diff --git a/cmd/gui/renderer/app.js b/cmd/gui/renderer/app.js
index 2f80252..7696700 100644
--- a/cmd/gui/renderer/app.js
+++ b/cmd/gui/renderer/app.js
@@ -44,6 +44,9 @@ const state = {
   pendingTools: [],
   eventSource: null,
   chatFinalIdx: -1,
+  chatOffset: 0, // 分段历史：当前已加载消息在服务端全量中的起始下标
+  chatTotal: 0, // 服务端历史总条数
+  chatHasMore: false, // 是否还有更早历史可向上加载
   sseLastEventID: "", // 最近一次 SSE 事件 id，断线重连时随 Last-Event-ID 头回传
   lang: localStorage.getItem("ha-lang") || "zh",
   connections: [],
@@ -1340,6 +1343,10 @@ function renderChat() {
       () => {
         state.chatStick =
           msgsEl.scrollHeight - msgsEl.scrollTop - msgsEl.clientHeight < 80;
+        // 触顶（近顶部 60px）且服务端还有更早历史 → 向上懒加载下一页
+        if (msgsEl.scrollTop < 60 && state.chatHasMore) {
+          loadOlderChat();
+        }
       },
       { passive: true },
     );
@@ -2351,13 +2358,26 @@ function switchChatPanel(tab, el) {
   if (tab === "knowledge") searchKnowledgeChat();
 }
 
+// 首屏分段加载条数：只拉最新 N 条，向上滚动触顶时再拉更早的。
+var CHAT_PAGE_SIZE = 40;
+
 async function loadChatHistory() {
   try {
-    var data = await api("/chat/history");
+    var data = await api("/chat/history?limit=" + CHAT_PAGE_SIZE);
     if (data && data.messages) {
       state.messages = data.messages;
+      state.chatOffset = typeof data.offset === "number" ? data.offset : 0;
+      state.chatTotal =
+        typeof data.total === "number" ? data.total : data.messages.length;
+      state.chatHasMore = !!data.has_more;
       if (window.homeagent && window.homeagent.log)
-        window.homeagent.log("history: loaded " + data.messages.length);
+        window.homeagent.log(
+          "history: loaded " +
+            data.messages.length +
+            "/" +
+            state.chatTotal +
+            (state.chatHasMore ? " (has more)" : ""),
+        );
     } else if (window.homeagent && window.homeagent.log) {
       window.homeagent.log("history: no messages field");
     }
@@ -2367,35 +2387,101 @@ async function loadChatHistory() {
   }
 }
 
