ccquota asked one vendor one question. The shape that generalizes is two parts:
a small set of probe adapters, and a declarative override for a provider nobody
anticipated.
Three adapters, each verified against a LIVE endpoint rather than assumed:
- deepseek family: GET /user/balance →
{is_available, balance_infos:[{currency,total,granted,topped_up}]}.
Granted vs topped-up is kept separate on purpose: gift credit expiring on a
schedule is a different operational fact from prepaid money.
- OpenAI console shape: /v1/dashboard/billing/subscription +
.../billing/usage. This reports LIMITS and SPEND, not a balance — the live
provider has hard_limit_usd 1e8. There is deliberately no `remaining`:
ceiling - spend would read as "$99,999,339 left", a number nobody should
believe. It renders as spend against a ceiling instead.
- commandcode: rolling 5h / weekly / monthly windows plus billing period.
A provider with no working adapter is SKIPPED and listed by name at the bottom.
Most of these upstreams are small re-sells whose dashboard paths return an HTML
page rather than JSON; a table full of "unknown" rows teaches the operator to
ignore the page entirely. Failures that are NOT "no endpoint" — a 403 on a key we
hold — stay visible, because those are the ones worth fixing.
## Cost control, and a lock the probe runs under
First load measured 101s for 21 sources: 3 adapters x 2 attempts x 4s, times 21.
Call() holds the plugin's write lock for the whole invocation, so every
__llmsproxy_http round trip inside it blocks request_end hooks — which is how
billing accrues money. A slow upstream is not just a slow card.
- Timeout cut to 4s (from 10s).
- A host->adapter memo, persisted in plugin.state, so the dead ends are
discovered once per process instead of on every page open. Steady state is
22s.
- The memo records NEGATIVE results too, which is what actually paid: only
recording successes left every silent source re-probing forever.
- Negatives carry a timestamp and expire after 30 minutes. "This upstream has
no balance endpoint" is a fact that changes — vendors ship dashboards — and a
permanent negative makes the page quietly stop reporting a source that
started working.
- The page calls collect_source concurrently, one call per source, so a single
502 over 21 sequential probes becomes 21 independent answers.
## Two bugs the page found
- deepseek reports topped_up_balance -0.02 against a positive grant; the `v <= 0
→ "0.00"` branch flattened a real debit to zero. Negative balances now render
as -¥0.02.
- PLAN_CAPS was declared below the adapter that reads it, so Command Code hit
"attempt to index global 'PLAN_CAPS' (a nil value)" on every probe. A Lua local
is nil until its assignment runs.
Verified on production: 5 sources report, 16 skipped by name, 1 visible failure
(tabitoken 403). deepseek ¥0.32 (赠送 0.34 · 充值 -0.02), justwoker $360,865.71
spent against an ∞ ceiling, both Command Code seats with their three windows.
Co-Authored-By: ModelRouter <noreply@modelrouter.dev>
The page showed a monthly BALANCE ($0.08 left) but no monthly ALLOWANCE, so
"how much do I get and how much of it is gone" had no answer — the most useful
number on a plan that is nearly exhausted.
The vendor's credits endpoint reports what is left and never the cap, so a
percentage is not derivable from it. Two things supply the missing half:
- planId from subscriptions (here individual-goat), mapped to the caps the
vendor publishes ($10 Go, $70 GOAT/Pro/Max).
- totalMonthlyCredits from usage/summary, which counts over the BILLING PERIOD
and so pairs with the monthly allowance rather than with the rolling 5h and
weekly windows.
Because the cap is a table lookup and not an API fact, it is cross-checked
against monthly_used + balance: if the two disagree by more than 2% + $0.50,
the card says the figure may be off rather than confidently printing a wrong
percentage. A planId absent from the table renders the balance with no bar at
all — inventing a denominator would be worse than admitting there isn't one.
Also surfaces what the endpoints were already returning and the page was
dropping: period_end (with a countdown), cancel_at_period_end, plan name, and
the vendor's own belowThreshold flag, which is distinct from "balance > 0" —
an account can hold $3 and still be under their threshold.
Verified on production: both accounts show 每月 100% ($69.94/$70, red bar),
5 小时 0/14, 每周 0.803/35 on one; plan goat, 周期至 2026-10-17 / 2026-10-19.
