feat(llm): implement structured JSON output for assistant responses

api/chat_service.py

@@ -1,4 +1,5 @@
-from typing import Dict, Any, Optional
+from typing import Dict, Any, Optional, Tuple
+import json
 import base64
 import threading
 from datetime import datetime
@@ -31,6 +32,37 @@ class ChatService:
         logger.info(f"Initializing Mem0 integration for user: {self.user_id}")
         self._initialized = True
     
+    def _parse_assistant_output(self, raw_text: Any) -> Tuple[str, str, Optional[str]]:
+        """Parse model JSON output into reply/action while reporting parse issues."""
+        if raw_text is None:
+            return "", "无", "empty_response"
+
+        if not isinstance(raw_text, str):
+            raw_text = str(raw_text)
+
+        stripped = raw_text.strip()
+        if not stripped:
+            return "", "无", "empty_response"
+
+        try:
+            payload = json.loads(stripped)
+        except json.JSONDecodeError as exc:
+            logger.warning("Failed to parse assistant JSON response: %s", exc)
+            return stripped, "无", f"json_decode_error: {exc}"
+        except Exception as exc:
+            logger.warning("Unexpected error parsing assistant response: %s", exc)
+            return stripped, "无", f"parse_error: {exc}"
+
+        reply = str(payload.get("reply", "")).strip()
+        action = str(payload.get("action", "")).strip()
+
+        if not reply:
+            reply = stripped
+        if not action:
+            action = "无"
+
+        return reply, action, None
+
     def chat(self, user_input: str, include_audio: bool = True) -> Dict[str, Any]:
         """处理用户输入并返回回复（包含音频）"""
         if not self._initialized:
@@ -51,13 +83,20 @@ class ChatService:
                 }
             
             assistant_response = result["response"]
+            reply, action, parse_error = self._parse_assistant_output(assistant_response)
+            if parse_error:
+                logger.warning("Assistant output parse_error=%s", parse_error)
+
+            response_text = reply.strip() if isinstance(reply, str) else str(reply)
+            if not response_text:
+                response_text = assistant_response.strip() if isinstance(assistant_response, str) else str(assistant_response)
             
             # Step 2: Generate audio if requested
             audio_data = None
             audio_error = None
-            if include_audio:
+            if include_audio and response_text:
                 try:
-                    success, message, base64_audio = text_to_speech(assistant_response, self.user_id)
+                    success, message, base64_audio = text_to_speech(response_text, self.user_id)
                     if success and base64_audio:
                         audio_data = base64_audio
                     else:
@@ -68,7 +107,9 @@ class ChatService:
             # Step 3: Prepare response
             response_data = {
                 "success": True,
-                "response": assistant_response,
+                "response": response_text,
+                "action": action,
+                "parse_error": parse_error,
                 "user_id": self.user_id
             }
             

api/main.py

@@ -32,6 +32,8 @@ class ChatRequest(BaseModel):
 class ChatResponse(BaseModel):
     success: bool
     response: Optional[str] = None
+    action: Optional[str] = None
+    parse_error: Optional[str] = None
     tokens: Optional[int] = None
     user_id: str
     error: Optional[str] = None