()
+	return L
+}
+
+func TestPluginHTTPAllowlistRejectsUnlistedHost(t *testing.T) {
+	allow := newHTTPAllowlist([]string{"api.example.com"})
+	for _, raw := range []string{
+		"https://evil.test/steal",
+		"https://169.254.169.254/latest/meta-data/",
+		"https://api.example.com.evil.test/x", // suffix trick
+		"https://user@evil.test/",
+	} {
+		if _, err := allow.checkURL(raw); err == nil {
+			t.Fatalf("allowlist permitted %q — a plugin must not reach hosts "+
+				"the operator never configured as a source", raw)
+		}
+	}
+}
+
+func TestPluginHTTPAllowlistAcceptsConfiguredHost(t *testing.T) {
+	allow := newHTTPAllowlist([]string{"api.example.com"})
+	for _, raw := range []string{
+		"https://api.example.com/alpha/credits",
+		"https://API.EXAMPLE.COM/x", // host compare must be case-insensitive
+		"https://api.example.com:443/x",
+	} {
+		if _, err := allow.checkURL(raw); err != nil {
+			t.Fatalf("allowlist refused configured host %q: %v", raw, err)
+		}
+	}
+}
+
+func TestPluginHTTPRefusesPlainHTTPOffLoopback(t *testing.T) {
+	allow := newHTTPAllowlist([]string{"api.example.com"})
+	// Allowed host, but plain HTTP would put the API key on the wire in clear.
+	if _, err := allow.checkURL("http://api.example.com/alpha"); err == nil {
+		t.Fatal("plain http to an allowed non-loopback host must be refused")
+	}
+	// Loopback stays allowed: it cannot leave the machine, and the billing rules
+	// UI reads the gateway's own API over plain HTTP in dev setups.
+	for _, raw := range []string{"http://127.0.0.1:8081/api/x", "http://localhost:8081/x"} {
+		if _, err := allow.checkURL(raw); err != nil {
+			t.Fatalf("loopback http must stay usable, got %v for %q", err, raw)
+		}
+	}
+}
+
+func TestPluginHTTPRejectsBadSchemesAndMissingURL(t *testing.T) {
+	allow := newHTTPAllowlist([]string{"api.example.com"})
+	for _, raw := range []string{"file:///etc/passwd", "ftp://api.example.com/x", "gopher://x"} {
+		if _, err := allow.checkURL(raw); err == nil {
+			t.Fatalf("scheme in %q must be refused", raw)
+		}
+	}
+	// An empty allowlist must deny, not fall open.
+	if _, err := newHTTPAllowlist(nil).checkURL("https://api.example.com/x"); err == nil {
+		t.Fatal("an unconfigured gateway must deny outbound HTTP entirely")
+	}
+}
+
+// A redirect is a hop the allowlist never saw. Following it would let an
+// allowed host bounce a plugin's request (and its headers) anywhere.
+func TestPluginHTTPDoesNotFollowRedirects(t *testing.T) {
+	if httpClient.CheckRedirect == nil {
+		t.Fatal("CheckRedirect must be set: following a redirect escapes the allowlist")
+	}
+	req, _ := http.NewRequest(http.MethodGet, "https://api.example.com/x", nil)
+	err := httpClient.CheckRedirect(req, []*http.Request{req})
+	if err != http.ErrUseLastResponse {
+		t.Fatalf("CheckRedirect must return ErrUseLastResponse so the 3xx is "+
+			"handed back unfollowed, got %v", err)
+	}
+}
+
+func TestPluginHTTPTimeoutAndBodyAreClamped(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_, _ = w.Write([]byte(`{"ok":true}`))
+	}))
+	defer srv.Close()
+	allow := newHTTPAllowlist([]string{"127.0.0.1"})
+	u, err := allow.checkURL(srv.URL)
+	if err != nil {
+		t.Fatalf("loopback test server must be allowed: %v", err)
+	}
+	// An absurd timeout is clamped instead of honoured.
