myagent-ai 1.17.0 → 1.17.2

package/agents/main_agent.py

@@ -644,6 +644,9 @@ class MainAgent(BaseAgent):
             else:
                 # [v1.16.12] 支持多模态消息（图片+文本）
                 user_images = context.metadata.get("user_images", [])
+                # [v1.17.1] 图片调试
+                if user_images:
+                    logger.info(f"[{task_id}] 检测到 {len(user_images)} 张图片, 构建多模态消息")
                 if user_images:
                     # OpenAI Vision 格式: [{type: "text"}, {type: "image_url"}]
                     multimodal_content = [{"type": "text", "text": context.user_message or "请描述这些图片。"}]
@@ -653,6 +656,8 @@ class MainAgent(BaseAgent):
                                 "type": "image_url",
                                 "image_url": {"url": img["url"]}
                             })
+                    logger.info(f"[{task_id}] 多模态消息构建完成: {len(multimodal_content)} parts, content_types={[p.get('type') for p in multimodal_content]}")
+                    logger.info(f"[{task_id}] LLM model={self.llm.model}, provider={self.llm.provider}")
                     messages.append(Message(role="user", content=multimodal_content))
                 else:
                     messages.append(Message(
@@ -701,9 +706,15 @@ 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"):
+                # [v1.16.13→17.0] 特殊处理模型不支持图片输入 — 去掉图片用纯文本重试
+                # 支持中英文错误关键词匹配（ModelScope 等国产 API 可能返回中文错误）
+                _vision_keywords = [
+                    "doesn't support image", "does not support image", "model_incompatible",
+                    "image input", "not support vision", "unsupported multimodal", "image capability",
+                    "不支持图片", "不支持图像", "图片输入", "图像输入", "不支持多模态",
+                    "视觉", "image_url", "multimodal", "vision",
+                ]
+                if any(kw.lower() 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

package/config.py

@@ -316,8 +316,21 @@ class ConfigManager:
             elif isinstance(value, list) and key == "models_library":
                 models = []
                 for item in value:
+                    # [v1.17.0] 兼容 "input" → "input_modes" 字段映射
+                    # 用户配置中常使用 "input": ["text", "image"]，内部字段名是 "input_modes"
+                    normalized = dict(item)
+                    if "input" in normalized and "input_modes" not in normalized:
+                        normalized["input_modes"] = normalized.pop("input")
+                    if "baseUrl" in normalized and "base_url" not in normalized:
+                        normalized["base_url"] = normalized.pop("baseUrl")
+                    if "apiKey" in normalized and "api_key" not in normalized:
+                        normalized["api_key"] = normalized.pop("apiKey")
+                    if "reasoning" in normalized:
+                        # 确保 reasoning 是 bool 类型
+                        v = normalized["reasoning"]
+                        normalized["reasoning"] = bool(v) if not isinstance(v, bool) else v
                     me = ModelEntry(**{
-                        k: v for k, v in item.items() if k in ModelEntry.__dataclass_fields__
+                        k: v for k, v in normalized.items() if k in ModelEntry.__dataclass_fields__
                     })
                     models.append(me)
                 setattr(target, key, models)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "myagent-ai",
-  "version": "1.17.0",
+  "version": "1.17.2",
   "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": {

package/web/api_server.py

@@ -1091,6 +1091,14 @@ window.toggleFullscreen = function() {{
         # 支持文件附件（文档等，提取文本后作为上下文）
         user_files = data.get("files", [])    # [{"name": "xxx.pdf", "type": "application/pdf", "data": "base64..."}, ...]
         
+        # [v1.17.1] 图片调试日志
+        if user_images:
+            _img_info = []
+            for _img in user_images:
+                _b64len = len(_img.get("data", "")) if isinstance(_img.get("data"), str) else 0
+                _img_info.append(f"{ _img.get('name','?') }(type={_img.get('type','?')}, b64len={_b64len})")
+            logger.info(f"[{session_id if 'session_id' in dir() else 'req'}] 收到图片: {_img_info}")
+        
         # 如果消息为空但有图片，设置默认提示
         if not message and user_images:
             message = "请查看这些图片"
@@ -1189,7 +1197,12 @@ window.toggleFullscreen = function() {{
                 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'}")
+                # [v1.17.1] 详细模型链日志
+                _chain_detail = []
+                for _mc in model_chain:
+                    _chain_detail.append(f"{_mc.get('name','?')}@{_mc.get('model','?')}(modes={_mc.get('input_modes',[])}provider={_mc.get('provider','?')})")
+                logger.info(f"[{session_id}] model_chain ({len(model_chain)} models): {_chain_detail}")
+                logger.info(f"[{session_id}] user_images={len(user_images) if user_images else 0}, has_images={bool(user_images)}")
 
                 task_plan_context = self._build_task_plan_context(agent_path, chat_mode, message, session_id=session_id)
                 clean_message, agent_system_prompt = self._build_agent_chat_context(agent_path, agent_cfg, message)
@@ -4079,6 +4092,8 @@ window.toggleFullscreen = function() {{
                             break
 
             logger.debug(f"模型 {model_name} (id={mc_id}) input_modes={input_modes}")
+            # [v1.17.1] 改为 info 级别，方便排查
+            logger.info(f"[reorder] 模型 {model_name} (id={mc_id}) input_modes={input_modes}")
             if "image" in input_modes:
                 vision_models.append(mc)
             else:
@@ -4371,7 +4386,9 @@ window.toggleFullscreen = function() {{
                         await stream_response.write(("data: " + json.dumps({"type": "user_files", "images": _image_file_ids}) + "\n\n").encode())
                     except Exception:
                         pass
-                logger.info(f"[{session_id}] 用户发送了 {len(_processed_images)} 张图片")
+                # [v1.17.1] 调试日志
+                logger.info(f"[{session_id}] 用户发送了 {len(_processed_images)} 张图片, data_uri_len={len(_processed_images[0].get('url',''))}")
+                logger.info(f"[{session_id}] context.metadata['user_images'] 已设置, 将传给 main_agent")
 
         # [v1.16.12→17] 处理用户文件附件 — 保存到磁盘 + 提取文本附加到消息
         if user_files: