autoglm-gui 1.5.0__py3-none-any.whl → 1.5.2__py3-none-any.whl

AutoGLM_GUI/agents/glm/async_agent.py

@@ -0,0 +1,515 @@
+"""AsyncGLMAgent - 异步 GLM Agent 实现，支持原生流式输出和立即取消。"""
+
+import asyncio
+import json
+import traceback
+from typing import Any, AsyncIterator, Callable
+
+from openai import AsyncOpenAI
+
+from AutoGLM_GUI.actions import ActionHandler, ActionResult
+from AutoGLM_GUI.config import AgentConfig, ModelConfig, StepResult
+from AutoGLM_GUI.device_protocol import DeviceProtocol
+from AutoGLM_GUI.logger import logger
+from AutoGLM_GUI.prompt_config import get_messages, get_system_prompt
+
+from .message_builder import MessageBuilder
+from .parser import GLMParser
+
+
+class AsyncGLMAgent:
+    """异步 GLM Agent 实现。
+
+    核心特性:
+    - 使用 AsyncOpenAI 进行异步 LLM 调用
+    - 原生支持流式输出 (async for)
+    - 支持立即取消 (asyncio.CancelledError)
+    - 使用 asyncio.to_thread 包装同步的设备操作
+
+    与 GLMAgent 的区别:
+    - stream() 方法返回 AsyncIterator，不需要 worker 线程
+    - cancel() 可以立即中断 HTTP 请求
+    - 不需要 monkey-patch thinking_callback
+    """
+
+    def __init__(
+        self,
+        model_config: ModelConfig,
+        agent_config: AgentConfig,
+        device: DeviceProtocol,
+        confirmation_callback: Callable[[str], bool] | None = None,
+        takeover_callback: Callable[[str], None] | None = None,
+    ):
+        self.model_config = model_config
+        self.agent_config = agent_config
+
+        # 使用 AsyncOpenAI
+        self.openai_client = AsyncOpenAI(
+            base_url=model_config.base_url,
+            api_key=model_config.api_key,
+            timeout=120,
+        )
+        self.parser = GLMParser()
+
+        self.device = device
+        self.action_handler = ActionHandler(
+            device=self.device,
+            confirmation_callback=confirmation_callback,
+            takeover_callback=takeover_callback,
+        )
+
+        # 取消机制
+        self._cancel_event = asyncio.Event()
+
+        # 状态
+        self._context: list[dict[str, Any]] = []
+        self._step_count = 0
+        self._is_running = False
+
+    async def stream(self, task: str) -> AsyncIterator[dict[str, Any]]:
+        """流式执行任务，支持取消。
+
+        Args:
+            task: 任务描述
+
+        Yields:
+            dict[str, Any]: 事件字典，格式为 {"type": str, "data": dict}
+
+        事件类型:
+        - "thinking": {"chunk": str}
+        - "step": {"step": int, "thinking": str, "action": dict, ...}
+        - "done": {"message": str, "steps": int, "success": bool}
+        - "cancelled": {"message": str}
+        - "error": {"message": str}
+        """
+        self._context = []
+        self._step_count = 0
+        self._is_running = True
+        self._cancel_event.clear()
+
+        try:
+            # 首步执行
+            async for event in self._execute_step_async(task, is_first=True):
+                yield event
+
+                # 检查是否完成
+                if event["type"] == "step" and event["data"].get("finished"):
+                    yield {
+                        "type": "done",
+                        "data": {
+                            "message": event["data"].get("message", "Task completed"),
+                            "steps": self._step_count,
+                            "success": event["data"].get("success", True),
+                        },
+                    }
+                    return
+
+            # 后续步骤
+            while self._step_count < self.agent_config.max_steps and self._is_running:
+                # 检查取消
+                if self._cancel_event.is_set():
+                    raise asyncio.CancelledError()
+
+                async for event in self._execute_step_async(None, is_first=False):
+                    yield event
+
+                    # 检查是否完成
+                    if event["type"] == "step" and event["data"].get("finished"):
+                        yield {
+                            "type": "done",
+                            "data": {
+                                "message": event["data"].get(
+                                    "message", "Task completed"
+                                ),
+                                "steps": self._step_count,
+                                "success": event["data"].get("success", True),
+                            },
+                        }
+                        return
