diff --git a/data/chroma/kb_2/714cba61-f48b-47c5-bcb8-eed8ffd400b0/data_level0.bin b/data/chroma/kb_2/714cba61-f48b-47c5-bcb8-eed8ffd400b0/data_level0.bin
index 6730e4e..6630d25 100644
Binary files a/data/chroma/kb_2/714cba61-f48b-47c5-bcb8-eed8ffd400b0/data_level0.bin and b/data/chroma/kb_2/714cba61-f48b-47c5-bcb8-eed8ffd400b0/data_level0.bin differ
diff --git a/data/chroma/kb_2/chroma.sqlite3 b/data/chroma/kb_2/chroma.sqlite3
index caa7dfa..d450540 100644
Binary files a/data/chroma/kb_2/chroma.sqlite3 and b/data/chroma/kb_2/chroma.sqlite3 differ
diff --git a/th_agenter/services/workflow_engine.py b/th_agenter/services/workflow_engine.py
index 0321df4..f786961 100644
--- a/th_agenter/services/workflow_engine.py
+++ b/th_agenter/services/workflow_engine.py
@@ -654,6 +654,9 @@ class WorkflowEngine:
         """执行LLM节点"""
         config = input_data.get('node_config', {})
         
+        # 标记是否已经使用了默认模型（用于错误时决定是否回退）
+        used_default_model = False
+        
         # 获取LLM配置
         model_id = config.get('model_id')
         if not model_id:
@@ -669,9 +672,9 @@ class WorkflowEngine:
                     result = await self.session.execute(
                         select(LLMConfig).where(LLMConfig.model_name == model_value)
                     )
-                    llm_config = result.scalar_one_or_none()
-                    if llm_config:
-                        model_id = llm_config.id
+                    llm_cfg = result.scalar_one_or_none()
+                    if llm_cfg:
+                        model_id = llm_cfg.id
         
         # 如果还是没有，尝试从节点定义本身获取
         if not model_id:
@@ -687,9 +690,9 @@ class WorkflowEngine:
                         result = await self.session.execute(
                             select(LLMConfig).where(LLMConfig.model_name == model_value)
                         )
-                        llm_config = result.scalar_one_or_none()
-                        if llm_config:
-                            model_id = llm_config.id
+                        llm_cfg = result.scalar_one_or_none()
+                        if llm_cfg:
+                            model_id = llm_cfg.id
         
         # 如果还是没有，尝试使用默认的LLM配置
         if not model_id:
@@ -698,6 +701,7 @@ class WorkflowEngine:
             default_config = await llm_config_service.get_default_chat_config(self.session)
             if default_config:
                 model_id = default_config.id
+                used_default_model = True
                 logger.info(f"LLM节点未指定模型配置，使用默认模型: {default_config.model_name} (ID: {model_id})")
             else:
                 raise ValueError(
@@ -826,8 +830,40 @@ class WorkflowEngine:
             }
             
         except Exception as e:
-            logger.error(f"LLM调用失败: {str(e)}")
-            raise ValueError(f"LLM调用失败: {str(e)}")
+            error_msg = str(e)
+            logger.error(f"LLM调用失败: {error_msg}")
+            
+            # 如果当前使用的不是默认模型，并且错误包含 Not Found / 404，则尝试回退到默认模型再调用一次
+            if (not used_default_model) and ("Not Found" in error_msg or "404" in error_msg):
+                try:
+                    from ..services.llm_config_service import LLMConfigService
+                    llm_config_service = LLMConfigService()
+                    default_config = await llm_config_service.get_default_chat_config(self.session)
+                    if default_config:
+                        logger.warning(
+                            f"LLM调用失败，模型可能不存在或端点错误，"
+                            f"尝试使用默认模型重试: {default_config.model_name} (ID: {default_config.id})"
+                        )
+                        fallback_response = await self.llm_service.chat_completion(
+                            model_config=default_config,
+                            messages=[{"role": "user", "content": prompt}],
+                            temperature=config.get('temperature', 0.7),
+                            max_tokens=config.get('max_tokens')
+                        )
+                        return {
+                            'success': True,
+                            'response': fallback_response,
+                            'prompt': prompt,
+                            'model': default_config.model_name,
+                            'tokens_used': getattr(fallback_response, 'usage', {}).get('total_tokens', 0)
+                            if hasattr(fallback_response, 'usage') else 0,
+                            'fallback_model_used': True
+                        }
+                except Exception as fallback_error:
+                    logger.error(f"使用默认模型重试LLM调用失败: {str(fallback_error)}")
+                    # 继续向下抛出原始错误
+            
+            raise ValueError(f"LLM调用失败: {error_msg}")
     
     def _substitute_variables(self, template: str, input_data: Dict[str, Any]) -> str:
         """变量替换函数"""