feat: restore source parity and harden agent runtime

backend-fastapi/ai_platform/nodes/builtin/dialog_nodes.py

@@ -570,7 +570,7 @@ class ConfirmNode(BaseNode):
         
         return False, False
     
-    def _llm_intent_recognition(self, user_input: str, context_content: str, model_id: str) -> bool:
+    def _llm_intent_recognition(self, user_input: str, context_content: str, model_id: str, db_session=None) -> bool:
         """
         使用 LLM 进行意图识别
         
@@ -595,7 +595,7 @@ class ConfirmNode(BaseNode):
 
 请判断用户意图："""
 
-            llm_service = LLMService()
+            llm_service = LLMService(db_session)
             response = llm_service.chat(
                 model_id=model_id,
                 messages=[
@@ -639,7 +639,9 @@ class ConfirmNode(BaseNode):
             if not matched and use_llm_intent and llm_model_id:
                 # 关键词未匹配，使用 LLM 意图识别
                 logger.info(f'关键词未匹配，使用 LLM 意图识别: {user_input}')
-                confirmed = self._llm_intent_recognition(str(user_input), content, llm_model_id)
+                confirmed = self._llm_intent_recognition(
+                    str(user_input), content, llm_model_id, context.db_session
+                )
             elif not matched:
                 # 关键词未匹配且未启用 LLM，默认为取消
                 logger.info(f'关键词未匹配，默认取消: {user_input}')

backend-fastapi/ai_platform/nodes/builtin/llm_node.py

@@ -313,6 +313,7 @@ class LLMNode(BaseNode):
                 messages.append({'role': 'user', 'content': user_prompt})
             
             llm_service = LLMService(context.db_session)
+            model_id = llm_service.resolve_chat_model_id_sync(model_id)
             output_var = self.config.get('output_variable', 'llm_response')
             
             # 根据输出模式选择执行方式

backend-fastapi/ai_platform/services/llm_service.py

@@ -64,6 +64,26 @@ class LLMService:
 
         logger.info("No model_id supplied, fallback to default chat model %s", model.id)
         return str(model.id)
+
+    @staticmethod
+    def _run_async_sync(coro):
+        import asyncio
+        import concurrent.futures
+
+        try:
+            loop = asyncio.get_event_loop()
+        except RuntimeError:
+            return asyncio.run(coro)
+
+        if loop.is_running():
+            with concurrent.futures.ThreadPoolExecutor() as executor:
+                return executor.submit(asyncio.run, coro).result()
+
+        return loop.run_until_complete(coro)
+
+    def resolve_chat_model_id_sync(self, model_id: Optional[str]) -> str:
+        """同步解析 chat 模型，供同步节点和流式生成器复用。"""
+        return self._run_async_sync(self.resolve_chat_model_id(model_id))
     
     async def _get_provider_async(self, model_id: str) -> tuple:
         """
@@ -272,6 +292,34 @@ class LLMService:
             return provider.chat(llm_messages, config)
         except Exception as exc:
             raise RuntimeError(self._format_provider_error(exc, provider, model_name)) from exc
+
+    def chat(
+        self,
+        model_id: Optional[str],
+        messages: List[Dict[str, str]],
+        temperature: float = 0.7,
+        max_tokens: int = 2048,
+        tools: List[Dict] = None,
+        tool_choice: str = 'auto',
+        **kwargs
+    ) -> LLMResponse:
+        """
+        同步对话入口。
+
+        旧节点仍会调用该方法；这里统一走 provider 解析和默认 chat 模型兜底，
+        避免不同节点各自处理 model_id 为空的情况。
+        """
+        provider, model_name = self._run_async_sync(self._get_provider_async(model_id))
+        return self.chat_with_provider(
+            provider=provider,
+            model_name=model_name,
+            messages=messages,
+            temperature=temperature,
+            max_tokens=max_tokens,
+            tools=tools,
+            tool_choice=tool_choice,
+            **kwargs,
+        )
     
     def _convert_messages(self, messages: List[Dict]) -> List[LLMMessage]:
         """转换消息格式，支持 tool 消息"""