Co-Authored-By: ModelRouter <noreply@modelrouter.dev>
The plugin page rendered nothing until the operator clicked refresh, and even
the click handler was never wired — while every unit test was green and the
API returned correct numbers. Four separate mistakes stacked here:
1. The script waited on DOMContentLoaded, an event that had already been
dispatched by the time the kernel injects a plugin mount. Replaced with
pluginAPI.onTabShown (the kernel's own mechanism, fires with the DOM in
place) PLUS an immediate call, because onTabShown alone leaves the page
blank when the tab is already open at boot. The billing page does both; I had
reduced it to one and broke first paint.
2. The re-entry guard flag lived in a closure variable. The tab's DOM is rebuilt
on re-entry, so the handler was wired against a node that no longer existed.
The flag now lives on the NODE (__ccqWired), which is what makes re-wiring
correct.
3. refresh() swallowed its fetch failure and rendered "no sources", so a broken
page and an empty account list looked identical. Failures now surface in the
header line, including when pluginAPI.call is missing (an old kernel).
4. THE ACTUAL REASON the page stayed blank: there were TWO ccquota plugins. The
embed (go:embed plugins/*.lua) version and the on-disk one both loaded, and
the UI injected the EMBED copy — 5182 characters, predating every fix — while
every edit I made and every hot reload went to the disk copy. The running
binary was serving stale plugin code no amount of reloading could fix. Only a
rebuild surfaced it, and only measuring the served <script> length (7053 vs
5182) showed which copy the browser had.
Lesson worth keeping: "the file on disk is what runs" is false for anything
under go:embed. Compare what the SERVER sends before editing the same file
twice and concluding the edit did nothing.
Verified in the browser: 2 account cards, 4 window bars, real balances, no
overflow, no console errors, refresh button 48x29 and functional.
Co-Authored-By: ModelRouter <noreply@modelrouter.dev>
A quota plugin cannot work on billing's model. Billing reports what the gateway
already measured, delivered on hooks during request forwarding. A plan balance
has no such arrival path: it exists only on the vendor's side, and the only way
to learn it is to ask — at the moment the operator clicks Refresh. No hook fires
on a click, and pluginAPI offered only state read/write.
So this adds the missing piece: POST /api/plugins/{name}/call/{fn}, plus
pluginAPI.call on the page side.
## Why the gate is an explicit export list
That endpoint is an HTTP surface into a Lua state. Without a declared allowlist
it would let any caller invoke the plugin's private helpers with crafted
arguments and read whatever it can see — for a plugin holding an API key, that
is handing the key to anyone who can POST. So a function is callable only if the
plugin listed it in plugin.exports. A plugin with no exports has no callable
surface; reading the missing list as "allow all" is exactly the failure mode
worth designing against. The endpoint is also admin-only.
## ccquota
One account per configured source on api.commandcode.ai, keyed by that source's
own key — two keys on one host are two seats, and both are shown. Reads
/alpha/whoami, /alpha/billing/credits, /alpha/billing/subscriptions and
/alpha/usage/summary. Each endpoint degrades independently: a vendor retiring
one route must not blank the others, and a key out of quota still reads credits
fine while inference 401s, so the report records WHICH part failed rather than
rendering a zero that reads as "no quota left".
The 401-is-everything trap: those endpoints need a bearer token plus three
headers. The token lives in the table now, per request. Passing the shared
header table straight through — my first version — yields 401 on every
endpoint, which is indistinguishable from a revoked key and sent me auditing the
account instead of the plugin.
## Also fixed: three real defects found by going live
- The bridge was registered as `httpGlobal` (the Go const's IDENTIFIER) instead
of `__llmsproxy_http` (its VALUE). Green unit tests, "attempt to call global
'__llmsproxy_http' (a nil value)" on first real request. Now a test boots a
real plugin and calls it the way a plugin does.
- The call router kept `call/` in the function name, so every call 502'd with
`does not export "call/collect_account"` while the exports list looked fine.
- readPluginExports used luaToJSON, which needs the `json` global — not yet
installed while a plugin is registering — so it silently returned an empty
list and every export looked undeclared. It uses luaTableToGoValue now.