+
+            # 达到最大步数
+            yield {
+                "type": "done",
+                "data": {
+                    "message": "Max steps reached",
+                    "steps": self._step_count,
+                    "success": False,
+                },
+            }
+
+        except asyncio.CancelledError:
+            yield {
+                "type": "cancelled",
+                "data": {"message": "Task cancelled by user"},
+            }
+            raise
+
+        finally:
+            self._is_running = False
+
+    async def _execute_step_async(
+        self, user_prompt: str | None, is_first: bool
+    ) -> AsyncIterator[dict[str, Any]]:
+        """执行单步，支持流式输出和取消。
+
+        Args:
+            user_prompt: 用户输入（首步必需，后续可选）
+            is_first: 是否是首步
+
+        Yields:
+            dict[str, Any]: 事件字典
+        """
+        self._step_count += 1
+
+        # 1. 截图和获取当前应用（使用线程池）
+        try:
+            screenshot = await asyncio.to_thread(self.device.get_screenshot)
+            current_app = await asyncio.to_thread(self.device.get_current_app)
+        except Exception as e:
+            logger.error(f"Failed to get device info: {e}")
+            yield {
+                "type": "error",
+                "data": {"message": f"Device error: {e}"},
+            }
+            yield {
+                "type": "step",
+                "data": {
+                    "step": self._step_count,
+                    "thinking": "",
+                    "action": None,
+                    "success": False,
+                    "finished": True,
+                    "message": f"Device error: {e}",
+                },
+            }
+            return
+
+        # 2. 构建消息
+        if is_first:
+            system_prompt = self.agent_config.system_prompt
+            if system_prompt is None:
+                system_prompt = get_system_prompt(self.agent_config.lang)
+
+            self._context.append(MessageBuilder.create_system_message(system_prompt))
+
+            screen_info = MessageBuilder.build_screen_info(current_app)
+            text_content = f"{user_prompt}\n\n{screen_info}"
+
+            self._context.append(
+                MessageBuilder.create_user_message(
+                    text=text_content, image_base64=screenshot.base64_data
+                )
+            )
+        else:
+            screen_info = MessageBuilder.build_screen_info(current_app)
+            if user_prompt:
+                text_content = f"{user_prompt}\n\n** Screen Info **\n\n{screen_info}"
+            else:
+                text_content = f"** Screen Info **\n\n{screen_info}"
+
+            self._context.append(
+                MessageBuilder.create_user_message(
+                    text=text_content, image_base64=screenshot.base64_data
+                )
+            )
+
+        # 3. 流式调用 OpenAI（真正的异步，可取消）
+        try:
+            if self.agent_config.verbose:
+                msgs = get_messages(self.agent_config.lang)
+                print("\n" + "=" * 50)
+                print(f"💭 {msgs['thinking']}:")
+                print("-" * 50)
+
+            thinking_parts = []
+            raw_content = ""
+
+            async for chunk_data in self._stream_openai(self._context):
+                # 检查取消
+                if self._cancel_event.is_set():
+                    raise asyncio.CancelledError()
+
+                if chunk_data["type"] == "thinking":
+                    thinking_parts.append(chunk_data["content"])
+
+                    # Yield thinking event
+                    yield {
+                        "type": "thinking",
+                        "data": {"chunk": chunk_data["content"]},
+                    }
+
+                    # Verbose output
+                    if self.agent_config.verbose:
+                        print(chunk_data["content"], end="", flush=True)
+
+                elif chunk_data["type"] == "raw":
+                    raw_content += chunk_data["content"]
+
+            thinking = "".join(thinking_parts)
+
+        except asyncio.CancelledError:
+            logger.info(f"Step {self._step_count} cancelled during LLM call")
+            raise
+
+        except Exception as e:
+            logger.error(f"LLM error: {e}")
+            if self.agent_config.verbose:
+                traceback.print_exc()
+
+            yield {
+                "type": "error",
+                "data": {"message": f"Model error: {e}"},
+            }
+            yield {
+                "type": "step",
+                "data": {
+                    "step": self._step_count,
+                    "thinking": "",
+                    "action": None,
+                    "success": False,
+                    "finished": True,
+                    "message": f"Model error: {e}",
+                },
+            }
+            return
+
+        # 4. 解析 action
+        _, action_str = self._parse_raw_response(raw_content)
+
+        try:
+            action = self.parser.parse(action_str)
+        except ValueError as e:
+            if self.agent_config.verbose:
+                logger.warning(f"Failed to parse action: {e}, treating as finish")
+            action = {"_metadata": "finish", "message": action_str}
+
+        if self.agent_config.verbose:
+            msgs = get_messages(self.agent_config.lang)
+            print()
+            print("-" * 50)
+            print(f"🎯 {msgs['action']}:")
+            print(json.dumps(action, ensure_ascii=False, indent=2))
+            print("=" * 50 + "\n")
+
+        # 5. 执行 action（使用线程池）
+        try:
+            result = await asyncio.to_thread(
+                self.action_handler.execute, action, screenshot.width, screenshot.height
+            )
+        except Exception as e:
+            logger.error(f"Action execution error: {e}")
+            if self.agent_config.verbose:
+                traceback.print_exc()
+            result = ActionResult(success=False, should_finish=True, message=str(e))
+
+        # 6. 更新上下文
+        self._context[-1] = MessageBuilder.remove_images_from_message(self._context[-1])
+
+        self._context.append(
+            MessageBuilder.create_assistant_message(
+                f"<think>{thinking}</think><answer>{action_str}</answer>"
+            )
+        )
+
+        # 7. 检查是否完成
+        finished = action.get("_metadata") == "finish" or result.should_finish
+
+        if finished and self.agent_config.verbose:
+            msgs = get_messages(self.agent_config.lang)
+            print("\n" + "🎉 " + "=" * 48)
+            print(
+                f"✅ {msgs['task_completed']}: {result.message or action.get('message', msgs['done'])}"
+            )
+            print("=" * 50 + "\n")
+
+        # 8. 返回步骤结果
+        yield {
+            "type": "step",
+            "data": {
+                "step": self._step_count,
+                "thinking": thinking,
+                "action": action,
+                "success": result.success,
+                "finished": finished,
+                "message": result.message or action.get("message"),
+            },
+        }
+
+    async def _stream_openai(
+        self, messages: list[dict[str, Any]]
+    ) -> AsyncIterator[dict[str, str]]:
+        """流式调用 OpenAI，yield thinking chunks。
+
+        Args:
+            messages: 消息列表
+
+        Yields:
+            dict[str, str]: {"type": "thinking" | "raw", "content": str}
+
+        Raises:
+            asyncio.CancelledError: 任务被取消
+        """
+        stream = await self.openai_client.chat.completions.create(
+            messages=messages,  # type: ignore[arg-type]
+            model=self.model_config.model_name,
+            max_tokens=self.model_config.max_tokens,
+            temperature=self.model_config.temperature,
+            top_p=self.model_config.top_p,
+            frequency_penalty=self.model_config.frequency_penalty,
+            extra_body=self.model_config.extra_body,
+            stream=True,
+        )
+
+        buffer = ""
+        action_markers = ["finish(message=", "do(action="]
+        in_action_phase = False
+
+        try:
+            async for chunk in stream:
+                # 检查取消
+                if self._cancel_event.is_set():
+                    await stream.close()  # 关键：关闭 HTTP 连接
+                    raise asyncio.CancelledError()
+
+                if len(chunk.choices) == 0:
+                    continue
+
+                if chunk.choices[0].delta.content is not None:
+                    content = chunk.choices[0].delta.content
+                    yield {"type": "raw", "content": content}
+
+                    if in_action_phase:
+                        continue
+
+                    buffer += content
+
+                    # 检查是否到达 action 标记
+                    marker_found = False
+                    for marker in action_markers:
+                        if marker in buffer:
+                            thinking_part = buffer.split(marker, 1)[0]
+                            yield {"type": "thinking", "content": thinking_part}
+                            in_action_phase = True
+                            marker_found = True
+                            break
+
+                    if marker_found:
+                        continue
+
+                    # 检查是否是潜在的 marker 前缀
+                    is_potential_marker = False
+                    for marker in action_markers:
+                        for i in range(1, len(marker)):
+                            if buffer.endswith(marker[:i]):
+                                is_potential_marker = True
+                                break
+                        if is_potential_marker:
+                            break
+
+                    if not is_potential_marker and len(buffer) > 0:
+                        yield {"type": "thinking", "content": buffer}
+                        buffer = ""
+
+        finally:
+            await stream.close()  # 确保资源释放
+
+    def _parse_raw_response(self, content: str) -> tuple[str, str]:
+        """解析原始响应，提取 thinking 和 action。
+
+        Args:
+            content: 原始响应内容
+
+        Returns:
+            tuple[str, str]: (thinking, action)
+        """
+        if "finish(message=" in content:
+            parts = content.split("finish(message=", 1)
+            thinking = parts[0].strip()
+            action = "finish(message=" + parts[1]
+            return thinking, action
+
+        if "do(action=" in content:
+            parts = content.split("do(action=", 1)
+            thinking = parts[0].strip()
+            action = "do(action=" + parts[1]
+            return thinking, action
+
+        if "<answer>" in content:
+            parts = content.split("<answer>", 1)
+            thinking = parts[0].replace("<think>", "").replace("</think>", "").strip()
+            action = parts[1].replace("</answer>", "").strip()
+            return thinking, action
+
+        return "", content
+
+    async def cancel(self) -> None:
+        """取消当前执行。
+
+        设置取消标志，中断正在进行的 HTTP 请求。
+        """
+        self._cancel_event.set()
+        self._is_running = False
+        logger.info("AsyncGLMAgent cancelled by user")
+
+    def reset(self) -> None:
+        """重置状态。"""
+        self._context = []
+        self._step_count = 0
+        self._is_running = False
+        self._cancel_event.clear()
+
+    async def run(self, task: str) -> str:
+        """运行完整任务（兼容接口）。
+
+        Args:
+            task: 任务描述
+
+        Returns:
+            str: 最终结果消息
+        """
+        final_message = ""
+        async for event in self.stream(task):
+            if event["type"] == "done":
+                final_message = event["data"].get("message", "")
+        return final_message
+
+    async def step(self, task: str | None = None) -> StepResult:
+        """执行单步（兼容接口）。
+
+        Args:
+            task: 任务描述（首步必需，后续可选）
+
+        Returns:
+            StepResult: 步骤结果
+        """
+        is_first = len(self._context) == 0
+        if is_first and not task:
+            raise ValueError("Task is required for the first step")
+
+        result = None
+        async for event in self._execute_step_async(task, is_first):
+            if event["type"] == "step":
+                result = StepResult(
+                    thinking=event["data"]["thinking"],
+                    action=event["data"]["action"],
+                    success=event["data"]["success"],
+                    finished=event["data"]["finished"],
+                    message=event["data"].get("message"),
+                )
+
+        if result is None:
+            raise RuntimeError("Step execution did not produce a result")
+
+        return result
+
+    @property
+    def step_count(self) -> int:
+        return self._step_count
+
+    @property
+    def context(self) -> list[dict[str, Any]]:
+        return self._context.copy()
+
+    @property
+    def is_running(self) -> bool:
+        return self._is_running