+// loadOlderChat 向上翻页：拉 offset 之前的一页，前置到 messages 头部。
+// 保持滚动位置补偿，避免视口跳动。
+var _loadingOlder = false;
+async function loadOlderChat() {
+  if (_loadingOlder || !state.chatHasMore) return;
+  _loadingOlder = true;
+  var msgsEl = document.getElementById("chat-msgs");
+  var prevH = msgsEl ? msgsEl.scrollHeight : 0;
+  var prevTop = msgsEl ? msgsEl.scrollTop : 0;
+  try {
+    var before = state.chatOffset || 0;
+    if (before <= 0) {
+      state.chatHasMore = false;
+      return;
+    }
+    var data = await api(
+      "/chat/history?limit=" + CHAT_PAGE_SIZE + "&before=" + before,
+    );
+    if (data && data.messages && data.messages.length) {
+      state.messages = data.messages.concat(state.messages);
+      state.chatOffset =
+        typeof data.offset === "number" ? data.offset : 0;
+      state.chatHasMore = !!data.has_more;
+      state.chatStick = false;
+      rerenderChatIfActive();
+      if (msgsEl) {
+        msgsEl.scrollTop = prevTop + (msgsEl.scrollHeight - prevH);
+      }
+      if (window.homeagent && window.homeagent.log)
+        window.homeagent.log(
+          "history: older " + data.messages.length + " (offset=" + state.chatOffset + ")",
+        );
+    } else {
+      state.chatHasMore = false;
+    }
+  } catch (e) {
+  } finally {
+    _loadingOlder = false;
+  }
+}
+
 // syncChatFromHistory 增量同步：对比服务端历史，仅追加新消息 DOM 节点，
 // 不重建已有消息 → 无闪烁。用于 SSE 断连恢复期间的轮询兜底（跨渠道消息补偿）。
 function syncChatFromHistory() {
-  return api("/chat/history").then(function (data) {
-    if (!data || !data.messages || data.messages.length === 0) return;
-    var serverMsgs = data.messages;
-    var localMsgs = state.messages;
-    if (localMsgs.length === 0) {
-      state.messages = serverMsgs;
-      rerenderChatIfActive();
-      return;
-    }
-    if (serverMsgs.length <= localMsgs.length) {
+  return api("/chat/history?limit=" + CHAT_PAGE_SIZE)
+    .then(function (data) {
+      if (!data || !data.messages || data.messages.length === 0) return;
+      var serverMsgs = data.messages;
+      var localMsgs = state.messages;
+      // 首次加载（空列表）→ 全量赋值
+      if (localMsgs.length === 0) {
+        state.messages = serverMsgs;
+        state.chatOffset = typeof data.offset === "number" ? data.offset : 0;
+        state.chatHasMore = !!data.has_more;
+        rerenderChatIfActive();
+        return;
+      }
+      // 分段拉取只回传最新页，本地可能已向上翻页加载更多，
+      // 因此不能用长度比较，改用末尾内容比对 + 重叠区对齐。
       var lastLocal = localMsgs[localMsgs.length - 1];
       var lastServer = serverMsgs[serverMsgs.length - 1];
       var localContent = lastLocal.content || lastLocal.Content || "";
       var serverContent = lastServer.content || lastServer.Content || "";
-      if (lastServer.role === "assistant" && localContent !== serverContent && serverContent) {
-        lastLocal.content = serverContent;
-        if (lastServer.ReasoningContent) lastLocal.reasoning_content = lastServer.ReasoningContent;
+      // 末尾一致 → 无新增
+      if (localContent === serverContent) return;
+      // 寻找重叠点（本地末尾 k 条在服务端页中的位置）
+      var overlap = -1;
+      var maxK = Math.min(3, serverMsgs.length - 1, localMsgs.length - 1);
+      for (var k = maxK; k >= 1; k--) {
+        var sMsg = serverMsgs[serverMsgs.length - 1 - k];
+        var lMsg = localMsgs[localMsgs.length - 1 - k];
+        if (
+          lMsg &&
+          sMsg &&
+          (lMsg.content || "") === (sMsg.content || "") &&
+          lMsg.role === sMsg.role
+        ) {
+          overlap = k;
+          break;
+        }
+      }
+      if (overlap >= 0) {
+        var newMsgs = serverMsgs.slice(serverMsgs.length - overlap);
+        if (newMsgs.length === 0) return;
+        Array.prototype.push.apply(state.messages, newMsgs);
+        rerenderChatIfActive();
+      } else {
+        // 无重叠点（本地领先过多/已失同步）→ 安全退化为全量刷新最新页
+        state.messages = serverMsgs;
+        state.chatOffset = typeof data.offset === "number" ? data.offset : 0;
+        state.chatHasMore = !!data.has_more;
         rerenderChatIfActive();
       }
-      return;
-    }
-    // 服务端消息更多 → 仅追加新消息对象（不重建已有 DOM）
-    var newMsgs = serverMsgs.slice(localMsgs.length);
-    Array.prototype.push.apply(state.messages, newMsgs);
-    rerenderChatIfActive();
-  }).catch(function () {});
+    })
+    .catch(function () {});
 }
 
 async function loadTerminals() {
diff --git a/internal/plugins/webui/dashboard.html b/internal/plugins/webui/dashboard.html
index a25ec7f..e6efb0e 100644
--- a/internal/plugins/webui/dashboard.html
+++ b/internal/plugins/webui/dashboard.html
@@ -2258,6 +2258,9 @@
         pendingTools: [],
         eventSource: null,
         chatFinalIdx: -1,
+        chatOffset: 0, // 分段历史：当前已加载消息在服务端全量中的起始下标
+        chatTotal: 0, // 服务端历史总条数
+        chatHasMore: false, // 是否还有更早历史可向上加载
         lang: localStorage.getItem("ha-lang") || "zh",
       };
 