Two tests hardcoded counts ("1 bundled plugin", "4 pages") and turned red the
moment a second bundled plugin existed. They now assert what they mean: billing
is among the bundled plugins, and the pages this test contributed all merged in
order. A count assertion is a changelog detector.
## Verified on production
commandcode 2198972886@qq.com active $0.0811 5h 0/14 wk 0.803/35 44,921 req
commmandcode2 jianf3112@gmail.com active $0.0700 5h 0/14 wk 0/35 43,827 req
Co-Authored-By: ModelRouter <noreply@modelrouter.dev>
Follow-on to the session-stability fix. "Per source" already made the
prefix cache hit, but it puts every conversation into one upstream session.
Using the client's own session id is not possible: capturing real agent
traffic (tcpdump on 127.0.0.1:8081) shows generic OpenAI clients send NO
session identifier at all — no user / session_id / conversation_id /
metadata in the body, and no session header (only X-Stainless-* plus
User-Agent: pi). The x-opencode-session the Go endpoint asks for is an
OpenCode native-client concept that a generic client cannot forward.
Since history is replayed every turn, the FIRST user message is invariant
for the life of a conversation, so it is used as the conversation
fingerprint. The session becomes stable within a conversation and distinct
across conversations; requests with no user message fall back to per-source
stability.
Measured through the gateway (same 5.7k-token prompt): 2nd call
cached_tokens=5504, and an unrelated conversation gets its own session.
Test: TestOpenCodeSessionIsStableForCache covers same-conversation
stability, cross-conversation separation, per-request request ids and the
sessionless fallback.
(cherry picked from commit 39b48e556e)
The opencode adapters derived x-opencode-session from meta.timestamp, i.e. a
brand new session on every request. The upstream prefix cache is
session-scoped, so no request could ever hit it, and the cache fields the
endpoint does report (prompt_tokens_details.cached_tokens,
prompt_cache_hit_tokens/prompt_cache_miss_tokens) always came back 0/absent.
Measured against the live endpoint, same 6032-token prompt:
fixed session id -> 2nd call: hit 5888, miss 144
rotating session id -> every call: hit 0, miss 6032
Fix: derive the session from the source name (stable), matching how
x-opencode-project is already derived. x-opencode-request stays unique per
request — it is only a request identifier, not part of the cache key.
Applied to both opencodego and opencodezen.
Through the gateway the same prompt now reports, on the 2nd call:
details={'cached_tokens': 5888} hit=5888 miss=144 (non-streaming)
prompt_tokens_details={'cached_tokens': 5888} (streaming)
Test: TestOpenCodeSessionIsStableForCache asserts the session is stable
across requests for one source while the request id differs.
(cherry picked from commit 791d198f47)
v1.5.4 stopped stripping reasoning_content, which fixes clients that send
it — but most agent clients (pi included) never store or replay their
reasoning, keeping only the tool call. OpenCode Go validates the field on
any assistant turn that carries tool_calls and rejects the whole request:
400 invalid_request_error: The `reasoning_content` in the thinking mode
must be passed back to the API.
Verified against the live endpoint that an EMPTY string satisfies the
check, so the adapter now fills in "" when a tool-calling assistant turn
has no reasoning_content. Nothing is fabricated: the reasoning shown to
the client is still exactly what the upstream returned for that turn.
Measured: with a tool_call + tool_result history and no reasoning_content,
all 25 configured Go models returned 400 before and all 25 answer
correctly now.
Test: TestOpenCodeGoVsZenReasoning also pins that a plain assistant turn
(no tool calls) must NOT gain the field.
(cherry picked from commit d1a72cd23a)
Zen (https://opencode.ai/zen/v1) and Go (https://opencode.ai/zen/go/v1)
are different services with different requirements, and one shared adapter
could not satisfy both.
The decisive difference is reasoning_content:
* OpenCode Go runs thinking models and REQUIRES the assistant turn's
reasoning_content to be echoed back. The shared adapter stripped it
(msg.reasoning_content = nil), so every replay of a thinking turn
failed with:
400 invalid_request_error: The `reasoning_content` in the thinking
mode must be passed back to the API.
Reproduced directly: the same request with reasoning_content -> 200,
without -> 400. That is why the Go tier never worked in an agent loop.
* The Zen free pool must not receive it, so it keeps stripping.
