Add Web Dashboard with multi-device control and callback hooks

Features:
- Web Dashboard: FastAPI-based dashboard with Vue.js frontend
  - Multi-device support (ADB, HDC, iOS)
  - Real-time WebSocket updates for task progress
  - Device management with status tracking
  - Task queue with execution controls (start/stop/re-execute)
  - Detailed task information display (thinking, actions, completion messages)
  - Screenshot viewing per device
  - LAN deployment support with configurable CORS

- Callback Hooks: Interrupt and modify task execution
  - step_callback: Called after each step with StepResult
  - before_action_callback: Called before executing action
  - Support for task interruption and dynamic task switching
  - Example scripts demonstrating callback usage

- Configuration: Environment-based configuration
  - .env file support for all settings
  - .env.example template with documentation
  - Model API configuration (base URL, model name, API key)
  - Dashboard configuration (host, port, CORS, device type)
  - Phone agent configuration (delays, max steps, language)

Technical improvements:
- Fixed forward reference issue with StepResult
- Added package exports for callback types and configs
- Enhanced dependencies with FastAPI, WebSocket support
- Thread-safe task execution with device locking
- Async WebSocket broadcasting from sync thread pool

Co-Authored-By: Claude <noreply@anthropic.com>

.env.example

@@ -0,0 +1,110 @@
+# Phone Agent Environment Configuration
+# Copy this file to .env and fill in your values
+
+# ============================================================================
+# Model Configuration / 模型配置
+# ============================================================================
+
+# Model API base URL / 模型 API 地址
+# Default: http://localhost:8000/v1
+# For BigModel (智谱AI): https://open.bigmodel.cn/api/paas/v4
+# For ModelScope: https://api-inference.modelscope.cn/v1
+PHONE_AGENT_BASE_URL=http://localhost:8000/v1
+
+# Model name / 模型名称
+# Default: autoglm-phone-9b
+PHONE_AGENT_MODEL=autoglm-phone-9b
+
+# API key for authentication / API 密钥
+# Default: EMPTY (no authentication)
+# For BigModel: your actual API key
+PHONE_AGENT_API_KEY=EMPTY
+
+# ============================================================================
+# Agent Configuration / Agent 配置
+# ============================================================================
+
+# Maximum steps per task / 每个任务的最大步数
+PHONE_AGENT_MAX_STEPS=100
+
+# Language for system prompt / 系统提示词语言
+# Options: cn (Chinese), en (English)
+PHONE_AGENT_LANG=cn
+
+# Verbose output / 详细输出
+# Options: true, false
+PHONE_AGENT_VERBOSE=true
+
+# ============================================================================
+# Device Configuration / 设备配置
+# ============================================================================
+
+# Device ID (for multi-device setups) / 设备 ID（多设备时使用）
+# Leave empty to auto-detect / 留空则自动检测
+PHONE_AGENT_DEVICE_ID=
+
+# Device type / 设备类型
+# Options: adb (Android), hdc (HarmonyOS), ios (iPhone)
+PHONE_AGENT_DEVICE_TYPE=adb
+
+# ============================================================================
+# iOS Specific Configuration / iOS 专用配置
+# ============================================================================
+
+# WebDriverAgent URL for iOS / iOS WebDriverAgent 地址
+PHONE_AGENT_WDA_URL=http://localhost:8100
+
+# ============================================================================
+# Timing Configuration (Optional) / 时序配置（可选）
+# ============================================================================
+
+# Action delays in seconds / 动作延迟（秒）
+PHONE_AGENT_KEYBOARD_SWITCH_DELAY=1.0
+PHONE_AGENT_TEXT_INPUT_DELAY=1.0
+
+# Device delays in seconds / 设备延迟（秒）
+PHONE_AGENT_TAP_DELAY=1.0
+PHONE_AGENT_SWIPE_DELAY=1.0
+PHONE_AGENT_LAUNCH_DELAY=1.0
+
+# Connection delays in seconds / 连接延迟（秒）
+PHONE_AGENT_ADB_RESTART_DELAY=2.0
+
+# ============================================================================
+# Web Dashboard Configuration / Web 仪表板配置
+# ============================================================================
+
+# Dashboard server host / 仪表板服务器主机
+# Use 0.0.0.0 for LAN access / 使用 0.0.0.0 以支持局域网访问
+DASHBOARD_HOST=0.0.0.0
+
+# Dashboard server port / 仪表板服务器端口
+DASHBOARD_PORT=8080
+
+# Dashboard debug mode / 仪表板调试模式
+# Options: true, false
+DASHBOARD_DEBUG=false
+
+# CORS allowed origins / CORS 允许的来源
+# Comma-separated list, use * for all / 逗号分隔列表，使用 * 表示全部
+# Examples: http://localhost:3000,http://192.168.1.*
+DASHBOARD_CORS_ORIGINS=*
+
+# Default device type for dashboard / 仪表板默认设备类型
+# Options: adb, hdc, ios
+DEFAULT_DEVICE_TYPE=adb
+
+# Maximum concurrent tasks / 最大并发任务数
+MAX_CONCURRENT_TASKS=10
+
+# Task timeout in seconds / 任务超时时间（秒）
+TASK_TIMEOUT_SECONDS=300
+
+# Screenshot quality (1-100) / 截图质量 (1-100)
+SCREENSHOT_QUALITY=80
+
+# Screenshot throttle delay in milliseconds / 截图节流延迟（毫秒）
+SCREENSHOT_THROTTLE_MS=500
+
+# Maximum task history to keep / 保留的最大任务历史数
+MAX_TASK_HISTORY=100

.gitignore

@@ -54,6 +54,11 @@ Thumbs.db
 *.log
 /tmp/
 screenshots/
+*.apk
+
+# Environment variables (keep .env.example as template)
+.env
+.env.local
 
 # Keep old files during transition
 call_model.py

dashboard/__init__.py

@@ -0,0 +1,8 @@
+"""
+AutoGLM Dashboard - Web-based multi-device control interface.
+
+This package provides a FastAPI web application for controlling multiple
+Android (ADB), HarmonyOS (HDC), and iOS devices through a web interface.
+"""
+
+__version__ = "0.1.0"

dashboard/api/__init__.py

@@ -0,0 +1,13 @@
+"""
+API endpoints for the dashboard.
+"""
+
+from dashboard.api.devices import router as devices_router
+from dashboard.api.tasks import router as tasks_router
+from dashboard.api.websocket import router as websocket_router
+
+__all__ = [
+    "devices_router",
+    "tasks_router",
+    "websocket_router",
+]

dashboard/api/devices.py

@@ -0,0 +1,193 @@
+"""
+Device management API endpoints.
+"""
+
+from typing import List
+
+from fastapi import APIRouter, Depends, HTTPException
+
+from dashboard.dependencies import get_device_manager, get_ws_manager
+from dashboard.models.device import DeviceSchema, DeviceStatus
+from dashboard.services.device_manager import DeviceManager
+from dashboard.services.websocket_manager import WebSocketManager
+
+router = APIRouter(prefix="/devices", tags=["devices"])
+
+
+@router.get("", response_model=List[DeviceSchema])
+async def list_devices(
+    device_manager: DeviceManager = Depends(get_device_manager),
+):
+    """List all connected devices.
+
+    Returns:
+        List of device schemas
+    """
+    devices = await device_manager.refresh_devices()
+    return [
+        DeviceSchema(
+            device_id=d.device_id,
+            status=d.status,
+            device_type=d.device_type,
+            model=d.model,
+            android_version=d.android_version,
+            current_app=d.current_app,
+            last_seen=d.last_seen,
+            is_connected=d.is_connected,
+        )
+        for d in devices
+    ]
+
+
+@router.get("/refresh")
+async def refresh_devices(
+    device_manager: DeviceManager = Depends(get_device_manager),
+    ws_manager: WebSocketManager = Depends(get_ws_manager),
+):
+    """Rescan for connected devices.
+
+    Returns:
+        Refresh confirmation
+    """
+    devices = await device_manager.refresh_devices()
+
+    # Broadcast device update
+    for device in devices:
+        await ws_manager.broadcast_device_update(
+            device.device_id,
+            {
+                "status": device.status,
+                "is_connected": device.is_connected,
+                "model": device.model,
+                "current_app": device.current_app,
+                "last_seen": device.last_seen.isoformat(),
+            },
+        )
+
+    return {"message": "Devices refreshed", "count": len(devices)}
+
+
+@router.get("/{device_id}", response_model=DeviceSchema)
+async def get_device(
+    device_id: str,
+    device_manager: DeviceManager = Depends(get_device_manager),
+):
+    """Get device details.
+
+    Args:
+        device_id: Device identifier
+
+    Returns:
+        Device schema
+
+    Raises:
+        HTTPException: If device not found
+    """
+    device = await device_manager.get_device(device_id)
+
+    if device is None:
+        raise HTTPException(status_code=404, detail=f"Device {device_id} not found")
+
+    return DeviceSchema(
+        device_id=device.device_id,
+        status=device.status,
+        device_type=device.device_type,
+        model=device.model,
+        android_version=device.android_version,
+        current_app=device.current_app,
+        last_seen=device.last_seen,
+        is_connected=device.is_connected,
+    )
+
+
+@router.post("/{device_id}/connect")
+async def connect_device(
+    device_id: str,
+    address: str | None = None,
+    device_manager: DeviceManager = Depends(get_device_manager),
+):
+    """Connect to device via WiFi.
+
+    Args:
+        device_id: Device identifier (for route matching)
+        address: Device address (IP:PORT)
+
+    Returns:
+        Connection result
+    """
+    if not address:
+        raise HTTPException(status_code=400, detail="Address is required")
+
+    success = await device_manager.connect_device(address)
+
+    if success:
+        return {"message": f"Connected to {address}"}
+    else:
+        raise HTTPException(status_code=500, detail=f"Failed to connect to {address}")
+
+
+@router.post("/{device_id}/disconnect")
+async def disconnect_device(
+    device_id: str,
+    address: str | None = None,
+    device_manager: DeviceManager = Depends(get_device_manager),
+):
+    """Disconnect from device.
+
+    Args:
+        device_id: Device identifier (for route matching)
+        address: Device address (IP:PORT)
+
+    Returns:
+        Disconnection result
+    """
+    if not address:
+        raise HTTPException(status_code=400, detail="Address is required")
+
+    success = await device_manager.disconnect_device(address)
+
+    if success:
+        return {"message": f"Disconnected from {address}"}
+    else:
+        raise HTTPException(
+            status_code=500, detail=f"Failed to disconnect from {address}"
+        )
+
+
+@router.get("/{device_id}/screenshot")
+async def get_device_screenshot(
+    device_id: str,
+    device_manager: DeviceManager = Depends(get_device_manager),
+):
+    """Get current device screenshot.
+
+    Args:
+        device_id: Device identifier
+
+    Returns:
+        Screenshot data (base64 encoded)
+    """
+    screenshot = await device_manager.get_screenshot(device_id)
+
+    if screenshot is None:
+        raise HTTPException(status_code=500, detail="Failed to capture screenshot")
+
+    return {"device_id": device_id, "screenshot": screenshot}
+
+
+@router.get("/{device_id}/current-app")
+async def get_current_app(
+    device_id: str,
+    device_manager: DeviceManager = Depends(get_device_manager),
+):
+    """Get current app for device.
+
+    Args:
+        device_id: Device identifier
+
+    Returns:
+        Current app package name
+    """
+    app = await device_manager.get_current_app(device_id)
+
+    return {"device_id": device_id, "current_app": app}