@@ -2996,6 +2999,10 @@
               state.chatStick =
                 msgsEl.scrollHeight - msgsEl.scrollTop - msgsEl.clientHeight <
                 80;
+              // 触顶（近顶部 60px）且服务端还有更早历史 → 向上懒加载下一页
+              if (msgsEl.scrollTop < 60 && state.chatHasMore) {
+                loadOlderChat();
+              }
             },
             { passive: true },
           );
@@ -4199,47 +4206,116 @@
         if (tab === "context") queryMemoryContext();
       }
 
+      // 首屏分段加载条数：只拉最新 N 条，向上滚动触顶时再拉更早的。
+      var CHAT_PAGE_SIZE = 40;
       async function loadChatHistory() {
         try {
-          var data = await api("/chat/history");
-          if (data && data.messages) state.messages = data.messages;
+          var data = await api("/chat/history?limit=" + CHAT_PAGE_SIZE);
+          if (data && data.messages) {
+            state.messages = data.messages;
+            state.chatOffset = typeof data.offset === "number" ? data.offset : 0;
+            state.chatTotal = typeof data.total === "number" ? data.total : data.messages.length;
+            state.chatHasMore = !!data.has_more;
+          }
         } catch (e) {}
       }
 
+      // loadOlderChat 向上翻页：拉 offset 之前的一页，前置到 messages 头部。
+      // 保持滚动位置（插入前后 scrollHeight 差值补偿），避免视口跳动。
+      var _loadingOlder = false;
+      async function loadOlderChat() {
+        if (_loadingOlder || !state.chatHasMore) return;
+        _loadingOlder = true;
+        var msgsEl = document.getElementById("chat-msgs");
+        var prevH = msgsEl ? msgsEl.scrollHeight : 0;
+        var prevTop = msgsEl ? msgsEl.scrollTop : 0;
+        try {
+          var before = state.chatOffset || 0;
+          if (before <= 0) {
+            state.chatHasMore = false;
+            return;
+          }
+          var data = await api(
+            "/chat/history?limit=" + CHAT_PAGE_SIZE + "&before=" + before,
+          );
+          if (data && data.messages && data.messages.length) {
+            state.messages = data.messages.concat(state.messages);
+            state.chatOffset = typeof data.offset === "number" ? data.offset : 0;
+            state.chatHasMore = !!data.has_more;
+            state.chatStick = false; // 向上翻页时不自动粘底
+            renderChat();
+            if (msgsEl) {
+              // 补偿新增内容高度，保持用户视觉位置不变
+              msgsEl.scrollTop = prevTop + (msgsEl.scrollHeight - prevH);
+            }
+          } else {
+            state.chatHasMore = false;
+          }
+        } catch (e) {
+        } finally {
+          _loadingOlder = false;
+        }
+      }
+
       // syncChatFromHistory 增量同步：对比服务端历史，仅追加新消息 DOM 节点，
       // 不重建已有消息 → 无闪烁。用于 SSE 断连恢复期间的轮询兜底。
       function syncChatFromHistory() {
-        return api("/chat/history").then(function (data) {
+        return api("/chat/history?limit=" + CHAT_PAGE_SIZE).then(function (data) {
           if (!data || !data.messages || data.messages.length === 0) return;
           var serverMsgs = data.messages;
           var localMsgs = state.messages;
-          // 空历史 → 全量加载（首次同步）
+          // 首次加载（空列表）→ 全量赋值
           if (localMsgs.length === 0) {
             state.messages = serverMsgs;
+            state.chatOffset = typeof data.offset === "number" ? data.offset : 0;
+            state.chatHasMore = !!data.has_more;
             rerenderChat(true);
             return;
           }
-          // 无新增消息 → 检查最后一条是否被改写
-          if (serverMsgs.length <= localMsgs.length) {
-            var lastLocal = localMsgs[localMsgs.length - 1];
-            var lastServer = serverMsgs[serverMsgs.length - 1];