Both adapters keep the earlier fixes they share (never drop an assistant
turn carrying tool_calls; send stream_options only when streaming; role
whitelist; multimodal strip) and the opencode client fingerprint headers —
the Go endpoint additionally REQUIRES x-opencode-session, which the
adapter already sends.
config: localzen -> opencodezen, gozen -> opencodego.
Verified: all 25 gozen models answer correctly through the gateway with a
thinking + tool_call + tool_result history (was 0/25 before), streaming
included; the Zen free models still pass.
Test: TestOpenCodeGoVsZenReasoning pins the Go-keeps / Zen-strips split.
(cherry picked from commit 71e9040a45)
OpenCode Go (and other strict OpenAI-compatible upstreams) reject a
non-streaming request that carries stream_options with
"stream_options should be set along with stream". The adapter attached it
unconditionally, so every non-stream call through the opencode adapter
failed on those upstreams.
Verified against OpenCode Go: 25/25 configured models now pass a real
completion through the gateway (they previously 400'd).
Test: TestOpenCodeStreamOptionsOnlyWhenStreaming (absent when
non-streaming, include_usage present when streaming).
(cherry picked from commit 0528941e24)
Three related forwarding defects found by auditing every adapter with a
tool-calling replay (assistant turn with content:[] + tool_calls).
1) content:[] -> content:{} (all 12 openai-adapter sources, plus
deepseek/trae/sensenova/agentrouter/github/groq/kimicode/mistral)
Lua adapters json.decode the request and re-encode it, and an empty Lua
table is indistinguishable from an empty JSON array — the encoder emits
{} for both. Agent clients serialise a tool-calling assistant turn with
no text as content:[], so every pass-through adapter rewrote it to
content:{} — not valid OpenAI (content is string|array|null). Verified
against a live upstream: content:[] produced "400 invalid arguments"
while content:"" was accepted.
Fixed once at the decode boundary (types.ChatMessage.UnmarshalJSON):
empty-array content normalises to "" and an empty tool_calls array is
dropped, so every adapter — including future ones — sees a valid shape.
2) gemini dropped tool_calls and never emitted functionCall /
functionResponse; the tool role also stayed as an invalid role inside
contents and system was not moved to systemInstruction.
3) ollama copied only role/content, dropping tool_calls and the call
attribution entirely (it needs tool_name, not tool_call_id).
Test: TestAdaptersPreserveToolCalls asserts, for every adapter, that the
call id (or function name where the wire format has no id), the function
name, the tool result and the trailing user turn all survive, plus a
negative control for plain text.
(cherry picked from commit 9114468753)
An agent client (pi) serialises an assistant turn whose content is only
[thinking, toolCall] as content:[] with tool_calls. The multimodal-strip
pass treated an empty content array as 'nothing left, drop the whole
message' and discarded the tool_calls with it.
The next message is that call's tool result, so it arrived orphaned: the
model saw a result for a call it had never made and re-issued the same
call on every turn — an endless repeated-tool-call loop. Reproduced
against a capture sink: content:[] lost tool_calls, while content:"" and
content:null kept them.
Only messages with neither usable content NOR a tool call now get
dropped. Content that collapses to empty but still has tool_calls or a
tool_call_id is emitted as "" instead.
Test: TestOpenCodeKeepsToolCallWithEmptyContent (plus a negative control
that an image-only message without tool calls is still dropped).
(cherry picked from commit 499f0cac2f)
A key with scope=[AUTO] could previously:
1. request ANY concrete model id directly (hasScopeModel/checkModelScope
treated AUTO as a wildcard)
2. see the full 56-model list on /v1/models (intersectModels considered
AUTO as grant-everything)
AUTO now only authorizes the AUTO routing mode. Direct requests to a
specific model require an explicit scope entry.
Also carries agentrouter.lua WAF fingerprint headers (Origin/Referer/
X-Requested-With) already staged on this branch.
Tests: TestHasScopeModelWithSourcePrefix updated; full suite green.
(cherry picked from commit 7fb8f96b82)
Full-repo audit after the anthropic/openai fix: agentrouter exposes
claude-opus-4-8, so it inherits Anthropic's tool id rule
^[a-zA-Z0-9_-]{1,64}$ and rejects the whole request on a violation, exactly
like justwoker/tabitoken/扇贝. It was the only remaining adapter serving Claude
models without the sanitizer, so a client that had picked up a dirty id (e.g.
"bash:0" from moonshotai/kimi-k3) would still lose every turn here.
Same shape as the other two — inbound tool_calls[].id + tool_call_id, outbound
non-streaming ids and the first streamed fragment — and the test now asserts
all three adapters rewrite an identical input identically, so a client mixing
sources within one session cannot end up with unpaired tool calls.
Audit result: every source exposing a claude/opus/sonnet model (qijiar, toter,
juziai, agentrouter, justwoker, api456, tabitoken) now routes through a
sanitizing adapter.
Anthropic and OpenAI disagree on what the prompt count means:
Anthropic: input_tokens EXCLUDES cached blocks; cache_read_input_tokens and
cache_creation_input_tokens are separate, additive, billed input.
OpenAI: prompt_tokens INCLUDES its cached_tokens subset.
anthropic.lua mapped input_tokens straight onto prompt, so a cache-heavy turn
was doubly wrong: the billed prompt was undercounted by the entire cache
portion, and cached_tokens could exceed prompt_tokens — a cache hit rate above
100% for any client that divides one by the other. cache_creation_input_tokens
was never read at all, so a cache-write turn silently lost those billed tokens.
Worse, the streaming path dropped the cache split entirely: message_delta
carries the FINAL usage and only mapped input/output, so every streamed
response reported no cache information even when the upstream sent it.
All three counts are now summed into prompt, with the read half exposed as
prompt_tokens_details.cached_tokens plus the DeepSeek-legacy hit/miss pair, via
one shared map_usage() used by transform_response, message_start and
message_delta. A reported zero stays distinguishable from "never reported": the
split is emitted whenever either cache field is present, and omitted entirely
when the upstream mentions neither (justwoker reports only input/output plus its
own cost fields, so its output is byte-identical to before). map_usage returns
nil for a countless object, preserving "no usage in this chunk means say
nothing" rather than reporting zeros.
message_start's placeholder count is still emitted: justwoker reports 160 there
and the real 6931 in message_delta, and the gateway's mergeUsage lets the later
non-zero value win.
Anthropic requires tool_use.id / tool_result.tool_use_id to match
^[a-zA-Z0-9_-]{1,64}$ and rejects the WHOLE request otherwise with
REQUEST_BODY_INVALID / "Invalid tool use format". OpenAI has no such rule, so
an OpenAI-compatible model can mint an id like "bash:0"
(xinjianya/moonshotai/kimi-k3 does exactly that).
In a fan-out router that id does not stay local: the client stores it in its
history and replays it to every other source. One such id therefore kills
every Claude slot at once — justwoker, tabitoken and 扇贝 are all
Claude-behind-{OpenAI,Anthropic} — and an AUTO request falls through all four
tiers to whatever tolerant model is left. Observed live: 4 consecutive 503
"all N auto providers failed" with tier 1/2/3 each reporting the same 400.
Both directions are sanitized, in both adapters:
- request: tool_calls[].id and tool_call_id, so poisoned history recovers
- response: non-streaming tool_calls[].id and the first streamed fragment,
so a bad id never enters a client session in the first place
safe_tool_id is pure and deterministic, so a call and its result are rewritten
identically within one request. A rewritten id keeps an 8-hex digest of the
original, without which distinct ids could collapse ("a:b" and "a_b") into a
duplicate/unpaired tool_use. Already-legal ids pass through byte-identical, so
well-behaved traffic is unaffected. openai.lua carries its own copy because
Lua adapters have no shared prelude.
Streamed argument fragments carry no id and must stay id-less, otherwise
index-based accumulation on the client breaks; a test pins that.
Root cause: trae-local-api is deployed to users without the "Fold past
assistant tool_calls into [Called tool: name({...})]" text format that
their own histories already contained. Trae-Local-API-LLM then mimics this
format in subsequent responses. The trae.lua parser only recognized
<tool_call>...</tool_call> or <toolcall>...</toolcall> tags, so
[Called tool: ...] responses were left unparsed and the client received
plain text where a structured tool_calls array should be.
Fix: Add a legacy pattern match at the END of parse_text_tool_calls to
catch the [Called tool: name({args})] shape and emit proper tool_calls.