dashboard/api/tasks.py

@@ -0,0 +1,156 @@
+"""
+Task management API endpoints.
+"""
+
+from typing import List
+
+from fastapi import APIRouter, Depends, HTTPException
+
+from dashboard.config import config
+from dashboard.dependencies import get_task_executor
+from dashboard.models.task import TaskCreateRequest, TaskSchema, TaskStatus
+from dashboard.services.task_executor import TaskExecutor
+
+router = APIRouter(prefix="/tasks", tags=["tasks"])
+
+
+@router.post("/execute", response_model=dict)
+async def execute_task(
+    request: TaskCreateRequest,
+    executor: TaskExecutor = Depends(get_task_executor),
+):
+    """Execute task on device.
+
+    Args:
+        request: Task creation request
+
+    Returns:
+        Task ID
+    """
+    # Fill in model config from environment if using defaults
+    if request.base_url == "http://localhost:8000/v1":
+        request.base_url = config.MODEL_BASE_URL
+    if request.model_name == "autoglm-phone-9b":
+        request.model_name = config.MODEL_NAME
+    if request.api_key == "EMPTY":
+        request.api_key = config.MODEL_API_KEY
+
+    task_id = await executor.execute_task(request)
+
+    return {"task_id": task_id, "message": "Task started"}
+
+
+@router.post("/{task_id}/stop")
+async def stop_task(
+    task_id: str,
+    executor: TaskExecutor = Depends(get_task_executor),
+):
+    """Stop running task.
+
+    Args:
+        task_id: Task identifier
+
+    Returns:
+        Stop confirmation
+    """
+    await executor.stop_task(task_id)
+
+    return {"message": f"Task {task_id} stop requested"}
+
+
+@router.get("", response_model=List[TaskSchema])
+async def list_tasks(
+    executor: TaskExecutor = Depends(get_task_executor),
+):
+    """List all tasks (active and recent).
+
+    Returns:
+        List of task schemas
+    """
+    return await executor.list_tasks()
+
+
+@router.get("/{task_id}", response_model=TaskSchema)
+async def get_task_status(
+    task_id: str,
+    executor: TaskExecutor = Depends(get_task_executor),
+):
+    """Get task status.
+
+    Args:
+        task_id: Task identifier
+
+    Returns:
+        Task schema
+
+    Raises:
+        HTTPException: If task not found
+    """
+    task = await executor.get_task_status(task_id)
+
+    if task is None:
+        raise HTTPException(status_code=404, detail=f"Task {task_id} not found")
+
+    return task
+
+
+@router.get("/{task_id}/screenshot")
+async def get_task_screenshot(
+    task_id: str,
+    executor: TaskExecutor = Depends(get_task_executor),
+):
+    """Get latest screenshot from task execution.
+
+    Args:
+        task_id: Task identifier
+
+    Returns:
+        Screenshot data (base64 encoded)
+
+    Raises:
+        HTTPException: If task not found or device unavailable
+    """
+    task = await executor.get_task_status(task_id)
+
+    if task is None:
+        raise HTTPException(status_code=404, detail=f"Task {task_id} not found")
+
+    # Get screenshot from device manager
+    from dashboard.dependencies import get_device_manager
+
+    device_manager = get_device_manager()
+    screenshot = await device_manager.get_screenshot(task.device_id)
+
+    if screenshot is None:
+        raise HTTPException(
+            status_code=500, detail="Failed to capture screenshot from device"
+        )
+
+    return {"task_id": task_id, "device_id": task.device_id, "screenshot": screenshot}
+
+
+@router.get("/stats/summary")
+async def get_task_stats(
+    executor: TaskExecutor = Depends(get_task_executor),
+):
+    """Get task execution statistics.
+
+    Returns:
+        Task statistics summary
+    """
+    tasks = await executor.list_tasks()
+
+    total = len(tasks)
+    running = sum(1 for t in tasks if t.status == TaskStatus.RUNNING)
+    completed = sum(1 for t in tasks if t.status == TaskStatus.COMPLETED)
+    failed = sum(1 for t in tasks if t.status == TaskStatus.FAILED)
+    stopped = sum(1 for t in tasks if t.status == TaskStatus.STOPPED)
+
+    return {
+        "total": total,
+        "running": running,
+        "completed": completed,
+        "failed": failed,
+        "stopped": stopped,
+        "active_count": executor.get_active_task_count(),
+    }

dashboard/api/websocket.py

@@ -0,0 +1,124 @@
+"""
+WebSocket API endpoints for real-time updates.
+"""
+
+from fastapi import APIRouter, WebSocket, WebSocketDisconnect, Depends, Query
+
+from dashboard.dependencies import get_ws_manager, get_device_manager
+from dashboard.services.websocket_manager import WebSocketManager
+from dashboard.services.device_manager import DeviceManager
+
+router = APIRouter(prefix="/ws", tags=["websocket"])
+
+
+@router.websocket("")
+async def websocket_endpoint(
+    websocket: WebSocket,
+    client_id: str | None = Query(None),
+    ws_manager: WebSocketManager = Depends(get_ws_manager),
+    device_manager: DeviceManager = Depends(get_device_manager),
+):
+    """WebSocket endpoint for real-time updates.
+
+    This endpoint provides real-time updates for:
+    - Device connection/disconnection
+    - Task execution progress
+    - Screenshot updates
+    - Task completion
+
+    Query parameters:
+        client_id: Optional client ID (auto-generated if not provided)
+
+    Message types (client -> server):
+        - {"type": "subscribe", "device_id": "device_id"} - Subscribe to device updates
+        - {"type": "unsubscribe", "device_id": "device_id"} - Unsubscribe from device updates
+        - {"type": "ping"} - Ping server
+
+    Message types (server -> client):
+        - {"type": "device_update", "data": {...}} - Device status update
+        - {"type": "task_started", "data": {...}} - Task started
+        - {"type": "task_step", "data": {...}} - Task step update
+        - {"type": "task_completed", "data": {...}} - Task completed
+        - {"type": "task_failed", "data": {...}} - Task failed
+        - {"type": "task_stopped", "data": {...}} - Task stopped
+        - {"type": "screenshot", "data": {...}} - Screenshot update
+        - {"type": "error", "data": {...}} - Error occurred
+        - {"type": "pong"} - Pong response
+    """
+    # Accept connection
+    client_id = await ws_manager.connect(websocket, client_id)
+
+    try:
+        # Send initial devices
+        devices = await device_manager.refresh_devices()
+        for device in devices:
+            await websocket.send_json({
+                "type": "device_update",
+                "data": {
+                    "device_id": device.device_id,
+                    "status": device.status.value,
+                    "device_type": device.device_type.value,
+                    "model": device.model,
+                    "android_version": device.android_version,
+                    "current_app": device.current_app,
+                    "is_connected": device.is_connected,
+                }
+            })
+
+        # Message loop
+        while True:
+            data = await websocket.receive_json()
+            msg_type = data.get("type")
+
+            if msg_type == "subscribe":
+                # Subscribe to device updates
+                device_id = data.get("device_id", "*")
+                ws_manager.subscribe_to_device(client_id, device_id)
+
+            elif msg_type == "unsubscribe":
+                # Unsubscribe from device updates
+                device_id = data.get("device_id")
+                if device_id:
+                    ws_manager.unsubscribe_from_device(client_id, device_id)
+
+            elif msg_type == "ping":
+                # Respond to ping
+                await websocket.send_json({"type": "pong"})
+
+    except WebSocketDisconnect:
+        ws_manager.disconnect(client_id)
+    except Exception:
+        ws_manager.disconnect(client_id)
+
+
+@router.websocket("/device/{device_id}")
+async def device_websocket(
+    device_id: str,
+    websocket: WebSocket,
+    ws_manager: WebSocketManager = Depends(get_ws_manager),
+):
+    """Device-specific WebSocket endpoint for real-time updates.
+
+    This endpoint provides real-time updates for a specific device.
+    Automatically subscribes to the device's updates.
+
+    Args:
+        device_id: Device identifier
+    """
+    # Accept connection and auto-subscribe to device
+    client_id = await ws_manager.connect(websocket)
+    ws_manager.subscribe_to_device(client_id, device_id)
+
+    try:
+        # Keep connection alive
+        while True:
+            data = await websocket.receive_json()
+
+            # Handle client messages
+            if data.get("type") == "ping":
+                await websocket.send_json({"type": "pong"})
+
+    except WebSocketDisconnect:
+        ws_manager.disconnect(client_id)
+    except Exception:
+        ws_manager.disconnect(client_id)

dashboard/config.py

@@ -0,0 +1,52 @@
+"""
+Dashboard configuration settings.
+"""
+
+import os
+from typing import List
+
+from dotenv import load_dotenv
+
+# Load .env file
+load_dotenv()
+
+
+class DashboardConfig:
+    """Dashboard configuration."""
+
+    # Server settings
+    HOST: str = os.getenv("DASHBOARD_HOST", "0.0.0.0")
+    PORT: int = int(os.getenv("DASHBOARD_PORT", "8080"))
+    DEBUG: bool = os.getenv("DASHBOARD_DEBUG", "false").lower() == "true"
+
+    # CORS settings
+    CORS_ORIGINS: List[str] = os.getenv(
+        "DASHBOARD_CORS_ORIGINS", "*"
+    ).split(",") if os.getenv("DASHBOARD_CORS_ORIGINS") else ["*"]
+
+    # Device settings
+    DEFAULT_DEVICE_TYPE: str = os.getenv("DEFAULT_DEVICE_TYPE", "adb")
+
+    # Task settings
+    MAX_CONCURRENT_TASKS: int = int(os.getenv("MAX_CONCURRENT_TASKS", "10"))
+    TASK_TIMEOUT_SECONDS: int = int(os.getenv("TASK_TIMEOUT_SECONDS", "300"))
+
+    # Screenshot settings
+    SCREENSHOT_QUALITY: int = int(os.getenv("SCREENSHOT_QUALITY", "80"))
+    SCREENSHOT_THROTTLE_MS: int = int(os.getenv("SCREENSHOT_THROTTLE_MS", "500"))
+
+    # Model settings (defaults from environment)
+    MODEL_BASE_URL: str = os.getenv("PHONE_AGENT_BASE_URL", "http://localhost:8000/v1")
+    MODEL_NAME: str = os.getenv("PHONE_AGENT_MODEL", "autoglm-phone-9b")
+    MODEL_API_KEY: str = os.getenv("PHONE_AGENT_API_KEY", "EMPTY")
+
+    # Task history
+    MAX_TASK_HISTORY: int = int(os.getenv("MAX_TASK_HISTORY", "100"))
+
+    class Config:
+        env_file = ".env"
+        env_file_encoding = "utf-8"
+
+
+# Global config instance
+config = DashboardConfig()

