diff --git a/graph_memory_tui/core/prompts/templates/system_prompt.md b/graph_memory_tui/core/prompts/templates/system_prompt.md
index ae937a0..0d775eb 100644
--- a/graph_memory_tui/core/prompts/templates/system_prompt.md
+++ b/graph_memory_tui/core/prompts/templates/system_prompt.md
@@ -2,6 +2,33 @@
 
 你是TrulyMEM，一个拥有长期记忆能力的AI助手。
 
+## ⚠️ 工具调用限制（严格执行）
+
+**每轮对话工具调用次数限制：**
+
+### 人设图工具
+- `memory_recall` (查询人设): **最多1次**
+- `persona_update` / `persona_clear`: **最多1次**
+
+### 工作记忆链工具
+- `memory_recall` (查询任务链): **最多10次**
+- `task_create` / `task_set_state` / `task_delete` / `task_link_info`: **最多5次**
+
+### 一般记忆工具
+- `memory_recall` (其他查询): **最多20次**
+- `memory_commit` / `memory_purge`: **最多10次**
+
+### 其他工具
+- `memory_introspect` / `memory_archive` / `memory_cleanup`: **最多1次**
+
+**重要提示：**
+- 单次 `memory_recall` 可以同时查询多个关键词（用逗号分隔）
+- 达到限制后立即停止工具调用，直接回复用户
+- 不要重复调用相同的工具
+- 优先合并查询，减少调用次数
+
+---
+
 ## ⚠️ 关键约束：无传统上下文系统
 
 **重要**: 你没有传统的对话上下文系统（没有消息历史数组）。
diff --git a/graph_memory_tui/services/chat_service.py b/graph_memory_tui/services/chat_service.py
index 039e543..e8b1853 100644
--- a/graph_memory_tui/services/chat_service.py
+++ b/graph_memory_tui/services/chat_service.py
@@ -1,6 +1,7 @@
 """聊天服务"""
 
 import asyncio
+import json
 from datetime import datetime
 from typing import AsyncIterator, TYPE_CHECKING
 from ..core.imports import GraphMemoryClient
@@ -14,6 +15,9 @@ if TYPE_CHECKING:
 class ChatService:
     """聊天业务服务"""
 
+    # 工具调用限制
+    MAX_TOOL_ITERATIONS = 5  # 最多5轮工具调用
+
     def __init__(
         self,
         graph: "Neo4jGraph",
@@ -34,29 +38,39 @@ class ChatService:
         }
 
         try:
-            # 2. 使用流式API调用
-            accumulated_content = ""
-            tool_calls_data = []
-            
-            # 流式处理响应
-            async for chunk in self._call_api_stream_async(user_input):
-                # 处理内容增量
-                if chunk.get("content_delta"):
-                    accumulated_content += chunk["content_delta"]
-                    yield {
-                        "type": "content_delta",
-                        "content": accumulated_content
-                    }
-                
-                # 处理工具调用
-                if chunk.get("tool_calls"):
-                    tool_calls_data = chunk["tool_calls"]
-            
-            # 3. 处理工具调用
-            tool_calls = None
-            tool_results = None
+            # 2. 处理工具调用循环
+            iteration = 0
+            current_input = user_input
+            all_tool_calls = []
+            all_tool_results = []
+            final_content = ""
 
-            if tool_calls_data:
+            while iteration < self.MAX_TOOL_ITERATIONS:
+                iteration += 1
+                
+                # 流式处理响应
+                accumulated_content = ""
+                tool_calls_data = []
+                
+                async for chunk in self._call_api_stream_async(current_input):
+                    # 处理内容增量
+                    if chunk.get("content_delta"):
+                        accumulated_content += chunk["content_delta"]
+                        yield {
+                            "type": "content_delta",
+                            "content": accumulated_content
+                        }
+                    
+                    # 处理工具调用
+                    if chunk.get("tool_calls"):
+                        tool_calls_data = chunk["tool_calls"]
+
+                # 如果没有工具调用，结束循环
+                if not tool_calls_data:
+                    final_content = accumulated_content
+                    break
+
+                # 处理工具调用
                 tool_calls = []
                 tool_results = []
 
@@ -87,12 +101,18 @@ class ChatService:
                         "log_entry": log_entry
                     }
 
-            # 4. 返回最终回复
+                all_tool_calls.extend(tool_calls)
+                all_tool_results.extend(tool_results)
+
+                # 准备下一轮输入（工具结果）
+                current_input = self._format_tool_results(tool_calls, tool_results)
+
+            # 3. 返回最终回复
             yield {
                 "type": "assistant_message",
-                "content": accumulated_content,
-                "tool_calls": tool_calls,
-                "tool_results": tool_results
+                "content": final_content,
+                "tool_calls": all_tool_calls if all_tool_calls else None,
+                "tool_results": all_tool_results if all_tool_results else None
             }
 
         except Exception as e:
@@ -102,6 +122,13 @@ class ChatService:
                 "error": str(e)
             }
 
+    def _format_tool_results(self, tool_calls: list[ToolCall], tool_results: list[ToolResult]) -> str:
+        """格式化工具结果为下一轮输入"""
+        results_text = []
+        for tc, tr in zip(tool_calls, tool_results):
+            results_text.append(f"工具 {tc.name} 执行结果:\n{tr.content}")
+        return "\n\n".join(results_text)
+
     async def _call_api_stream_async(self, message: str) -> AsyncIterator[dict]:
         """异步流式调用 API"""
         loop = asyncio.get_event_loop()