This is a fallback; models should emit <tool_call> tags per system prompt,
but we tolerate the mimicked form for robustness.
Four adapters handled tool_calls in transform_stream_chunk but lost them in
transform_response, so any NON-streaming tool-using conversation broke on its
second request: the client received finish_reason:"tool_calls" with no
tool_calls payload, replayed an assistant message whose function
name/arguments were empty, and the upstream rejected the next turn with
400 invalid tool_call function, function/name/arguments cannot be empty
The production audit trail shows 46 such failures on sensenova alone.
- sensenova.lua: forward message.tool_calls, decoding the arguments JSON string
into an object as the unified shape expects.
- gemini.lua: collect functionCall parts from candidates[].content.parts. Also
correct finish_reason, since Gemini reports "STOP" even when it emitted a
function call and clients keyed on it treat that as a finished answer.
- ollama.lua: the field was initialized to an empty table and never filled;
fill it and likewise correct done_reason "stop" -> "tool_calls".
trae is a different failure with the same symptom: trae-local-api's OpenAI
endpoint (/v1/chat/completions, src/server.js:353) never reads the request's
`tools` array — only its Anthropic endpoint does — so the relayed model is never
told the tool schema and instead PRINTS a <tool_call>{...}</tool_call> block into
content, leaving message.tool_calls null and finish_reason "stop". An OpenAI
client sees an ordinary completion and its agent loop ends mid-conversation.
trae.lua now recovers the structured call from that text, strips the block from
user-visible content, and corrects finish_reason. Both tag spellings
(<tool_call>/<toolcall>, the latter is what the same codebase's Anthropic prompt
asks for) and all three argument key names (arguments/params/input) are accepted.
This is a defensive fallback: fixing the upstream shim to honour `tools` remains
the real fix, since the model still guesses parameter names.
Tests: TestNonStreamToolCallsPreserved covers all ten OpenAI-shaped adapters,
TestGeminiNonStreamToolCalls and TestOllamaNonStreamToolCalls cover their native
shapes, TestTraeTextToolCallRecovery covers both tag spellings, prose around the
block, and asserts a plain text answer never gains tool_calls.
Verified end-to-end against mock upstreams reproducing each shape: a full
two-round agent loop (tool call -> tool result -> final answer) now completes for
both the structured and the text-emitted variants.
Production oscillated between ~46 MB and ~56 MB RSS with the openai pool
cycling 1 -> 10..12 -> 1 states every couple of minutes, while peak_in_use
never went above 2.
Cause: the batch prewarm sized itself purely on the adapter's ceiling. With
max_concurrent summing to 76, growStep is 8, so any two overlapping requests
warmed 8 states — 6 more than anything was waiting for. A minute later the
janitor correctly reclaimed the surplus, the next pair of overlapping requests
warmed 8 again, and the pool churned boot/discard forever. The elasticity was
working; the growth signal was simply wrong.
Prewarm is now bounded by BOTH limits: the ceiling still caps the step, but the
batch never exceeds p.waiting, the number of goroutines actually blocked on the
pool. Overlapping-but-not-queued traffic (the common case) creates exactly the
states it uses; a genuinely queued burst still ramps in one jump.
TestContentionBatchPrewarms is rewritten to queue real waiters instead of
relying on the ceiling to imply demand, and TestNoPrewarmWithoutWaiters pins the
production shape: two overlapping requests against a 76-wide adapter must create
exactly 2 states.
Production after the first deploy showed the openai pool stuck at 9 states with
peak_in_use=1: a startup burst grew it, and then it never shrank again. The
shrink grace counter was reset by every CHECKOUT, so on a gateway that always
has a request in flight the counter never reached shrinkGraceRounds and the
burst's leftover states were pinned indefinitely — the same monotonic-growth
behaviour this series set out to remove, just with an extra step.
Grace is now reset by GROWTH (a miss that had to boot a state), which is the
actual signal that capacity is short. Ordinary sequential traffic no longer
defers reclaim, while two consecutive quiet-ish rounds are still required so a
gap between two bursts does not tear the pool down.