dashboard/dependencies.py

@@ -0,0 +1,39 @@
+"""
+Dependency injection for the dashboard.
+"""
+
+from typing import AsyncGenerator
+
+from dashboard.config import config
+from dashboard.services.device_manager import DeviceManager
+from dashboard.services.task_executor import TaskExecutor
+from dashboard.services.websocket_manager import WebSocketManager
+
+# Global service instances
+_device_manager: DeviceManager | None = None
+_task_executor: TaskExecutor | None = None
+_ws_manager: WebSocketManager | None = None
+
+
+def get_device_manager() -> DeviceManager:
+    """Get the device manager instance."""
+    global _device_manager
+    if _device_manager is None:
+        _device_manager = DeviceManager(device_type=config.DEFAULT_DEVICE_TYPE)
+    return _device_manager
+
+
+def get_task_executor() -> TaskExecutor:
+    """Get the task executor instance."""
+    global _task_executor
+    if _task_executor is None:
+        _task_executor = TaskExecutor(get_device_manager())
+    return _task_executor
+
+
+def get_ws_manager() -> WebSocketManager:
+    """Get the WebSocket manager instance."""
+    global _ws_manager
+    if _ws_manager is None:
+        _ws_manager = WebSocketManager()
+    return _ws_manager

dashboard/main.py

@@ -0,0 +1,175 @@
+"""
+AutoGLM Dashboard - FastAPI Main Application.
+
+This is the main entry point for the web dashboard.
+Run with: uvicorn dashboard.main:app --host 0.0.0.0 --port 8080 --reload
+"""
+
+import os
+from contextlib import asynccontextmanager
+from datetime import datetime
+from pathlib import Path
+
+from dotenv import load_dotenv
+from fastapi import FastAPI, Request
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import FileResponse, JSONResponse
+from fastapi.staticfiles import StaticFiles
+
+from dashboard.api import devices_router, tasks_router, websocket_router
+from dashboard.config import config
+from dashboard.dependencies import (
+    get_device_manager,
+    get_task_executor,
+    get_ws_manager,
+)
+from dashboard.services.device_manager import DeviceManager
+from dashboard.services.task_executor import TaskExecutor
+from dashboard.services.websocket_manager import WebSocketManager
+
+# Load .env file
+load_dotenv()
+
+
+@asynccontextmanager
+async def lifespan(app: FastAPI):
+    """Application lifespan manager.
+
+    Handles startup and shutdown events.
+    """
+    # Startup
+    print("=" * 50)
+    print("AutoGLM Dashboard Starting...")
+    print("=" * 50)
+
+    # Initialize services
+    device_manager = get_device_manager()
+    task_executor = get_task_executor()
+    ws_manager = get_ws_manager()
+
+    # Link services
+    task_executor.set_ws_manager(ws_manager)
+
+    # Scan for devices on startup
+    print("Scanning for devices...")
+    try:
+        devices = await device_manager.refresh_devices()
+        print(f"Found {len(devices)} device(s)")
+        for device in devices:
+            status = "connected" if device.is_connected else "disconnected"
+            print(f"  - {device.device_id} ({status})")
+    except Exception as e:
+        print(f"Error scanning devices: {e}")
+
+    print("=" * 50)
+    print(f"Dashboard running on http://{config.HOST}:{config.PORT}")
+    print(f"API docs: http://{config.HOST}:{config.PORT}/docs")
+    print("=" * 50)
+
+    yield
+
+    # Shutdown
+    print("Shutting down dashboard...")
+
+
+# Create FastAPI app
+app = FastAPI(
+    title="AutoGLM Dashboard",
+    description="Web-based multi-device control interface for AutoGLM",
+    version="0.1.0",
+    lifespan=lifespan,
+)
+
+# Configure CORS for LAN access
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=config.CORS_ORIGINS,
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+
+# Exception handlers
+@app.exception_handler(Exception)
+async def global_exception_handler(request: Request, exc: Exception):
+    """Global exception handler."""
+    return JSONResponse(
+        status_code=500,
+        content={"error": str(exc), "type": type(exc).__name__},
+    )
+
+
+# Include routers
+app.include_router(devices_router, prefix="/api")
+app.include_router(tasks_router, prefix="/api")
+app.include_router(websocket_router)
+
+
+# Health check
+@app.get("/health")
+async def health_check():
+    """Health check endpoint."""
+    device_manager = get_device_manager()
+    task_executor = get_task_executor()
+    ws_manager = get_ws_manager()
+
+    devices = device_manager.list_all_devices()
+    connected_devices = sum(1 for d in devices if d.is_connected)
+
+    return {
+        "status": "healthy",
+        "timestamp": datetime.now().isoformat(),
+        "devices": {
+            "total": len(devices),
+            "connected": connected_devices,
+        },
+        "tasks": {
+            "active": task_executor.get_active_task_count(),
+        },
+        "websocket": {
+            "connections": ws_manager.get_connection_count(),
+        },
+    }
+
+
+# Root endpoint - serve dashboard or API info
+@app.get("/")
+async def root():
+    """Root endpoint - returns API info or serves dashboard."""
+    # Check if static files exist
+    static_path = Path(__file__).parent / "static" / "index.html"
+
+    if static_path.exists():
+        return FileResponse(static_path)
+
+    # Return API info if dashboard not built
+    return {
+        "name": "AutoGLM Dashboard API",
+        "version": "0.1.0",
+        "docs": "/docs",
+        "endpoints": {
+            "devices": "/api/devices",
+            "tasks": "/api/tasks",
+            "websocket": "/ws",
+            "health": "/health",
+        },
+    }
+
+
+# Mount static files for dashboard (if exists)
+static_path = Path(__file__).parent / "static"
+if static_path.exists():
+    app.mount("/static", StaticFiles(directory=str(static_path)), name="static")
+
+
+# Run script entry point
+if __name__ == "__main__":
+    import uvicorn
+
+    uvicorn.run(
+        "dashboard.main:app",
+        host=config.HOST,
+        port=config.PORT,
+        reload=config.DEBUG,
+    )

dashboard/models/__init__.py

@@ -0,0 +1,37 @@
+"""
+Data models for the dashboard API.
+
+Includes Pydantic schemas for devices, tasks, and WebSocket messages.
+"""
+
+from dashboard.models.device import (
+    DeviceType,
+    DeviceStatus,
+    DeviceSchema,
+    DeviceInfo,
+)
+from dashboard.models.task import (
+    TaskStatus,
+    TaskRequest,
+    TaskSchema,
+    TaskCreateRequest,
+)
+from dashboard.models.ws_messages import (
+    WSMessageType,
+    WSMessage,
+    StepUpdate,
+)
+
+__all__ = [
+    "DeviceType",
+    "DeviceStatus",
+    "DeviceSchema",
+    "DeviceInfo",
+    "TaskStatus",
+    "TaskRequest",
+    "TaskSchema",
+    "TaskCreateRequest",
+    "WSMessageType",
+    "WSMessage",
+    "StepUpdate",
+]

dashboard/models/device.py

@@ -0,0 +1,67 @@
+"""
+Device data models for the dashboard.
+"""
+
+from datetime import datetime
+from enum import Enum
+from typing import Optional
+
+from pydantic import BaseModel, Field
+
+
+class DeviceType(str, Enum):
+    """Device connection type."""
+
+    ADB = "adb"
+    HDC = "hdc"
+    IOS = "ios"
+
+
+class DeviceStatus(str, Enum):
+    """Device status."""
+
+    ONLINE = "online"
+    OFFLINE = "offline"
+    BUSY = "busy"
+    ERROR = "error"
+
+
+class DeviceSchema(BaseModel):
+    """Device schema for API responses."""
+
+    device_id: str = Field(..., description="Unique device identifier")
+    status: DeviceStatus = Field(default=DeviceStatus.OFFLINE, description="Device status")
+    device_type: DeviceType = Field(..., description="Device connection type")
+    model: Optional[str] = Field(None, description="Device model name")
+    android_version: Optional[str] = Field(None, description="Android/iOS version")
+    current_app: Optional[str] = Field(None, description="Currently active app")
+    last_seen: datetime = Field(default_factory=datetime.now, description="Last connection time")
+    is_connected: bool = Field(True, description="Whether device is connected")
+    screenshot: Optional[str] = Field(None, description="Base64 encoded screenshot")
+
+    class Config:
+        json_schema_extra = {
+            "example": {
+                "device_id": "emulator-5554",
+                "status": "online",
+                "device_type": "adb",
+                "model": "sdk_gphone64_x86_64",
+                "android_version": "14",
+                "current_app": "com.android.launcher3",
+                "is_connected": True,
+            }
+        }
+
+
+class DeviceInfo(BaseModel):
+    """Extended device information."""
+
+    device_id: str
+    status: DeviceStatus
+    device_type: DeviceType
+    model: Optional[str] = None
+    android_version: Optional[str] = None
+    current_app: Optional[str] = None
+    last_seen: datetime
+    screenshot: Optional[str] = None
+    is_connected: bool = True