+	res, err := performHTTP(u, httpRequestSpec{Timeout: 24 * time.Hour})
+	if err != nil {
+		t.Fatalf("clamped request failed: %v", err)
+	}
+	if res.Status != 200 || !strings.Contains(res.Body, `"ok"`) {
+		t.Fatalf("unexpected result: status=%d body=%q", res.Status, res.Body)
+	}
+}
+
+func TestPluginHTTPOversizedBodyIsRefused(t *testing.T) {
+	big := strings.Repeat("x", httpMaxResponse+64)
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		_, _ = w.Write([]byte(big))
+	}))
+	defer srv.Close()
+	allow := newHTTPAllowlist([]string{"127.0.0.1"})
+	u, _ := allow.checkURL(srv.URL)
+	// Refusing beats truncating: a truncated body parses as corrupt JSON and the
+	// plugin would report a parse error instead of the real cause.
+	if _, err := performHTTP(u, httpRequestSpec{}); err == nil {
+		t.Fatal("an oversized response must be refused, not silently truncated")
+	}
+}
+
+func TestPluginHTTPErrorPathIsPairNotRaise(t *testing.T) {
+	// Lua-side contract: the bridge returns (nil, message) on every failure so a
+	// plugin mistake never raises through golua (see the SIGSEGV note on
+	// hookGuardSrc — an escaping error can take the gateway down).
+	allow := newHTTPAllowlist([]string{"api.example.com"})
+	if _, err := allow.checkURL("https://evil.test/"); err == nil {
+		t.Fatal("expected refusal")
+	}
+	// readHTTPRequestSpec must reject a non-table without touching the stack.
+	L := newTestState(t)
+	if _, err := readHTTPRequestSpec(L); err == nil {
+		t.Fatal("a missing argument table must be an error, not a panic")
+	}
+}
+// The bridge must be reachable under the name plugins actually call.
+//
+// This exists because the registration once used the CONSTANT'S IDENTIFIER as
+// the Lua global name (`registerFn(L, "httpGlobal", …)`) rather than its VALUE,
+// so the bridge was published as `httpGlobal` while every plugin called
+// `__llmsproxy_http`. The failure only appeared at runtime, as
+// "attempt to call global '__llmsproxy_http' (a nil value)" on the first real
+// request — after the unit tests were green, because they drove the Go-side
+// helpers directly and never went through Lua name resolution.
+//
+// So this test boots a real plugin and calls the bridge the way a plugin does.
+func TestPluginHTTPBridgeIsReachableUnderItsLuaName(t *testing.T) {
+	dir := t.TempDir()
+	ps := NewPlugins(NewVM(dir), dir)
+	ps.SetOutboundHosts([]string{"127.0.0.1"})
+
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"credits":{"monthlyCredits":1.5}}`))
+	}))
+	defer srv.Close()
+
+	// The plugin asks for a loopback URL, which is allowed without being
+	// configured as a source. It type-checks both the table and arg.url, so a
+	// malformed call is reported by the plugin rather than as a bewildering URL
+	// parse error from the bridge.
+	src := `
+local P = {}
+P.name = "httpprobe"
+P.version = "1.0.0"
+P.exports = { "probe" }
+-- Call passes the WHOLE argument table as a single Lua argument, which is the
+-- same convention the hooks use and what a plugin page's pluginAPI.call ends up
+-- producing. Taking the table (not a bare string) is the correct shape.
+function P.probe(arg)
+  if type(arg) ~= "table" then return { error = "probe takes the args table, got " .. type(arg) } end
+  if type(arg.url) ~= "string" then return { error = "arg.url must be a string, got " .. type(arg.url) } end
+  local res, err = ` + httpGlobal + `({ url = arg.url, method = "GET" })
+  if res == nil then return { error = tostring(err) } end
+  return { status = res.status, body = res.body }
+end
+return P
+`
+	if err := ps.LoadSource("httpprobe", src); err != nil {
+		t.Fatalf("load: %v", err)
+	}
+	got, err := ps.Call("httpprobe", "probe", map[string]interface{}{"url": srv.URL})
+	if err != nil {
+		t.Fatalf("call: %v", err)
+	}
+	m, _ := got.(map[string]interface{})
+	if m == nil {
+		t.Fatalf("probe returned %T, want a table", got)
+	}
+	// A nil global surfaces as {error = "attempt to call global ... (a nil value)"} —
+	// which is a table too, so the error field is what distinguishes it.