-            var localContent = lastLocal.content || lastLocal.Content || "";
-            var serverContent = lastServer.content || lastServer.Content || "";
+          // 以「本地末尾 vs 服务端末尾」做依据。分段拉取只会回传最新页，
+          // 本地已向上翻页加载了更早内容，长度上本地可能 ≥ 服务端。
+          var lastLocal = localMsgs[localMsgs.length - 1];
+          var lastServer = serverMsgs[serverMsgs.length - 1];
+          var localContent = lastLocal.content || lastLocal.Content || "";
+          var serverContent = lastServer.content || lastServer.Content || "";
+          // 服务端末尾和本地末尾内容相同 → 无新增（可能裁剪过的轻量项因不含 tool_calls args/result，
+          // 依赖 content+role+time 足以识别）
+          if (localContent === serverContent) {
+            // 检查是否本地末尾被服务端改写（如 stage 插件改写最终文本）
             if (lastServer.role === "assistant" && localContent !== serverContent && serverContent) {
               lastLocal.content = serverContent;
               if (lastServer.ReasoningContent) lastLocal.reasoning_content = lastServer.ReasoningContent;
-              // 仅更新最后一条消息 DOM，不全量重建
-              var msgsEl = document.getElementById("chat-msgs");
-              if (msgsEl && msgsEl.lastElementChild) {
-                var el = msgsEl.lastElementChild;
-                var textEl = el.querySelector(".msg-bubble .text");
-                if (textEl) textEl.innerHTML = renderMd(serverContent);
+              var msgsEl0 = document.getElementById("chat-msgs");
+              if (msgsEl0 && msgsEl0.lastElementChild) {
+                var textEl0 = msgsEl0.lastElementChild.querySelector(".msg-bubble .text");
+                if (textEl0) textEl0.innerHTML = renderMd(serverContent);
               }
             }
             return;
           }
-          // 有新增消息：追加到 state.messages + DOM（仅追加节点，不触碰已有）
-          var newMsgs = serverMsgs.slice(localMsgs.length);
+          // 有新增消息：服务端末尾与本地末尾不同 → 找出本地新增的差异，追加到 DOM。
+          // 服务端只回传最新页，我们需在 localMsgs 末尾取交集起点（本地末尾在服务端页的位置）
+          // 简化：取服务端末尾3条与本地末尾3条比对，找到重含点的下标；若找不到全量替换最新页。
+          var overlap = -1;
+          for (var k = Math.min(3, serverMsgs.length - 1, localMsgs.length - 1); k >= 1; k--) {
+            var sMsg = serverMsgs[serverMsgs.length - 1 - k];
+            var lMsg = localMsgs[localMsgs.length - 1 - k];
+            if (lMsg && (lMsg.content || "") === (sMsg.content || "") && lMsg.role === sMsg.role) {
+              overlap = k;
+              break;
+            }
+          }
+          var newMsgs;
+          if (overlap >= 0) {
+            // 本地末尾 overlap 条在服务端页里，之后的部分是新增
+            newMsgs = serverMsgs.slice(serverMsgs.length - overlap);
+            if (newMsgs.length === 0) return; // 无新增（内容改写走上面的分支）
+          } else {
+            // 找不到重含点（本地领先太多，超出服务端页）→ 无法精确差异，安全退化为全量刷新
+            state.messages = serverMsgs;
+            rerenderChat(true);
+            return;
+          }
           var msgsEl = document.getElementById("chat-msgs");
           if (msgsEl) {
             var aiAvatar = '<img src="/mascot.webp" alt="小宅">';
diff --git a/internal/plugins/webui/handler.go b/internal/plugins/webui/handler.go
index 95ecde6..205c2df 100644
--- a/internal/plugins/webui/handler.go
+++ b/internal/plugins/webui/handler.go
@@ -174,6 +174,9 @@ type ChatToolCall struct {
 	Result interface{} `json:"result,omitempty"`
 	Status string      `json:"status,omitempty"`
 	Plugin string      `json:"plugin,omitempty"`
+	// Truncated 标记本条工具调用的 args/result 已被服务端裁剪（lean 模式），
+	// 前端可据此显示「详情需展开加载」而不是把空结果当成执行失败。
+	Truncated bool `json:"truncated,omitempty"`
 }
 
 type CmdExec struct {
@@ -1382,12 +1385,87 @@ func (h *Handler) addChatMsg(msg ChatMsg) {
 	h.chatMu.Unlock()
 }
 
