myagent-ai 1.16.13 → 1.16.15

package/agents/main_agent.py

@@ -693,6 +693,41 @@ class MainAgent(BaseAgent):
                         )
                     break
 
+                # [v1.16.13] 特殊处理模型不支持图片输入 — 去掉图片用纯文本重试
+                _vision_keywords = ["doesn't support image", "does not support image", "model_incompatible", "image input", "not support vision", "unsupported multimodal", "image capability"]
+                if any(kw in _llm_error.lower() for kw in _vision_keywords) and context.metadata.get("user_images"):
+                    logger.warning(f"[{task_id}] 模型不支持图片输入，去掉图片用纯文本重试")
+                    context.metadata["user_images"] = []
+                    # 用纯文本消息替换最后一条多模态消息
+                    _text_only_msg = context.user_message or "请处理上述上下文。"
+                    if len(messages) > 0 and isinstance(messages[-1].content, list):
+                        messages[-1] = Message(role="user", content=_text_only_msg)
+                    # 重试 LLM 调用
+                    if stream_response and self.llm:
+                        response = await self._call_llm_stream(
+                            messages, text_delta_callback=text_delta_callback,
+                            stream_response=stream_response,
+                        )
+                    else:
+                        response = await self._call_llm(messages)
+                    if response.success:
+                        # 纯文本重试成功，给回复加上提示前缀
+                        _vision_prefix = "⚠️ 当前模型不支持图片识别，已自动使用纯文本模式处理（图片未发送给模型）。\n\n"
+                        llm_raw = _vision_prefix + response.content
+                        context.working_memory["final_response"] = llm_raw
+                        await self._emit_v2_event("v2_reasoning", {"content": llm_raw}, stream_callback)
+                        if self.memory:
+                            self.memory.add_session(
+                                session_id=context.session_id,
+                                role="assistant",
+                                content=llm_raw,
+                            )
+                        break
+                    else:
+                        # 纯文本也失败了，走下面的通用错误处理
+                        _llm_error = response.error or ""
+                        logger.error(f"[{task_id}] 纯文本重试也失败: {_llm_error}")
+
                 # 其他 LLM 错误
                 error_msg = f"LLM 调用失败: {response.error}"
                 context.working_memory["final_response"] = error_msg

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "myagent-ai",
-  "version": "1.16.13",
+  "version": "1.16.15",
   "description": "本地桌面端执行型AI助手 - Open Interpreter 风格 | Local Desktop Execution-Oriented AI Assistant",
   "main": "main.py",
   "bin": {
@@ -43,4 +43,4 @@
     "python": ">=3.10",
     "node": ">=18"
   }
-}
+}

package/web/api_server.py

@@ -827,6 +827,8 @@ class ApiServer:
             try:
                 agent_cfg = self._read_agent_config(agent_path)
                 model_chain = self._build_model_chain(agent_cfg, agent_path)
+                # [v1.16.14] 有图片时，优先使用支持 vision 的模型
+                model_chain = self._reorder_model_chain_for_images(model_chain, bool(user_images))
                 logger.info(f"[{session_id}] model_chain={'有' if model_chain else '无'}, llm={'有' if self.core.llm else '无'}")
 
                 task_plan_context = self._build_task_plan_context(agent_path, chat_mode, message, session_id=session_id)
@@ -3572,6 +3574,38 @@ class ApiServer:
 
         return chain
 
+    def _reorder_model_chain_for_images(self, model_chain: list[dict], has_images: bool) -> list[dict]:
+        """[v1.16.14] 当消息包含图片时，将支持 vision 的模型优先排列
+
+        检查每个模型的 input_modes 字段，如果包含 "image"，则优先使用。
+        如果都没有 vision 能力，则保持原序（会触发 main_agent 的纯文本降级）。
+        """
+        if not has_images or not model_chain:
+            return model_chain
+
+        # 从 models_library 获取每个模型的 input_modes
+        vision_models = []
+        text_only_models = []
+        for mc in model_chain:
+            mc_id = mc.get("id", "")
+            input_modes = ["text"]  # 默认
+            if mc_id:
+                for me in self.core.config.models_library:
+                    if me.id == mc_id:
+                        input_modes = me.input_modes or ["text"]
+                        break
+            if "image" in input_modes:
+                vision_models.append(mc)
+            else:
+                text_only_models.append(mc)
+
+        if vision_models:
+            logger.info(f"消息含图片，优先使用 vision 模型: {[m.get('name', m.get('model')) for m in vision_models]}")
+            return vision_models + text_only_models
+        else:
+            logger.info("消息含图片，但模型链中无 vision 模型，将尝试当前主模型（失败后自动降级纯文本）")
+            return model_chain
+
     async def _try_model_chain(self, model_chain: list[dict], message: str, session_id: str,
                                  agent_path: str = None, agent_system_prompt: str = None,
                                  chat_mode: str = "") -> str: