diff --git a/main_v2.py b/main_v2.py
index ca2fcf0..6afb527 100644
--- a/main_v2.py
+++ b/main_v2.py
@@ -659,58 +659,45 @@ async def websocket_endpoint(websocket: WebSocket, user_id: str):
                     # 获取对话历史
                     history = conv_service.get_conversation_history(conversation_id, limit=agent_config['agent'].get('max_history', 20))
                     
-                    # 使用流式调用LLM
+                    # 使用非流式调用LLM（简化版本，确保稳定）
                     try:
-                        thinking_buffer = ""
-                        content_buffer = ""
-                        in_thinking = False
+                        # 先发送思考开始
+                        if agent_config['agent'].get('enable_thinking') and enable_thinking:
+                            await websocket.send_json({
+                                "type": "thinking_start",
+                                "conversation_id": conversation_id
+                            })
                         
-                        # 流式生成
-                        async for chunk in llm_service.chat_stream(
+                        # 调用LLM（非流式）
+                        response, thinking_content = await llm_service.chat(
                             messages=history,
                             provider_config=agent_config['provider'],
                             agent_config=agent_config['agent'],
                             enable_thinking=enable_thinking
-                        ):
-                            chunk_type = chunk.get('type', 'content')
-                            chunk_text = chunk.get('text', '')
-                            
-                            if chunk_type == 'thinking':
-                                if not in_thinking:
-                                    in_thinking = True
-                                    await websocket.send_json({
-                                        "type": "thinking_start",
-                                        "conversation_id": conversation_id
-                                    })
-                                thinking_buffer += chunk_text
-                                await websocket.send_json({
-                                    "type": "thinking_stream",
-                                    "conversation_id": conversation_id,
-                                    "text": chunk_text
-                                })
-                            elif chunk_type == 'content':
-                                if in_thinking:
-                                    in_thinking = False
-                                    await websocket.send_json({
-                                        "type": "thinking_end",
-                                        "conversation_id": conversation_id,
-                                        "content": thinking_buffer  # 发送完整思考内容用于折叠
-                                    })
-                                content_buffer += chunk_text
-                                await websocket.send_json({
-                                    "type": "content_stream",
-                                    "conversation_id": conversation_id,
-                                    "text": chunk_text
-                                })
+                        )
                         
-                        # 如果思考没结束就结束了，发送思考结束
-                        if in_thinking:
+                        # 发送思考内容
+                        if thinking_content:
                             await websocket.send_json({
-                                "type": "thinking_end",
+                                "type": "thinking_content",
                                 "conversation_id": conversation_id,
-                                "content": thinking_buffer
+                                "content": thinking_content
                             })
                         
+                        # 发送思考结束
+                        await websocket.send_json({
+                            "type": "thinking_end",
+                            "conversation_id": conversation_id,
+                            "content": thinking_content
+                        })
+                        
+                        # 直接发送完整回复
+                        await websocket.send_json({
+                            "type": "content_stream",
+                            "conversation_id": conversation_id,
+                            "text": response
+                        })
+                        
                         # 发送完成通知
                         await websocket.send_json({
                             "type": "stream_end",
@@ -721,9 +708,9 @@ async def websocket_endpoint(websocket: WebSocket, user_id: str):
                         assistant_msg = conv_service.add_message(
                             conversation_id=conversation.id,
                             role='assistant',
-                            content=content_buffer,
+                            content=response,
                             source='web',
-                            thinking_content=thinking_buffer if thinking_buffer else None,
+                            thinking_content=thinking_content if thinking_content else None,
                             agent_id=current_agent_id,
                             model_used=agent_config['provider'].get('default_model')
                         )
@@ -734,8 +721,8 @@ async def websocket_endpoint(websocket: WebSocket, user_id: str):
                             "message": {
                                 "id": assistant_msg.id,
                                 "role": "assistant",
-                                "content": content_buffer,
-                                "thinking_content": thinking_buffer if thinking_buffer else None,
+                                "content": response,
+                                "thinking_content": thinking_content if thinking_content else None,
                                 "source": "web",
                                 "agent_id": current_agent_id,
                                 "agent_name": agent_config['agent'].get('display_name'),