+// handleChatHistory 返回对话历史，支持分段懒加载。
+//
+// 查询参数（全部可选，省略时保持旧行为=返回全量，向后兼容旧客户端）：
+//   - limit:  返回条数上限（1..maxChatHistory）。带 limit 时默认取「最新的 limit 条」。
+//   - before: 游标，只返回下标 < before 的消息（配合 limit 向上翻页取更早历史）。
+//   - lean:   "1"/"true" 时裁剪重负载字段（tool_calls 的 args/result 置空、reasoning_content 省略），
+//     体积可降约 80%（实测 tool_calls 占 ~72%、reasoning ~12%）。用于移动端/弱网首屏。
+//
+// 响应额外返回 total / offset / has_more，供前端判断是否继续向上加载。
 func (h *Handler) handleChatHistory(w http.ResponseWriter, r *http.Request) {
+	q := r.URL.Query()
+	limit := parseIntDefault(q.Get("limit"), 0)
+	if limit < 0 {
+		limit = 0
+	}
+	if limit > maxChatHistory {
+		limit = maxChatHistory
+	}
+	lean := q.Get("lean") == "1" || strings.EqualFold(q.Get("lean"), "true")
+
 	h.chatMu.Lock()
-	result := make([]ChatMsg, len(h.chatHistory))
-	copy(result, h.chatHistory)
+	total := len(h.chatHistory)
+	// before 游标：默认取到末尾（最新）
+	end := parseIntDefault(q.Get("before"), total)
+	if end < 0 || end > total {
+		end = total
+	}
+	start := 0
+	if limit > 0 && end-limit > 0 {
+		start = end - limit
+	}
+	result := make([]ChatMsg, end-start)
+	copy(result, h.chatHistory[start:end])
 	h.chatMu.Unlock()
-	writeJSON(w, http.StatusOK, map[string]interface{}{"messages": result})
+
+	if lean {
+		result = leanChatMsgs(result)
+	}
+	writeJSON(w, http.StatusOK, map[string]interface{}{
+		"messages": result,
+		"total":    total,
+		"offset":   start,
+		"has_more": start > 0,
+	})
+}
+
+// parseIntDefault 解析十进制整数，失败/空串返回 def。
+func parseIntDefault(s string, def int) int {
+	if s == "" {
+		return def
+	}
+	n, err := strconv.Atoi(s)
+	if err != nil {
+		return def
+	}
+	return n
+}
+
+// leanChatMsgs 裁剪重负载字段用于移动端/弱网首屏：
+// tool_calls 只保留 tool/name/status/plugin（args/result 置 nil 并标记 truncated），
+// reasoning_content 整体省略。前端需要完整内容时按 before/limit 拉非 lean 分段。
+func leanChatMsgs(in []ChatMsg) []ChatMsg {
+	out := make([]ChatMsg, len(in))
+	for i, m := range in {
+		m.ReasoningContent = ""
+		if len(m.ToolCalls) > 0 {
+			tcs := make([]ChatToolCall, len(m.ToolCalls))
+			for j, tc := range m.ToolCalls {
+				tcs[j] = ChatToolCall{
+					Tool:      tc.Tool,
+					Name:      tc.Name,
+					Status:    tc.Status,
+					Plugin:    tc.Plugin,
+					Truncated: true,
+				}
+			}
+			m.ToolCalls = tcs
+		}
+		out[i] = m
+	}
+	return out
 }
 
 // handleChatFile 处理用户经 webui 上传文件并附带消息注入 agent。