TestCheckoutResetsGrace asserted the old behaviour and is replaced by
TestGrowthResetsGrace (sequential traffic must NOT defer, growth must) plus
TestBurstLeftoverIsReclaimedUnderSteadyTraffic, which reproduces the production
shape: 9 concurrent holds, then one request per janitor round, and the pool must
still fall back to the resident floor.
sensenova: the deployed /etc/llmsproxy/adapters/sensenova.lua carried a fix that
never made it back into the repo — sensenova-6.8-flash-lite reports its chain of
thought in `reasoning` rather than `reasoning_content`, in both the single-shot
response and the stream deltas. Without this the model's output looked empty.
Repo and deployment now match byte for byte.
trae: the adapter was in use on this deployment but untracked, so a fresh
install had no way to serve the trae source.
ConfigureConcurrency() set each adapter pool's target to the sum of
max_concurrent over its sources (108 on this deployment) and `created` only
ever went UP: once a Lua state was booted it was parked forever, so a
long-running gateway's resident state count was a high-water mark of all
traffic it had ever seen, never of what it currently needs.
Pools are now sized from three live inputs:
* the adapter's MAX CONCURRENCY (sum of max_concurrent) is a ceiling, not a
preallocation, and it sets the growth step:
growStep = clamp(ceil(maxW/8), 1, 8). A 64-wide adapter warms 8 states at
once on a spike; an 8-wide one creeps up one at a time.
* the LIVE CONNECTION COUNT (inUse, i.e. checked-out states) sets the shrink
step: shrinkStep = clamp(ceil(excess/(1+inUse)), 1, excess). With no
connections the slack collapses in a single round; a busy adapter gives up
one state per round so the hot path keeps its warm states.
* how many states already exist (created / len(idle)) decides how much room
is left to grow and how much can be reclaimed.
Batch prewarm only fires on genuine contention (a miss while every existing
state is checked out), so a single sequential caller keeps reusing one state
rather than burning a whole grow step on a cold start. A single VM-level
janitor goroutine (not one per adapter) reclaims idle states every 30s, and
shrinkGraceRounds=2 plus idleHeadroom=1 keep a gap between requests from being
mistaken for the end of a load period; any checkout resets the grace counter.
`created` now decrements on reclaim and on shutdown, and release() closes a
state outright when the ceiling was lowered underneath it, so shrinking
max_concurrent in the config gives memory back immediately instead of parking
orphans until restart.
PoolStats() exposes created/idle/in_use/waiting/max/resident/grow_step/
shrink_step/peak_in_use for the status API.
Measured on the test instance (single mock source, max_concurrent=64):
idle created=1; 50 concurrent requests -> created=10 (peak_in_use=5, ceiling
respected); after 95s of silence -> created=1. On production after deploy: 13
adapters, 1 resident Lua state total with ceilings up to 76.
Live testing across the zen pool showed models report
prompt_tokens_details.cached_tokens even when the hit count is 0 (e.g.
nemotron-3-ultra-free returns cached_tokens:0, audio_tokens:0,
cache_write_tokens:0). The previous >0 guard dropped those objects, so a
cache-enabled upstream looked identical to one without cache support.
- types: PromptTokensDetails.CachedTokens always emitted (drop inner
omitempty) so clients see cached_tokens:0 explicitly; dsh reads it as
a 0% hit instead of 'no data'
- adapters (9): forward prompt_tokens_details whenever the upstream
provides it (presence check instead of >0)
- Req: add cache_reported flag set when usage carried cache accounting;
WebUI shows an amber 0% tag for reported-but-missed rows and keeps
the em-dash only for sources that never report cache data
Live testing proved both sensenova and zen DO return cache fields:
- zen laguna-s-2.1-free: usage.prompt_tokens_details.cached_tokens = 32
(real hit), plus cache_write_tokens/audio_tokens
- sensenova glm-5.2: prompt_tokens_details.cached_tokens present (0 on
short prompts)
The previous round only patched deepseek/openai/anthropic/gemini.lua;
sensenova/opencode (localzen!) and the other adapters still dropped them.
- sensenova/opencode/groq/mistral/github/kimicode: stream + response
cache passthrough (same pattern as openai.lua)
- agentrouter: response passthrough + NEW stream usage forwarding (it
previously dropped the terminal usage-only chunk entirely)
- ollama skipped intentionally: its native API has no cache fields
Verified end-to-end through the gateway: localzen/laguna-s-2.1-free now
returns prompt_tokens_details.cached_tokens=32 to clients, and the request
record carries cache_hit_tokens (both chat and stream paths).
dsh displays cache-hit %, but llmsproxy dropped every upstream's cache
fields — deepseek prompt_cache_hit_tokens, OpenAI prompt_tokens_details.
cached_tokens, anthropic cache_read_input_tokens, gemini cachedContentTokenCount.
Changes:
- TokenUsage: add PromptTokensDetails (with CachedTokens) + PromptCacheHit/Miss
- MarshalJSON: emit prompt_tokens_details.cached_tokens (OpenAI v2 standard)
and prompt_cache_hit/miss_tokens (DeepSeek legacy) — dsh reads the former
first, falls back to the latter
- mergeUsage: preserve cache fields across stream chunks
- standardSSEChunk: parse the upstream raw prompt_tokens_details too
- deepseek.lua: forward prompt_cache_hit/miss_tokens + create
prompt_tokens_details from them
- openai.lua: forward prompt_tokens_details.cached_tokens and legacy
prompt_cache_hit/miss_tokens; normalize legacy hits into the standard
object so dsh sees them regardless of upstream format
- anthropic.lua: map cache_read_input_tokens → prompt_tokens_details
- gemini.lua: map cachedContentTokenCount → prompt_tokens_details
Every upstream formats errors differently, which is adapter territory:
the protocol gains an optional transform_error(status, body) hook and all
built-in adapters implement their own envelope parsing (zen free-pool
labels, anthropic/gemini/ollama/mistral shapes, sensenova quota notes,
agentrouter WAF pages). The core keeps a single uniform fallback: when no
hook yields a reason clients get "api error <status>: unknown error" and
the raw body goes to server logs only.
The gateway hardcoded "stop" on every terminating stream chunk, so
tool-call rounds reported finish_reason=stop and length caps were
invisible to clients. UnifiedChunk now carries finish_reason; adapters
emit it (with empty-string finish reasons like sensenova treated as
non-terminal), standardSSEChunk passes it through for un-adapted
upstreams, [DONE] no longer emits a duplicate reason-less done chunk,
and both streaming paths emit the real reason with "stop" as fallback.
Also vendor sensenova/agentrouter adapters into the repo: they were
WebUI-only uploads and a deploy sync silently removed them while live
AUTO-chain slots still referenced them.
zen fingerprints clients via UA + x-opencode-client/session/request/project
headers; requests without them are routed as anonymous and fail with a
single-chunk network_error stream (tools+stream) or 503 Endpoint is
unavailable (non-stream), which the adapter laundered into empty-but-valid
replies. Derive per-request session/request ids from build_headers meta
(sandbox has no os/math), bump UA to the real client format, map zen
reasoning field to reasoning_content, and set stream_options.include_usage.
The prior usage-passthrough fix only covered openai/opencode; the same
empty-choices+usage drop bug remained in the 5 sibling OpenAI-compatible
adapters, and non-OpenAI providers (anthropic/gemini/ollama) never surfaced
streaming usage at all.
- deepseek/github/groq/kimicode/mistral: preserve usage on empty-choices
chunks and attach it to normal chunks (same pattern as openai.lua)
- anthropic: emit usage from message_start (prompt) and message_delta
(completion); gateway merges split usage additively
- gemini: read usageMetadata in the stream path
- ollama: fix non-streaming key (usage -> token_usage, matches
UnifiedResponse json tag) and read prompt_eval_count/eval_count;
surface counts from the done stream chunk
- gateway: mergeUsage combines usage across chunks (non-zero fields win,
total recomputed from prompt+completion) so split usage doesn't lose
the prompt half; single-chunk case (OpenAI) preserved exactly
- usage-only chunks: done=false (no redundant terminal stop), matching
the Go fallback standardSSEChunk
Streaming responses now carry the upstream's real token usage instead of
gateway estimates:
- UnifiedChunk gains an optional Usage field; adapters (opencode, openai)
extract usage from upstream stream chunks (including the final chunk with
empty choices) and pass it through.
- standardSSEChunk preserves usage for passthrough adapters.
- Gateway emits the exact usage in the final stream chunk when available,
falling back to estimates only when the upstream provided none.
Non-streaming usage was already fixed to emit OpenAI-standard keys.