AutoGLM_GUI/agents/glm/parser.py

@@ -98,13 +98,15 @@ class GLMParser:
 
         return params
 
-    def _parse_value(self, value_str: str) -> Any:
+    def _parse_value(
+        self, value_str: str
+    ) -> str | int | float | bool | list | dict | None:
         value_str = value_str.strip()
 
         if not value_str:
             return ""
 
         try:
-            return ast.literal_eval(value_str)
+            return ast.literal_eval(value_str)  # type: ignore[no-any-return]
         except (ValueError, SyntaxError):
             return value_str

AutoGLM_GUI/agents/mai/agent.py

@@ -96,6 +96,9 @@ class InternalMAIAgent:
 
         if is_first:
             self.traj_memory.task_goal = task or ""
+        elif task:
+            # 多轮对话：有新的用户消息，更新 task_goal
+            self.traj_memory.task_goal = task
 
         return self._execute_step(task, is_first)
 

AutoGLM_GUI/agents/protocols.py

@@ -1,10 +1,27 @@
 from __future__ import annotations
 
-from typing import Any, Protocol
+import inspect
+from typing import Any, AsyncIterator, Protocol
 
 from AutoGLM_GUI.config import AgentConfig, ModelConfig, StepResult
 
 
+def is_async_agent(agent: AsyncAgent | BaseAgent) -> bool:
+    """Check if an agent implements the AsyncAgent interface.
+
+    Uses runtime inspection to detect async capabilities since static
+    type narrowing is not possible with Protocol union types.
+
+    Args:
+        agent: Agent instance to check
+
+    Returns:
+        True if agent has async stream() method, False otherwise
+    """
+    stream_method = getattr(agent, "stream", None)
+    return stream_method is not None and inspect.isasyncgenfunction(stream_method)
+
+
 class BaseAgent(Protocol):
     model_config: ModelConfig
     agent_config: AgentConfig
@@ -25,3 +42,96 @@ class BaseAgent(Protocol):
 
     @property
     def is_running(self) -> bool: ...
+
+
+class AsyncAgent(Protocol):
+    """异步 Agent 接口，原生支持流式输出和取消。
+
+    核心特性:
+    - stream() 方法返回 AsyncIterator[dict]，支持原生 async for
+    - cancel() 方法使用 asyncio 取消机制，可立即中断 HTTP 请求
+    - 不需要 worker 线程、queue、monkey-patch
+
+    使用示例:
+        async for event in agent.stream("打开微信"):
+            if event["type"] == "thinking":
+                print(event["data"]["chunk"])
+            elif event["type"] == "step":
+                print(f"Step {event['data']['step']}")
+            elif event["type"] == "done":
+                break
+    """
+
+    model_config: ModelConfig
+    agent_config: AgentConfig
+
+    async def run(self, task: str) -> str:
+        """运行完整任务，返回最终结果。
+
+        Args:
+            task: 任务描述
+
+        Returns:
+            str: 最终结果消息
+        """
+        ...
+
+    async def step(self, task: str | None = None) -> StepResult:
+        """执行单步，返回步骤结果。
+
+        Args:
+            task: 任务描述（首步必需，后续可选）
+
+        Returns:
+            StepResult: 步骤结果
+        """
+        ...
+
+    async def stream(self, task: str) -> AsyncIterator[dict[str, Any]]:
+        """流式执行任务，yield 事件字典。
+
+        这是核心方法，支持:
+        - 实时流式输出 (thinking chunks)
+        - 立即取消 (通过 asyncio.CancelledError)
+        - 不需要额外的线程或队列
+
+        事件类型:
+        - "thinking": {"chunk": str} - 思考过程片段
+        - "step": {"step": int, "thinking": str, "action": dict, ...} - 步骤完成
+        - "done": {"message": str, "steps": int, "success": bool} - 任务完成
+        - "cancelled": {"message": str} - 任务取消
+        - "error": {"message": str} - 错误
+
+        Args:
+            task: 任务描述
+
+        Yields:
+            dict[str, Any]: 事件字典，格式为 {"type": str, "data": dict}
+
+        Raises:
+            asyncio.CancelledError: 任务被取消
+        """
+        ...
+
+    async def cancel(self) -> None:
+        """取消当前执行（立即中断网络请求）。
+
+        使用 asyncio 的取消机制，会:
+        1. 设置内部取消标志
+        2. 关闭正在进行的 HTTP 连接
+        3. 抛出 asyncio.CancelledError
+        """
+        ...
+
+    def reset(self) -> None:
+        """重置状态（同步方法，只清理内存）。"""
+        ...
+
+    @property
+    def step_count(self) -> int: ...
+
+    @property
+    def context(self) -> list[dict[str, Any]]: ...
+
+    @property
+    def is_running(self) -> bool: ...