diff --git a/agent/guideline_middleware.py b/agent/guideline_middleware.py
index 9e0eb9e..f142447 100644
--- a/agent/guideline_middleware.py
+++ b/agent/guideline_middleware.py
@@ -8,15 +8,10 @@ from typing import Any, Callable
 from langchain_core.callbacks import BaseCallbackHandler
 from langchain_core.outputs import LLMResult
 import logging
+import re
 
 logger = logging.getLogger('app')
 
-class ThinkingCallbackHandler(BaseCallbackHandler):
-    """自定义回调处理器，用于将模型响应内容转换为thinking格式"""
-    def on_llm_end(self, response: LLMResult, **kwargs) -> None:
-        """在LLM调用结束后处理响应，将内容转换为thinking格式"""
-        logger.info("Successfully converted response content to thinking format")
-
 class GuidelineMiddleware(AgentMiddleware):
     def __init__(self, bot_id: str, model:BaseChatModel, prompt: str, robot_type: str, language: str, user_identifier: str):
         self.model = model
@@ -98,17 +93,17 @@ Action: Provide concise, friendly, and personified natural responses.
         
         guideline_prompt = self.get_guideline_prompt(convert_to_openai_messages(state['messages']))
         
-        # 创建回调处理器实例
-        thinking_handler = ThinkingCallbackHandler()
-        
         # 使用回调处理器调用模型
         response = self.model.invoke(
             guideline_prompt, 
-            config={"callbacks": [thinking_handler]}
+            config={"callbacks": [BaseCallbackHandler()]}
         )
 
-        response.additional_kwargs["thinking"] = response.content
-
+        # 提取<think>与</think>之间的内容作为thinking
+        
+        match = re.search(r'<think>(.*?)</think>', response.content, re.DOTALL)
+        response.additional_kwargs["thinking"] = match.group(1).strip() if match else response.content
+        
         messages = state['messages']+[response]
         return {
             "messages": messages
@@ -123,16 +118,12 @@ Action: Provide concise, friendly, and personified natural responses.
         # 使用回调处理器调用模型
         response = await self.model.ainvoke(
             guideline_prompt, 
-            config={"callbacks": [ThinkingCallbackHandler()]}
+            config={"callbacks": [BaseCallbackHandler()]}
         )
         
         # 提取<think>与</think>之间的内容作为thinking
-        import re
         match = re.search(r'<think>(.*?)</think>', response.content, re.DOTALL)
-        if match:
-            response.additional_kwargs["thinking"] = match.group(1).strip()
-        else:
-            response.additional_kwargs["thinking"] = response.content
+        response.additional_kwargs["thinking"] = match.group(1).strip() if match else response.content
 
         messages = state['messages']+[response]
         return {