dashboard/models/task.py

@@ -0,0 +1,81 @@
+"""
+Task data models for the dashboard.
+"""
+
+from datetime import datetime
+from enum import Enum
+from typing import Any, Dict, Optional
+
+from pydantic import BaseModel, Field
+
+from phone_agent.model import ModelConfig
+
+
+class TaskStatus(str, Enum):
+    """Task execution status."""
+
+    PENDING = "pending"
+    RUNNING = "running"
+    COMPLETED = "completed"
+    FAILED = "failed"
+    STOPPED = "stopped"
+
+
+class TaskCreateRequest(BaseModel):
+    """Request to create a new task."""
+
+    device_id: str = Field(..., description="Target device ID")
+    task: str = Field(..., description="Task description")
+    max_steps: int = Field(100, description="Maximum execution steps")
+    lang: str = Field("cn", description="Language (cn or en)")
+
+    # Model config - use dict to avoid validation issues with ModelConfig
+    base_url: str = Field(
+        default="http://localhost:8000/v1", description="Model API base URL"
+    )
+    model_name: str = Field(default="autoglm-phone-9b", description="Model name")
+    api_key: str = Field(default="EMPTY", description="API key")
+    max_tokens: int = Field(default=3000, description="Max tokens per response")
+    temperature: float = Field(default=0.0, description="Sampling temperature")
+    top_p: float = Field(default=0.85, description="Top-p sampling parameter")
+    frequency_penalty: float = Field(default=0.2, description="Frequency penalty")
+
+
+class TaskSchema(BaseModel):
+    """Task schema for API responses."""
+
+    task_id: str = Field(..., description="Unique task identifier")
+    device_id: str = Field(..., description="Target device ID")
+    task: str = Field(..., description="Task description")
+    status: TaskStatus = Field(..., description="Task status")
+    current_step: int = Field(0, description="Current step number")
+    max_steps: int = Field(100, description="Maximum steps")
+    current_action: Optional[Dict[str, Any]] = Field(None, description="Current action")
+    thinking: Optional[str] = Field(None, description="Current thinking/reasoning")
+    started_at: datetime = Field(..., description="Task start time")
+    updated_at: datetime = Field(..., description="Last update time")
+    finished_at: Optional[datetime] = Field(None, description="Task completion time")
+    error: Optional[str] = Field(None, description="Error message if failed")
+    completion_message: Optional[str] = Field(
+        None, description="Task completion message with details"
+    )
+
+    class Config:
+        json_schema_extra = {
+            "example": {
+                "task_id": "task_123456",
+                "device_id": "emulator-5554",
+                "task": "Open WeChat",
+                "status": "running",
+                "current_step": 3,
+                "max_steps": 100,
+                "current_action": {"action": "Tap", "element": "WeChat icon"},
+                "thinking": "Looking for WeChat icon on home screen",
+                "started_at": "2024-01-09T10:00:00",
+                "updated_at": "2024-01-09T10:00:15",
+            }
+        }
+
+
+# For backward compatibility
+TaskRequest = TaskCreateRequest

dashboard/models/ws_messages.py

@@ -0,0 +1,70 @@
+"""
+WebSocket message models for real-time updates.
+"""
+
+from datetime import datetime
+from enum import Enum
+from typing import Any, Dict, Optional
+
+from pydantic import BaseModel, Field
+
+
+class WSMessageType(str, Enum):
+    """WebSocket message types."""
+
+    DEVICE_UPDATE = "device_update"
+    TASK_STARTED = "task_started"
+    TASK_STEP = "task_step"
+    TASK_COMPLETED = "task_completed"
+    TASK_FAILED = "task_failed"
+    TASK_STOPPED = "task_stopped"
+    SCREENSHOT = "screenshot"
+    ERROR = "error"
+    PING = "ping"
+    PONG = "pong"
+
+
+class WSMessage(BaseModel):
+    """Base WebSocket message."""
+
+    type: WSMessageType = Field(..., description="Message type")
+    data: Dict[str, Any] = Field(default_factory=dict, description="Message data")
+    timestamp: datetime = Field(default_factory=datetime.now, description="Message timestamp")
+
+    class Config:
+        json_schema_extra = {
+            "example": {
+                "type": "task_step",
+                "data": {
+                    "task_id": "task_123",
+                    "device_id": "emulator-5554",
+                    "step": 5,
+                    "action": {"action": "Tap"},
+                    "thinking": "Tapping on button",
+                },
+                "timestamp": "2024-01-09T10:00:00",
+            }
+        }
+
+
+class StepUpdate(BaseModel):
+    """Step update message data."""
+
+    task_id: str = Field(..., description="Task ID")
+    device_id: str = Field(..., description="Device ID")
+    step: int = Field(..., description="Step number")
+    action: Optional[Dict[str, Any]] = Field(None, description="Action taken")
+    thinking: Optional[str] = Field(None, description="AI reasoning")
+    finished: bool = Field(False, description="Whether task is finished")
+    success: bool = Field(True, description="Whether step succeeded")
+    message: Optional[str] = Field(None, description="Status message")
+
+
+class ScreenshotUpdate(BaseModel):
+    """Screenshot update message data."""
+
+    device_id: str = Field(..., description="Device ID")
+    screenshot: str = Field(..., description="Base64 encoded screenshot")
+    width: int = Field(..., description="Screenshot width")
+    height: int = Field(..., description="Screenshot height")
+    timestamp: datetime = Field(default_factory=datetime.now)

dashboard/services/__init__.py

@@ -0,0 +1,15 @@
+"""
+Services for the dashboard.
+
+Includes device management, task execution, and WebSocket management.
+"""
+
+from dashboard.services.device_manager import DeviceManager
+from dashboard.services.task_executor import TaskExecutor
+from dashboard.services.websocket_manager import WebSocketManager
+
+__all__ = [
+    "DeviceManager",
+    "TaskExecutor",
+    "WebSocketManager",
+]

dashboard/services/device_manager.py

@@ -0,0 +1,336 @@
+"""
+Device manager for handling device pool and connections.
+"""
+
+import asyncio
+from datetime import datetime
+from enum import Enum
+from threading import Lock
+from typing import Dict, List, Optional
+
+from phone_agent.adb.connection import ADBConnection
+from phone_agent.adb.screenshot import get_screenshot as adb_screenshot
+from phone_agent.device_factory import (
+    DeviceFactory,
+    DeviceType,
+    get_device_factory,
+    set_device_type,
+)
+from phone_agent.hdc.connection import HDCConnection
+from phone_agent.hdc.screenshot import get_screenshot as hdc_screenshot
+
+from dashboard.models.device import DeviceInfo, DeviceStatus, DeviceSchema
+
+
+class DeviceConnectionError(Exception):
+    """Raised when device connection fails."""
+
+    pass
+
+
+class DeviceManager:
+    """Manage device pool and connections."""
+
+    def __init__(self, device_type: str = "adb"):
+        """Initialize the device manager.
+
+        Args:
+            device_type: Default device type (adb, hdc, ios)
+        """
+        self._devices: Dict[str, DeviceInfo] = {}
+        self._device_locks: Dict[str, Lock] = {}
+        self._device_type = self._parse_device_type(device_type)
+        self._screenshot_cache: Dict[str, tuple[str, datetime]] = {}  # (base64, timestamp)
+        self._cache_ttl_seconds = 2.0  # Cache screenshots for 2 seconds
+
+        # Set the global device type
+        set_device_type(self._device_type)
+
+    def _parse_device_type(self, device_type: str) -> DeviceType:
+        """Parse device type string to enum."""
+        try:
+            return DeviceType[device_type.upper()]
+        except KeyError:
+            return DeviceType.ADB
+
+    @property
+    def factory(self) -> DeviceFactory:
+        """Get the device factory instance."""
+        return get_device_factory()
+
+    async def refresh_devices(self) -> List[DeviceInfo]:
+        """Scan and return all connected devices.
+
+        Returns:
+            List of device info objects
+        """
+        try:
+            # Run device listing in thread pool to avoid blocking
+            loop = asyncio.get_event_loop()
+            devices = await loop.run_in_executor(None, self.factory.list_devices)
+
+            # Update device cache
+            current_time = datetime.now()
+            for device in devices:
+                device_id = device.device_id
+
+                # Initialize lock if needed
+                if device_id not in self._device_locks:
+                    self._device_locks[device_id] = Lock()
+
+                # Update or create device info
+                if device_id in self._devices:
+                    # Update existing device
+                    self._devices[device_id].last_seen = current_time
+                    self._devices[device_id].is_connected = True
+                    self._devices[device_id].status = (
+                        DeviceStatus.BUSY
+                        if not self._is_device_available(device_id)
+                        else DeviceStatus.ONLINE
+                    )
+                else:
+                    # New device
+                    self._devices[device_id] = DeviceInfo(
+                        device_id=device_id,
+                        status=DeviceStatus.ONLINE,
+                        device_type=self._device_type_to_schema(device.connection_type),
+                        model=device.model,
+                        android_version=device.android_version,
+                        current_app=None,
+                        last_seen=current_time,
+                        is_connected=True,
+                    )
+
+            # Mark disconnected devices
+            connected_ids = {d.device_id for d in devices}
+            for device_id, device_info in self._devices.items():
+                if device_id not in connected_ids:
+                    device_info.is_connected = False
+                    device_info.status = DeviceStatus.OFFLINE
+
+            return list(self._devices.values())
+
+        except Exception as e:
+            raise DeviceConnectionError(f"Failed to refresh devices: {e}")
+
+    async def get_device(self, device_id: str) -> Optional[DeviceInfo]:
+        """Get device info by ID.
+
+        Args:
+            device_id: Device identifier
+
+        Returns:
+            Device info or None if not found
+        """
+        return self._devices.get(device_id)
+
+    def acquire_device(self, device_id: str) -> bool:
+        """Acquire lock on device.
+
+        Args:
+            device_id: Device identifier
+
+        Returns:
+            True if acquired, False if device is busy
+        """
+        if device_id not in self._device_locks:
+            return False
+
+        lock = self._device_locks[device_id]
+        acquired = lock.acquire(blocking=False)
+
+        if acquired:
+            # Update device status
+            if device_id in self._devices:
+                self._devices[device_id].status = DeviceStatus.BUSY
+
+        return acquired
+
+    def release_device(self, device_id: str):
+        """Release device lock.
+
+        Args:
+            device_id: Device identifier
+        """
+        if device_id in self._device_locks:
+            self._device_locks[device_id].release()
+
+            # Update device status
+            if device_id in self._devices:
+                device = self._devices[device_id]
+                if device.is_connected:
+                    device.status = DeviceStatus.ONLINE
+                else:
+                    device.status = DeviceStatus.OFFLINE
+
+    def is_device_available(self, device_id: str) -> bool:
+        """Check if device is available for task execution.
+
+        Args:
+            device_id: Device identifier
+
+        Returns:
+            True if available, False if busy or offline
+        """
+        if device_id not in self._devices:
+            return False
+
+        device = self._devices[device_id]
+        if not device.is_connected:
+            return False
+
+        if device.status == DeviceStatus.BUSY:
+            return False
+
+        return True
+
+    def _is_device_available(self, device_id: str) -> bool:
+        """Internal check if device lock is available."""
+        if device_id not in self._device_locks:
+            return True
+        lock = self._device_locks[device_id]
+        return not lock.locked()
+
+    async def get_screenshot(self, device_id: str) -> Optional[str]:
+        """Get screenshot for device.
+
+        Args:
+            device_id: Device identifier
+
+        Returns:
+            Base64 encoded screenshot or None
+        """
+        # Check cache
+        if device_id in self._screenshot_cache:
+            screenshot, timestamp = self._screenshot_cache[device_id]
+            age = (datetime.now() - timestamp).total_seconds()
+            if age < self._cache_ttl_seconds:
+                return screenshot
+
+        try:
+            loop = asyncio.get_event_loop()
+
+            if self._device_type == DeviceType.HDC:
+                result = await loop.run_in_executor(
+                    None, hdc_screenshot, device_id, 10
+                )
+            else:  # ADB or default
+                result = await loop.run_in_executor(
+                    None, adb_screenshot, device_id, 10
+                )
+
+            if result:
+                # Cache the screenshot
+                self._screenshot_cache[device_id] = (result.base64_data, datetime.now())
+                return result.base64_data
+
+        except Exception:
+            pass
+
+        return None
+
+    async def get_current_app(self, device_id: str) -> Optional[str]:
+        """Get current app for device.
+
+        Args:
+            device_id: Device identifier
+
+        Returns:
+            Current app package name or None
+        """
+        try:
+            loop = asyncio.get_event_loop()
+            app = await loop.run_in_executor(
+                None, self.factory.get_current_app, device_id
+            )
+
+            # Update device info
+            if device_id in self._devices and app:
+                self._devices[device_id].current_app = app
+
+            return app
+
+        except Exception:
+            return None
+
+    async def connect_device(self, address: str) -> bool:
+        """Connect to device via WiFi.
+
+        Args:
+            address: Device address (IP:PORT)
+
+        Returns:
+            True if connected successfully
+        """
+        try:
+            loop = asyncio.get_event_loop()
+
+            if self._device_type == DeviceType.HDC:
+                conn = HDCConnection()
+            else:
+                conn = ADBConnection()
+
+            success, _ = await loop.run_in_executor(None, conn.connect, address)
+
+            if success:
+                # Refresh devices after connecting
+                await self.refresh_devices()
+
+            return success
+
+        except Exception:
+            return False
+
+    async def disconnect_device(self, address: str) -> bool:
+        """Disconnect from device.
+
+        Args:
+            address: Device address (IP:PORT)
+
+        Returns:
+            True if disconnected successfully
+        """
+        try:
+            loop = asyncio.get_event_loop()
+
+            if self._device_type == DeviceType.HDC:
+                conn = HDCConnection()
+            else:
+                conn = ADBConnection()
+
+            success, _ = await loop.run_in_executor(None, conn.disconnect, address)
+
+            if success:
+                # Refresh devices after disconnecting
+                await self.refresh_devices()
+
+            return success
+
+        except Exception:
+            return False
+
+    def _device_type_to_schema(self, connection_type) -> "DeviceSchema":
+        """Convert connection type to DeviceType enum."""
+        from dashboard.models.device import DeviceType as SchemaDeviceType
+
+        # Mapping from phone_agent connection types to schema types
+        type_map = {
+            "USB": SchemaDeviceType.ADB,
+            "WIFI": SchemaDeviceType.ADB,
+            "REMOTE": SchemaDeviceType.ADB,
+        }
+
+        # Try to get string value from enum
+        try:
+            type_str = connection_type.value if hasattr(connection_type, "value") else str(connection_type)
+            return type_map.get(type_str.upper(), SchemaDeviceType.ADB)
+        except (AttributeError, KeyError):
+            return SchemaDeviceType.ADB
+
+    def list_all_devices(self) -> List[DeviceInfo]:
+        """Get all cached devices.
+
+        Returns:
+            List of all device info
+        """
+        return list(self._devices.values())