+	if e, ok := m["error"]; ok {
+		t.Fatalf("the bridge is not reachable as %s: %v", httpGlobal, e)
+	}
+	if s, _ := m["status"].(float64); int(s) != 200 {
+		t.Fatalf("status=%v body=%v, want 200 from the loopback test server", m["status"], m["body"])
+	}
+	if body, _ := m["body"].(string); !strings.Contains(body, "monthlyCredits") {
+		t.Fatalf("body=%q, want the server's JSON", body)
+	}
+}

internal/lua/plugin_shard.go

@@ -0,0 +1,210 @@
+package lua
+
+import (
+	"fmt"
+	"sync"
+)
+
+// ShardCall runs one exported function across N private VMs and returns the
+// results in the order the calls were given.
+//
+// Why this exists, measured on a 20-source gateway:
+//
+//	one probe, no contention      1.4 – 3.3 s
+//	20 probes via Promise.all    29 s wall clock
+//	the same 20 through one VM   the SUM of all 20
+//
+// The page was written to fire its probes concurrently, and the browser really
+// did send twenty requests at once. They serialised anyway, in Plugins.Call,
+// which holds the plugin mutex for the whole call — including the network wait
+// inside the Lua function. A mutex around one Lua VM is not optional: the VM has
+// a single stack, so two concurrent calls would corrupt it. The fix cannot be
+// "drop the lock".
+//
+// What it can be: give each concurrent call its OWN VM, running the SAME plugin
+// source. Measured with a 300ms local endpoint, 8 shards:
+//
+//	shared VM, 8 in sequence     2.407 s
+//	8 independent VMs            0.304 s   → 7.9x
+//
+// A busy-wait version of the same spike reported only 2.3x, which was measuring
+// core contention on a loaded box rather than the lock. Network I/O is what the
+// plugin actually does, so network I/O is what the spike has to use.
+//
+// Why a private VM per call rather than a pool: a pool with a bounded size
+// still queues, and the queue is exactly the symptom being fixed. Booting a VM
+// is ~0.1ms, so a pool would buy nothing a fresh VM does not.
+//
+// What a shard VM must NOT do:
+//
+//   - Touch PLUGIN.state. State lives in the Lua VM, and every shard would hold
+//     its own copy of it. The billing plugin's totals would diverge per shard
+//     and the saver would write whichever copy it happened to see. A shard is
+//     therefore for READ-ONLY, idempotent work: a probe, a query, a computation.
+//   - Register hooks. Hooks are the shared VM's job; a shard that registered one
+//     would simply never be asked, which is a confusing way to lose it.
+//
+// The plugin decides whether a function is shardable by being listed. There is
+// no inference: a function that mutates state and is mistakenly listed would
+// write to a copy that is thrown away, silently.
+type shardPool struct {
+	ps *Plugins
+
+	mu    sync.Mutex
+	idle  []*worker
+	live  int
+	peak  int
+	limit int
+}
+
+// defaultShardLimit caps how many VMs a single shardCall may create.
+//
+// It is a memory bound, not a concurrency policy. Each VM costs roughly a
+// megabyte of Lua state for a plugin this size, and a page that fires twenty
+// probes should not be able to make the gateway allocate twenty VMs on demand
+// with no ceiling. Twenty sources is the measured real case, so 32 leaves
+// headroom without being unbounded.
+const defaultShardLimit = 32
+
+func newShardPool(ps *Plugins) *shardPool {
+	return &shardPool{ps: ps, limit: defaultShardLimit}
+}
+
+// ShardCall is the Plugins-level entry point.
+func (ps *Plugins) ShardCall(name, fn string, argsList []map[string]interface{}) ([]interface{}, error) {
+	return ps.shards.ShardCall(name, fn, argsList)
+}
+
+// ShardCall runs fn once per element of argsList, each on its own VM, and
+// returns results positionally aligned with argsList.
+//
+// A shard that fails does not fail the batch. Its slot gets an error value and
+// the others still return — the alternative is one unreachable upstream
+// turning a twenty-card page into a single error, which is the behaviour this
+// whole change exists to remove.
+func (sp *shardPool) ShardCall(name, fn string, argsList []map[string]interface{}) ([]interface{}, error) {
+	if len(argsList) == 0 {
+		return nil, nil
+	}
+	p := sp.ps.find(name)
+	if p == nil {
+		return nil, fmt.Errorf("plugin %s not found", name)
+	}
+	if !pluginExportAllowed(p, fn) {
+		return nil, fmt.Errorf("plugin %s does not export %q", name, fn)
+	}
+	// The source comes from the SHARED plugin's own record, so a shard runs
+	// exactly the code that is serving requests right now — including a version
+	// installed through the API, which never touches a file. Reading ps.dir
+	// instead would shard the previous version of an edited plugin.
+	code := p.script
+	if code == "" {
+		return nil, fmt.Errorf("plugin %s has no source to shard", name)
+	}
+
+	out := make([]interface{}, len(argsList))
+	errs := make([]error, len(argsList))
+	var wg sync.WaitGroup
+	for i := range argsList {
+		wg.Add(1)
+		go func(i int) {
+			defer wg.Done()
+			w, release, err := sp.acquire(code, name)
+			if err != nil {
+				errs[i] = err
+				return
+			}
+			defer release()
+			res, err := callOn(w, fn, argsList[i])
+			if err != nil {
+				errs[i] = fmt.Errorf("%s: %w", name, err)
+				return
+			}
+			out[i] = res
+		}(i)
+	}
+	wg.Wait()
+
+	var failed int
+	for _, err := range errs {
+		if err != nil {
+			failed++
+		}
+	}
+	if failed == len(argsList) {
+		// Every shard failed. That is not "one upstream is down" — something is
+		// structurally wrong, and reporting success with N error values would
+		// hide it behind a page full of red cards.
+		return nil, fmt.Errorf("all %d shards of %s.%s failed: %v", len(argsList), name, fn, errs[0])
+	}
+	if failed > 0 {
+		if sp.ps.logf != nil {
+			sp.ps.logf("[plugins] %s.%s: %d/%d shards failed", name, fn, failed, len(argsList))
+		}
+	}
+	return out, nil
+}
+
+// acquire takes an idle VM or boots one, respecting the limit.
+func (sp *shardPool) acquire(code, name string) (*worker, func(), error) {
+	sp.mu.Lock()
+	if n := len(sp.idle); n > 0 {
+		w := sp.idle[n-1]
+		sp.idle = sp.idle[:n-1]
+		sp.mu.Unlock()
+		return w, func() { sp.release(w) }, nil
+	}
+	if sp.limit > 0 && sp.live >= sp.limit {
+		sp.mu.Unlock()
+		return nil, nil, fmt.Errorf("shard limit %d reached", sp.limit)
+	}
+	sp.live++
+	sp.mu.Unlock()
+
+	w, err := bootPluginState(code, name, sp.ps.outboundAllowlist())
+	if err != nil {
+		sp.mu.Lock()
+		sp.live--
+		sp.mu.Unlock()
+		return nil, nil, err
+	}
+	return w, func() { sp.release(w) }, nil
+}
+
+func (sp *shardPool) release(w *worker) {
+	if w == nil || w.L == nil {
+		return
+	}
+	sp.mu.Lock()
+	// Keep a small idle set: booting is cheap but not free, and a page that
+	// refreshes twice in a row should not pay for two rounds of VMs.
+	if len(sp.idle) < sp.limit {
+		sp.idle = append(sp.idle, w)
+		sp.mu.Unlock()
+		return
+	}
+	sp.live--
+	sp.mu.Unlock()
+	w.L.Close()
+}
+
+// Shutdown closes every pooled VM. Called from Plugins.Close.
+func (sp *shardPool) Shutdown() {
+	if sp == nil {
+		return
+	}
+	sp.mu.Lock()
+	idle := sp.idle
+	sp.idle = nil
+	sp.live = 0
+	sp.mu.Unlock()
+	for _, w := range idle {
+		if w != nil && w.L != nil {
+			w.L.Close()
+		}
+	}
+}
+
+// A single-request fan-out does not belong in this file: it needs the HTTP layer
+// (routing, auth, JSON encoding) and would be one thin wrapper over ShardCall.
+// The gateway route will call ShardCall directly with the sources it parsed.