dashboard/services/task_executor.py

@@ -0,0 +1,410 @@
+"""
+Task executor for running tasks on devices with thread pool.
+"""
+
+import asyncio
+import threading
+import uuid
+from concurrent.futures import ThreadPoolExecutor, Future
+from copy import deepcopy
+from dataclasses import dataclass, field
+from datetime import datetime
+from typing import Callable, Dict, Optional
+
+from phone_agent import AgentConfig, PhoneAgent
+from phone_agent.agent import StepResult
+from phone_agent.model import ModelConfig
+
+from dashboard.config import config
+from dashboard.models.task import TaskCreateRequest, TaskSchema, TaskStatus
+from dashboard.services.device_manager import DeviceManager
+from dashboard.services.websocket_manager import WebSocketManager
+
+
+def _run_async(coro):
+    """Run async coroutine from sync context and wait for completion."""
+    try:
+        loop = asyncio.get_running_loop()
+        # We're already in an async context, use run_coroutine_threadsafe
+        import concurrent.futures
+        future = asyncio.run_coroutine_threadsafe(coro, loop)
+        # Wait for completion with timeout to avoid hanging
+        try:
+            future.result(timeout=5)
+        except concurrent.futures.TimeoutError:
+            pass
+        except Exception:
+            pass
+    except RuntimeError:
+        # No running loop, use asyncio.run
+        asyncio.run(coro)
+
+
+@dataclass
+class ActiveTask:
+    """An active task being executed."""
+
+    task_id: str
+    device_id: str
+    task: str
+    status: TaskStatus
+    current_step: int
+    max_steps: int
+    started_at: datetime
+    updated_at: datetime
+    finished_at: Optional[datetime] = None
+    future: Optional[Future] = None
+    stop_requested: bool = False
+    current_action: Optional[dict] = None
+    thinking: Optional[str] = None
+    error: Optional[str] = None
+    completion_message: Optional[str] = None
+
+
+class TaskExecutor:
+    """Execute tasks on devices with thread pool."""
+
+    def __init__(
+        self,
+        device_manager: DeviceManager,
+        ws_manager: Optional[WebSocketManager] = None,
+        max_workers: int = 10,
+    ):
+        """Initialize the task executor.
+
+        Args:
+            device_manager: Device manager instance
+            ws_manager: WebSocket manager for real-time updates (optional)
+            max_workers: Maximum number of concurrent tasks
+        """
+        self.device_manager = device_manager
+        self.ws_manager = ws_manager
+        self.executor = ThreadPoolExecutor(max_workers=max_workers)
+        self.active_tasks: Dict[str, ActiveTask] = {}
+        self.task_history: Dict[str, TaskSchema] = {}
+        self._lock = threading.Lock()
+
+    def set_ws_manager(self, ws_manager: WebSocketManager):
+        """Set the WebSocket manager.
+
+        Args:
+            ws_manager: WebSocket manager instance
+        """
+        self.ws_manager = ws_manager
+
+    async def execute_task(self, request: TaskCreateRequest) -> str:
+        """Execute task on device.
+
+        Args:
+            request: Task creation request
+
+        Returns:
+            Task ID
+        """
+        task_id = f"task_{uuid.uuid4().hex[:8]}"
+
+        # Create active task
+        active_task = ActiveTask(
+            task_id=task_id,
+            device_id=request.device_id,
+            task=request.task,
+            status=TaskStatus.RUNNING,
+            current_step=0,
+            max_steps=request.max_steps,
+            started_at=datetime.now(),
+            updated_at=datetime.now(),
+        )
+
+        with self._lock:
+            self.active_tasks[task_id] = active_task
+
+        # Notify WebSocket
+        if self.ws_manager:
+            await self.ws_manager.broadcast_task_started(task_id, request.dict())
+
+        # Submit to thread pool
+        future = self.executor.submit(
+            self._run_task,
+            task_id,
+            request,
+        )
+        active_task.future = future
+
+        return task_id
+
+    def _run_task(self, task_id: str, request: TaskCreateRequest):
+        """Run task in thread pool.
+
+        Args:
+            task_id: Task ID
+            request: Task creation request
+        """
+        import os
+
+        # Get model config from request
+        model_config = ModelConfig(
+            base_url=request.base_url,
+            model_name=request.model_name,
+            api_key=request.api_key,
+            max_tokens=request.max_tokens,
+            temperature=request.temperature,
+            top_p=request.top_p,
+            frequency_penalty=request.frequency_penalty,
+            lang=request.lang,
+        )
+
+        # Get agent config
+        agent_config = AgentConfig(
+            max_steps=request.max_steps,
+            device_id=request.device_id,
+            lang=request.lang,
+            step_callback=lambda result: self._step_callback(task_id, result),
+            before_action_callback=lambda action: self._before_action_callback(
+                task_id, action
+            ),
+        )
+
+        # Create agent
+        agent = PhoneAgent(model_config=model_config, agent_config=agent_config)
+
+        try:
+            # Acquire device
+            if not self.device_manager.acquire_device(request.device_id):
+                raise Exception(f"Device {request.device_id} is not available")
+
+            try:
+                # Run task
+                result = agent.run(request.task)
+
+                # Update task status
+                with self._lock:
+                    if task_id in self.active_tasks:
+                        task = self.active_tasks[task_id]
+                        task.status = (
+                            TaskStatus.COMPLETED if result else TaskStatus.FAILED
+                        )
+                        task.finished_at = datetime.now()
+                        task.updated_at = datetime.now()
+
+            finally:
+                # Release device
+                self.device_manager.release_device(request.device_id)
+
+                # Broadcast device status update
+                if self.ws_manager:
+                    device = self.device_manager._devices.get(request.device_id)
+                    if device:
+                        _run_async(
+                            self.ws_manager.broadcast_device_update(
+                                request.device_id,
+                                {
+                                    "status": device.status.value,
+                                    "is_connected": device.is_connected,
+                                    "current_app": device.current_app,
+                                },
+                            )
+                        )
+
+            # Notify completion
+            if self.ws_manager:
+                with self._lock:
+                    task = self.active_tasks.get(task_id)
+                    task_status = task.status if task else TaskStatus.COMPLETED
+                    message = result.message if hasattr(result, "message") else None
+
+                _run_async(
+                    self.ws_manager.broadcast_task_completed(
+                        task_id,
+                        request.device_id,
+                        task_status,
+                        message,
+                    )
+                )
+
+        except Exception as e:
+            # Release device on error
+            self.device_manager.release_device(request.device_id)
+
+            # Broadcast device status update
+            if self.ws_manager:
+                device = self.device_manager._devices.get(request.device_id)
+                if device:
+                    _run_async(
+                        self.ws_manager.broadcast_device_update(
+                            request.device_id,
+                            {
+                                "status": device.status.value,
+                                "is_connected": device.is_connected,
+                            },
+                        )
+                    )
+
+            # Update task status
+            with self._lock:
+                if task_id in self.active_tasks:
+                    task = self.active_tasks[task_id]
+                    task.status = TaskStatus.FAILED
+                    task.error = str(e)
+                    task.finished_at = datetime.now()
+                    task.updated_at = datetime.now()
+
+            # Notify error
+            if self.ws_manager:
+                _run_async(
+                    self.ws_manager.broadcast_task_failed(
+                        task_id, request.device_id, str(e)
+                    )
+                )
+
+        finally:
+            # Move to history
+            with self._lock:
+                if task_id in self.active_tasks:
+                    active_task = self.active_tasks.pop(task_id)
+                    task_schema = self._active_task_to_schema(active_task)
+                    self.task_history[task_id] = task_schema
+
+                    # Trim history
+                    if len(self.task_history) > config.MAX_TASK_HISTORY:
+                        oldest = min(self.task_history.items(), key=lambda x: x[1].started_at)
+                        del self.task_history[oldest[0]]
+
+    def _step_callback(
+        self, task_id: str, result: StepResult
+    ) -> Optional[str]:
+        """Callback after each step.
+
+        Args:
+            task_id: Task ID
+            result: Step result
+
+        Returns:
+            "stop" to interrupt, new task to switch, None to continue
+        """
+        with self._lock:
+            if task_id not in self.active_tasks:
+                return None
+
+            task = self.active_tasks[task_id]
+
+            # Check if stop was requested
+            if task.stop_requested:
+                return "stop"
+
+            # Update task state
+            task.current_step = result.step_count
+            task.updated_at = datetime.now()
+            task.thinking = result.thinking
+            task.current_action = result.action
+
+            # Store completion message when finished
+            if result.finished and result.message:
+                task.completion_message = result.message
+
+        # Notify WebSocket
+        if self.ws_manager:
+            _run_async(
+                self.ws_manager.broadcast_step_update(
+                    task_id=task_id,
+                    device_id=task.device_id,
+                    step=result.step_count,
+                    action=result.action,
+                    thinking=result.thinking,
+                    finished=result.finished,
+                    success=result.success,
+                    message=result.message,
+                )
+            )
+
+        return None
+
+    def _before_action_callback(self, task_id: str, action: dict) -> Optional[dict]:
+        """Callback before executing action.
+
+        Args:
+            task_id: Task ID
+            action: Action to execute
+
+        Returns:
+            Modified action or None to proceed as-is
+        """
+        # Can be used for action validation/logging
+        return None
+
+    async def stop_task(self, task_id: str):
+        """Stop running task.
+
+        Args:
+            task_id: Task ID
+        """
+        with self._lock:
+            if task_id not in self.active_tasks:
+                return
+
+            task = self.active_tasks[task_id]
+            task.stop_requested = True
+
+    async def get_task_status(self, task_id: str) -> Optional[TaskSchema]:
+        """Get task status.
+
+        Args:
+            task_id: Task ID
+
+        Returns:
+            Task schema or None if not found
+        """
+        with self._lock:
+            if task_id in self.active_tasks:
+                return self._active_task_to_schema(self.active_tasks[task_id])
+            elif task_id in self.task_history:
+                return self.task_history[task_id]
+
+        return None
+
+    async def list_tasks(self) -> list[TaskSchema]:
+        """List all tasks (active and recent).
+
+        Returns:
+            List of task schemas
+        """
+        with self._lock:
+            active_schemas = [
+                self._active_task_to_schema(t) for t in self.active_tasks.values()
+            ]
+            history_schemas = list(self.task_history.values())
+
+        return active_schemas + history_schemas
+
+    def _active_task_to_schema(self, active_task: ActiveTask) -> TaskSchema:
+        """Convert ActiveTask to TaskSchema.
+
+        Args:
+            active_task: Active task
+
+        Returns:
+            Task schema
+        """
+        return TaskSchema(
+            task_id=active_task.task_id,
+            device_id=active_task.device_id,
+            task=active_task.task,
+            status=active_task.status,
+            current_step=active_task.current_step,
+            max_steps=active_task.max_steps,
+            current_action=active_task.current_action,
+            thinking=active_task.thinking,
+            started_at=active_task.started_at,
+            updated_at=active_task.updated_at,
+            finished_at=active_task.finished_at,
+            error=active_task.error,
+            completion_message=active_task.completion_message,
+        )
+
+    def get_active_task_count(self) -> int:
+        """Get number of active tasks.
+
+        Returns:
+            Active task count
+        """
+        with self._lock:
+            return len(self.active_tasks)

dashboard/services/websocket_manager.py

@@ -0,0 +1,338 @@
+"""
+WebSocket manager for real-time updates.
+"""
+
+import asyncio
+import json
+import uuid
+from datetime import datetime
+from typing import Any, Dict, Optional
+
+from fastapi import WebSocket
+
+from dashboard.models.ws_messages import (
+    ScreenshotUpdate,
+    WSMessage,
+    WSMessageType,
+)
+
+
+class WebSocketManager:
+    """Manage WebSocket connections for real-time updates."""
+
+    def __init__(self):
+        """Initialize the WebSocket manager."""
+        self.active_connections: Dict[str, WebSocket] = {}
+        self.client_subscriptions: Dict[str, set[str]] = {}  # client_id -> set of device_ids
+
+    async def connect(self, websocket: WebSocket, client_id: Optional[str] = None) -> str:
+        """Accept and store connection.
+
+        Args:
+            websocket: WebSocket connection
+            client_id: Optional client ID (auto-generated if not provided)
+
+        Returns:
+            Client ID
+        """
+        await websocket.accept()
+
+        if client_id is None:
+            client_id = f"client_{uuid.uuid4().hex[:8]}"
+
+        self.active_connections[client_id] = websocket
+        self.client_subscriptions[client_id] = set()
+
+        return client_id
+
+    def disconnect(self, client_id: str):
+        """Remove connection.
+
+        Args:
+            client_id: Client ID
+        """
+        if client_id in self.active_connections:
+            del self.active_connections[client_id]
+
+        if client_id in self.client_subscriptions:
+            del self.client_subscriptions[client_id]
+
+    async def send_to_client(
+        self, client_id: str, message: WSMessage | Dict[str, Any]
+    ):
+        """Send message to specific client.
+
+        Args:
+            client_id: Client ID
+            message: Message to send
+        """
+        if client_id not in self.active_connections:
+            return
+
+        websocket = self.active_connections[client_id]
+
+        # Convert dict to WSMessage if needed
+        if isinstance(message, dict):
+            message = WSMessage(
+                type=WSMessageType(message.get("type", "error")),
+                data=message.get("data", {}),
+                timestamp=message.get("timestamp", datetime.now()),
+            )
+
+        try:
+            await websocket.send_json(message.model_dump(mode="json"))
+        except Exception:
+            # Connection may be closed
+            self.disconnect(client_id)
+
+    async def broadcast(self, message: WSMessage | Dict[str, Any]):
+        """Broadcast message to all connected clients.
+
+        Args:
+            message: Message to broadcast
+        """
+        # Convert dict to WSMessage if needed
+        if isinstance(message, dict):
+            message = WSMessage(
+                type=WSMessageType(message.get("type", "error")),
+                data=message.get("data", {}),
+                timestamp=message.get("timestamp", datetime.now()),
+            )
+
+        # Create list of clients to avoid modification during iteration
+        clients = list(self.active_connections.items())
+
+        for client_id, websocket in clients:
+            try:
+                await websocket.send_json(message.model_dump(mode="json"))
+            except Exception:
+                self.disconnect(client_id)
+
+    async def broadcast_to_device_subscribers(
+        self, device_id: str, message: WSMessage | Dict[str, Any]
+    ):
+        """Broadcast message to clients subscribed to a device.
+
+        Args:
+            device_id: Device ID
+            message: Message to broadcast
+        """
+        # Convert dict to WSMessage if needed
+        if isinstance(message, dict):
+            message = WSMessage(
+                type=WSMessageType(message.get("type", "error")),
+                data=message.get("data", {}),
+                timestamp=message.get("timestamp", datetime.now()),
+            )
+
+        # Find clients subscribed to this device
+        for client_id, subscriptions in self.client_subscriptions.items():
+            if device_id in subscriptions or "*" in subscriptions:
+                await self.send_to_client(client_id, message)
+
+    def subscribe_to_device(self, client_id: str, device_id: str):
+        """Subscribe client to device updates.
+
+        Args:
+            client_id: Client ID
+            device_id: Device ID (use "*" for all devices)
+        """
+        if client_id in self.client_subscriptions:
+            self.client_subscriptions[client_id].add(device_id)
+
+    def unsubscribe_from_device(self, client_id: str, device_id: str):
+        """Unsubscribe client from device updates.
+
+        Args:
+            client_id: Client ID
+            device_id: Device ID
+        """
+        if client_id in self.client_subscriptions:
+            self.client_subscriptions[client_id].discard(device_id)
+
+    # Convenience methods for specific message types
+
+    async def broadcast_device_update(self, device_id: str, device_data: Dict[str, Any]):
+        """Broadcast device update.
+
+        Args:
+            device_id: Device ID
+            device_data: Device data
+        """
+        await self.broadcast_to_device_subscribers(
+            device_id,
+            WSMessage(
+                type=WSMessageType.DEVICE_UPDATE,
+                data={"device_id": device_id, **device_data},
+            ),
+        )
+
+    async def broadcast_task_started(self, task_id: str, task_data: Dict[str, Any]):
+        """Broadcast task started.
+
+        Args:
+            task_id: Task ID
+            task_data: Task data
+        """
+        await self.broadcast(
+            WSMessage(
+                type=WSMessageType.TASK_STARTED,
+                data={"task_id": task_id, **task_data},
+            )
+        )
+
+    async def broadcast_step_update(
+        self,
+        task_id: str,
+        device_id: str,
+        step: int,
+        action: Optional[Dict] = None,
+        thinking: Optional[str] = None,
+        finished: bool = False,
+        success: bool = True,
+        message: Optional[str] = None,
+    ):
+        """Broadcast task step update.
+
+        Args:
+            task_id: Task ID
+            device_id: Device ID
+            step: Step number
+            action: Current action
+            thinking: AI reasoning
+            finished: Whether task is finished
+            success: Whether step succeeded
+            message: Status message
+        """
+        await self.broadcast_to_device_subscribers(
+            device_id,
+            WSMessage(
+                type=WSMessageType.TASK_STEP,
+                data={
+                    "task_id": task_id,
+                    "device_id": device_id,
+                    "step": step,
+                    "action": action,
+                    "thinking": thinking,
+                    "finished": finished,
+                    "success": success,
+                    "message": message,
+                },
+            ),
+        )
+
+    async def broadcast_task_completed(
+        self, task_id: str, device_id: str, status: str, message: Optional[str] = None
+    ):
+        """Broadcast task completed.
+
+        Args:
+            task_id: Task ID
+            device_id: Device ID
+            status: Task status
+            message: Completion message
+        """
+        await self.broadcast(
+            WSMessage(
+                type=WSMessageType.TASK_COMPLETED,
+                data={
+                    "task_id": task_id,
+                    "device_id": device_id,
+                    "status": status,
+                    "message": message,
+                },
+            )
+        )
+
+    async def broadcast_task_failed(self, task_id: str, device_id: str, error: str):
+        """Broadcast task failed.
+
+        Args:
+            task_id: Task ID
+            device_id: Device ID
+            error: Error message
+        """
+        await self.broadcast(
+            WSMessage(
+                type=WSMessageType.TASK_FAILED,
+                data={
+                    "task_id": task_id,
+                    "device_id": device_id,
+                    "error": error,
+                },
+            )
+        )
+
+    async def broadcast_task_stopped(self, task_id: str, device_id: str):
+        """Broadcast task stopped.
+
+        Args:
+            task_id: Task ID
+            device_id: Device ID
+        """
+        await self.broadcast(
+            WSMessage(
+                type=WSMessageType.TASK_STOPPED,
+                data={
+                    "task_id": task_id,
+                    "device_id": device_id,
+                },
+            )
+        )
+
+    async def broadcast_screenshot(
+        self, device_id: str, screenshot: str, width: int, height: int
+    ):
+        """Broadcast screenshot update.
+
+        Args:
+            device_id: Device ID
+            screenshot: Base64 encoded screenshot
+            width: Image width
+            height: Image height
+        """
+        await self.broadcast_to_device_subscribers(
+            device_id,
+            WSMessage(
+                type=WSMessageType.SCREENSHOT,
+                data={
+                    "device_id": device_id,
+                    "screenshot": screenshot,
+                    "width": width,
+                    "height": height,
+                },
+            ),
+        )
+
+    async def broadcast_error(self, error: str, details: Optional[Dict] = None):
+        """Broadcast error.
+
+        Args:
+            error: Error message
+            details: Additional error details
+        """
+        await self.broadcast(
+            WSMessage(
+                type=WSMessageType.ERROR,
+                data={
+                    "error": error,
+                    "details": details,
+                },
+            )
+        )
+
+    def get_connection_count(self) -> int:
+        """Get number of active connections.
+
+        Returns:
+            Connection count
+        """
+        return len(self.active_connections)
+
+    def get_client_ids(self) -> list[str]:
+        """Get list of connected client IDs.
+
+        Returns:
+            List of client IDs
+        """
+        return list(self.active_connections.keys())

dashboard/static/css/dashboard.css

@@ -0,0 +1,539 @@
+/* AutoGLM Dashboard Styles */
+
+:root {
+    --primary-color: #6366f1;
+    --primary-hover: #4f46e5;
+    --success-color: #10b981;
+    --warning-color: #f59e0b;
+    --danger-color: #ef4444;
+    --bg-color: #0f172a;
+    --card-bg: #1e293b;
+    --border-color: #334155;
+    --text-primary: #f1f5f9;
+    --text-secondary: #94a3b8;
+    --shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.3);
+    --shadow-lg: 0 10px 15px -3px rgba(0, 0, 0, 0.4);
+}
+
+* {
+    margin: 0;
+    padding: 0;
+    box-sizing: border-box;
+}
+
+body {
+    font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, sans-serif;
+    background-color: var(--bg-color);
+    color: var(--text-primary);
+    line-height: 1.6;
+    min-height: 100vh;
+}
+
+/* Header */
+.header {
+    background-color: var(--card-bg);
+    border-bottom: 1px solid var(--border-color);
+    padding: 1rem 2rem;
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    position: sticky;
+    top: 0;
+    z-index: 100;
+    box-shadow: var(--shadow);
+}
+
+.header-content {
+    display: flex;
+    align-items: center;
+    gap: 2rem;
+}
+
+.header h1 {
+    font-size: 1.5rem;
+    font-weight: 600;
+    color: var(--text-primary);
+}
+
+.stats {
+    display: flex;
+    gap: 1.5rem;
+}
+
+.stat {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    font-size: 0.875rem;
+    color: var(--text-secondary);
+}
+
+.stat svg {
+    opacity: 0.7;
+}
+
+.ws-status {
+    color: var(--danger-color);
+}
+
+.ws-status.connected {
+    color: var(--success-color);
+}
+
+.header-actions {
+    display: flex;
+    gap: 0.75rem;
+}
+
+/* Main Content */
+.main-content {
+    padding: 2rem;
+    max-width: 1600px;
+    margin: 0 auto;
+}
+
+.main-content h2 {
+    font-size: 1.25rem;
+    font-weight: 600;
+    margin-bottom: 1rem;
+    color: var(--text-primary);
+}
+
+/* Device Grid */
+.device-grid {
+    display: grid;
+    grid-template-columns: repeat(auto-fill, minmax(350px, 1fr));
+    gap: 1.5rem;
+    margin-bottom: 3rem;
+}
+
+.device-card {
+    background-color: var(--card-bg);
+    border: 1px solid var(--border-color);
+    border-radius: 12px;
+    padding: 1.25rem;
+    transition: box-shadow 0.2s, border-color 0.2s;
+}
+
+.device-card:hover {
+    box-shadow: var(--shadow-lg);
+}
+
+.device-card.busy {
+    border-color: var(--warning-color);
+}
+
+.device-card.offline {
+    opacity: 0.6;
+}
+
+.device-header {
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    margin-bottom: 1rem;
+}
+
+.device-header h3 {
+    font-size: 1rem;
+    font-weight: 600;
+    color: var(--text-primary);
+    word-break: break-all;
+}
+
+.status-badge {
+    padding: 0.25rem 0.75rem;
+    border-radius: 9999px;
+    font-size: 0.75rem;
+    font-weight: 600;
+    text-transform: uppercase;
+}
+
+.status-badge.online {
+    background-color: rgba(16, 185, 129, 0.2);
+    color: var(--success-color);
+}
+
+.status-badge.offline {
+    background-color: rgba(239, 68, 68, 0.2);
+    color: var(--danger-color);
+}
+
+.status-badge.busy {
+    background-color: rgba(245, 158, 11, 0.2);
+    color: var(--warning-color);
+}
+
+.device-info {
+    margin-bottom: 1rem;
+}
+
+.device-info p {
+    font-size: 0.875rem;
+    color: var(--text-secondary);
+    margin-bottom: 0.25rem;
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+}
+
+.device-info p svg {
+    flex-shrink: 0;
+}
+
+.screenshot {
+    margin-bottom: 1rem;
+    border-radius: 8px;
+    overflow: hidden;
+    background-color: #000;
+    aspect-ratio: 9/16;
+}
+
+.screenshot img {
+    width: 100%;
+    height: 100%;
+    object-fit: contain;
+}
+
+.device-actions {
+    display: flex;
+    gap: 0.5rem;
+}
+
+.device-actions input {
+    flex: 1;
+    padding: 0.625rem 0.875rem;
+    background-color: var(--bg-color);
+    border: 1px solid var(--border-color);
+    border-radius: 8px;
+    color: var(--text-primary);
+    font-size: 0.875rem;
+}
+
+.device-actions input:focus {
+    outline: none;
+    border-color: var(--primary-color);
+}
+
+.device-actions input:disabled {
+    opacity: 0.5;
+    cursor: not-allowed;
+}
+
+/* Task List */
+.task-list {
+    display: flex;
+    flex-direction: column;
+    gap: 0.75rem;
+}
+
+.task-item {
+    background-color: var(--card-bg);
+    border: 1px solid var(--border-color);
+    border-radius: 8px;
+    padding: 1rem;
+    display: flex;
+    justify-content: space-between;
+    align-items: start;
+    gap: 1rem;
+    transition: border-color 0.2s;
+}
+
+.task-item.active {
+    border-color: var(--primary-color);
+    box-shadow: 0 0 0 1px var(--primary-color);
+}
+
+.task-info {
+    flex: 1;
+}
+
+.task-header {
+    display: flex;
+    justify-content: space-between;
+    align-items: center;
+    margin-bottom: 0.5rem;
+}
+
+.task-id {
+    font-size: 0.75rem;
+    color: var(--text-secondary);
+    font-family: monospace;
+}
+
+.task-status {
+    padding: 0.125rem 0.5rem;
+    border-radius: 4px;
+    font-size: 0.7rem;
+    font-weight: 600;
+    text-transform: uppercase;
+}
+
+.task-status.running {
+    background-color: rgba(99, 102, 241, 0.2);
+    color: var(--primary-color);
+}
+
+.task-status.completed {
+    background-color: rgba(16, 185, 129, 0.2);
+    color: var(--success-color);
+}
+
+.task-status.failed {
+    background-color: rgba(239, 68, 68, 0.2);
+    color: var(--danger-color);
+}
+
+.task-status.stopped {
+    background-color: rgba(148, 163, 184, 0.2);
+    color: var(--text-secondary);
+}
+
+.task-description {
+    font-size: 0.95rem;
+    color: var(--text-primary);
+    margin-bottom: 0.5rem;
+}
+
+.task-meta {
+    display: flex;
+    gap: 1rem;
+    font-size: 0.8rem;
+    color: var(--text-secondary);
+}
+
+.task-progress {
+    height: 4px;
+    background-color: var(--bg-color);
+    border-radius: 2px;
+    overflow: hidden;
+    margin-top: 0.5rem;
+}
+
+.progress-bar {
+    height: 100%;
+    background-color: var(--primary-color);
+    transition: width 0.3s ease;
+}
+
+.task-thinking {
+    display: flex;
+    align-items: start;
+    gap: 0.5rem;
+    margin-top: 0.5rem;
+    padding: 0.5rem;
+    background-color: var(--bg-color);
+    border-radius: 6px;
+    font-size: 0.85rem;
+    color: var(--text-secondary);
+}
+
+.task-thinking svg {
+    flex-shrink: 0;
+    margin-top: 2px;
+}
+
+.task-action {
+    display: flex;
+    align-items: center;
+    gap: 0.5rem;
+    margin-top: 0.5rem;
+    padding: 0.5rem;
+    background-color: rgba(99, 102, 241, 0.1);
+    border-radius: 6px;
+    font-size: 0.85rem;
+    color: var(--primary-color);
+}
+
+.task-action svg {
+    flex-shrink: 0;
+}
+
+.task-action strong {
+    color: var(--text-primary);
+}
+
+.task-message {
+    display: flex;
+    align-items: start;
+    gap: 0.5rem;
+    margin-top: 0.5rem;
+    padding: 0.5rem;
+    border-radius: 6px;
+    font-size: 0.85rem;
+    line-height: 1.4;
+}
+
+.task-message svg {
+    flex-shrink: 0;
+    margin-top: 2px;
+}
+
+.task-message.completed {
+    background-color: rgba(16, 185, 129, 0.1);
+    color: var(--success-color);
+}
+
+.task-message.failed {
+    background-color: rgba(239, 68, 68, 0.1);
+    color: var(--danger-color);
+}
+
+.task-message.stopped {
+    background-color: rgba(148, 163, 184, 0.1);
+    color: var(--text-secondary);
+}
+
+.task-actions {
+    flex-shrink: 0;
+}
+
+/* Buttons */
+.btn {
+    padding: 0.625rem 1rem;
+    border: none;
+    border-radius: 8px;
+    font-size: 0.875rem;
+    font-weight: 500;
+    cursor: pointer;
+    transition: all 0.2s;
+    display: inline-flex;
+    align-items: center;
+    gap: 0.5rem;
+}
+
+.btn:disabled {
+    opacity: 0.5;
+    cursor: not-allowed;
+}
+
+.btn-primary {
+    background-color: var(--primary-color);
+    color: white;
+}
+
+.btn-primary:hover:not(:disabled) {
+    background-color: var(--primary-hover);
+}
+
+.btn-secondary {
+    background-color: var(--card-bg);
+    color: var(--text-primary);
+    border: 1px solid var(--border-color);
+}
+
+.btn-secondary:hover:not(:disabled) {
+    background-color: var(--border-color);
+}
+
+.btn-danger {
+    background-color: var(--danger-color);
+    color: white;
+}
+
+.btn-danger:hover:not(:disabled) {
+    background-color: #dc2626;
+}
+
+.btn-sm {
+    padding: 0.5rem 0.75rem;
+    font-size: 0.8rem;
+}
+
+.spinning {
+    animation: spin 1s linear infinite;
+}
+
+@keyframes spin {
+    from { transform: rotate(0deg); }
+    to { transform: rotate(360deg); }
+}
+
+/* Empty State */
+.empty-state {
+    text-align: center;
+    padding: 4rem 2rem;
+    color: var(--text-secondary);
+}
+
+.empty-state svg {
+    margin-bottom: 1rem;
+    opacity: 0.3;
+}
+
+.empty-state p {
+    margin-bottom: 1.5rem;
+}
+
+.empty-state .hint {
+    font-size: 0.875rem;
+    opacity: 0.7;
+}
+
+/* Toast */
+.toast-container {
+    position: fixed;
+    bottom: 2rem;
+    right: 2rem;
+    z-index: 1000;
+    display: flex;
+    flex-direction: column;
+    gap: 0.5rem;
+}
+
+.toast {
+    padding: 0.75rem 1rem;
+    border-radius: 8px;
+    font-size: 0.875rem;
+    box-shadow: var(--shadow-lg);
+    animation: slideIn 0.3s ease;
+}
+
+.toast.success {
+    background-color: var(--success-color);
+    color: white;
+}
+
+.toast.error {
+    background-color: var(--danger-color);
+    color: white;
+}
+
+.toast.info {
+    background-color: var(--primary-color);
+    color: white;
+}
+
+@keyframes slideIn {
+    from {
+        transform: translateX(100%);
+        opacity: 0;
+    }
+    to {
+        transform: translateX(0);
+        opacity: 1;
+    }
+}
+
+/* Responsive */
+@media (max-width: 768px) {
+    .header {
+        flex-direction: column;
+        gap: 1rem;
+        padding: 1rem;
+    }
+
+    .header-content {
+        flex-direction: column;
+        gap: 0.75rem;
+        align-items: start;
+    }
+
+    .device-grid {
+        grid-template-columns: 1fr;
+    }
+
+    .main-content {
+        padding: 1rem;
+    }
+}

dashboard/static/index.html

@@ -0,0 +1,219 @@
+<!DOCTYPE html>
+<html lang="zh-CN">
+<head>
+    <meta charset="UTF-8">
+    <meta name="viewport" content="width=device-width, initial-scale=1.0">
+    <title>AutoGLM Dashboard</title>
+    <!-- Vue.js 3 -->
+    <script src="https://unpkg.com/vue@3/dist/vue.global.js"></script>
+    <!-- Axios for API requests -->
+    <script src="https://unpkg.com/axios/dist/axios.min.js"></script>
+    <!-- CSS -->
+    <link rel="stylesheet" href="/static/css/dashboard.css">
+</head>
+<body>
+    <div id="app">
+        <!-- Header -->
+        <header class="header">
+            <div class="header-content">
+                <h1>AutoGLM Dashboard</h1>
+                <div class="stats">
+                    <span class="stat" title="Connected Devices">
+                        <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                            <rect x="5" y="2" width="14" height="20" rx="2" ry="2"></rect>
+                            <line x1="12" y1="18" x2="12" y2="18"></line>
+                        </svg>
+                        {{ connectedDeviceCount }} / {{ devices.length }}
+                    </span>
+                    <span class="stat" title="Active Tasks">
+                        <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                            <polyline points="22 12 18 12 15 21 9 3 6 12 2 12"></polyline>
+                        </svg>
+                        {{ activeTasks.length }}
+                    </span>
+                    <span class="stat ws-status" :class="{ connected: wsConnected }" title="WebSocket">
+                        <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                            <circle cx="12" cy="12" r="10"></circle>
+                            <line x1="2" y1="12" x2="22" y2="12"></line>
+                            <path d="M12 2a15.3 15.3 0 0 1 4 10 15.3 15.3 0 0 1-4 10 15.3 15.3 0 0 1-4-10 15.3 15.3 0 0 1 4-10z"></path>
+                        </svg>
+                    </span>
+                </div>
+            </div>
+            <div class="header-actions">
+                <button @click="refreshDevices" class="btn btn-secondary" :disabled="refreshing">
+                    <svg width="16" height="16" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" :class="{ spinning: refreshing }">
+                        <polyline points="23 4 23 10 17 10"></polyline>
+                        <polyline points="1 20 1 14 7 14"></polyline>
+                        <path d="M3.51 9a9 9 0 0 1 14.85-3.36L23 10M1 14l4.64 4.36A9 9 0 0 0 20.49 15"></path>
+                    </svg>
+                    Refresh
+                </button>
+            </div>
+        </header>
+
+        <!-- Main Content -->
+        <main class="main-content">
+            <!-- Device Grid -->
+            <section class="devices-section">
+                <h2>Devices</h2>
+                <div class="device-grid" v-if="devices.length > 0">
+                    <div
+                        class="device-card"
+                        v-for="device in devices"
+                        :key="device.device_id"
+                        :class="{ busy: device.status === 'busy', offline: !device.is_connected }"
+                    >
+                        <div class="device-header">
+                            <h3>{{ device.device_id }}</h3>
+                            <span class="status-badge" :class="device.status">
+                                {{ device.status }}
+                            </span>
+                        </div>
+                        <div class="device-info">
+                            <p v-if="device.model">{{ device.model }}</p>
+                            <p v-if="device.android_version" class="version">
+                                <svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                    <path d="M12 2L2 7l10 5 10-5-10-5z"></path>
+                                    <path d="M2 17l10 5 10-5"></path>
+                                    <path d="M2 12l10 5 10-5"></path>
+                                </svg>
+                                {{ device.android_version }}
+                            </p>
+                            <p v-if="device.current_app" class="app">
+                                <svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                    <rect x="3" y="3" width="18" height="18" rx="2"></rect>
+                                </svg>
+                                {{ formatAppName(device.current_app) }}
+                            </p>
+                        </div>
+                        <div class="screenshot" v-if="device.screenshot">
+                            <img :src="'data:image/png;base64,' + device.screenshot" alt="Screen">
+                        </div>
+                        <div class="device-actions">
+                            <input
+                                v-model="device.taskInput"
+                                type="text"
+                                placeholder="Enter task..."
+                                @keyup.enter="executeTask(device)"
+                                :disabled="device.status === 'busy' || !device.is_connected"
+                            >
+                            <button
+                                @click="executeTask(device)"
+                                class="btn btn-primary"
+                                :disabled="device.status === 'busy' || !device.is_connected || !device.taskInput"
+                            >
+                                {{ device.status === 'busy' ? 'Running...' : 'Execute' }}
+                            </button>
+                        </div>
+                    </div>
+                </div>
+                <div class="empty-state" v-else>
+                    <svg width="64" height="64" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1">
+                        <rect x="5" y="2" width="14" height="20" rx="2" ry="2"></rect>
+                        <line x1="12" y1="18" x2="12" y2="18"></line>
+                    </svg>
+                    <p>No devices found</p>
+                    <button @click="refreshDevices" class="btn btn-secondary">Scan for Devices</button>
+                </div>
+            </section>
+
+            <!-- Task Queue -->
+            <section class="tasks-section">
+                <h2>Task Queue</h2>
+                <div class="task-list" v-if="tasks.length > 0">
+                    <div
+                        class="task-item"
+                        v-for="task in tasks"
+                        :key="task.task_id"
+                        :class="{ active: task.status === 'running' }"
+                    >
+                        <div class="task-info">
+                            <div class="task-header">
+                                <span class="task-id">{{ task.task_id }}</span>
+                                <span class="task-status" :class="task.status">{{ task.status }}</span>
+                            </div>
+                            <p class="task-description">{{ task.task }}</p>
+                            <div class="task-meta">
+                                <span>Device: {{ task.device_id }}</span>
+                                <span>Step: {{ task.current_step }}/{{ task.max_steps }}</span>
+                            </div>
+                            <div class="task-progress" v-if="task.status === 'running'">
+                                <div class="progress-bar" :style="{ width: (task.current_step / task.max_steps * 100) + '%' }"></div>
+                            </div>
+                            <!-- Current Action -->
+                            <div class="task-action" v-if="task.current_action && task.status === 'running'">
+                                <svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                    <polyline points="22 12 18 12 15 21 9 3 6 12 2 12"></polyline>
+                                </svg>
+                                <strong>{{ task.current_action.action }}</strong>
+                                <span v-if="task.current_action.element"> on {{ task.current_action.element }}</span>
+                                <span v-if="task.current_action.text">: "{{ task.current_action.text }}"</span>
+                            </div>
+                            <!-- Thinking -->
+                            <div class="task-thinking" v-if="task.thinking">
+                                <svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                    <circle cx="12" cy="12" r="10"></circle>
+                                    <line x1="12" y1="16" x2="12" y2="12"></line>
+                                    <line x1="12" y1="8" x2="12.01" y2="8"></line>
+                                </svg>
+                                {{ task.thinking }}
+                            </div>
+                            <!-- Completion Message -->
+                            <div class="task-message" v-if="task.completion_message && (task.status === 'completed' || task.status === 'failed' || task.status === 'stopped')">
+                                <svg width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                    <path v-if="task.status === 'completed'" d="M22 11.08V12a10 10 0 1 1-5.93-9.14"></path>
+                                    <polyline v-if="task.status === 'completed'" points="22 4 12 14.01 9 11.01"></polyline>
+                                    <circle v-if="task.status === 'failed'" cx="12" cy="12" r="10"></circle>
+                                    <line v-if="task.status === 'failed'" x1="15" y1="9" x2="9" y2="15"></line>
+                                    <line v-if="task.status === 'failed'" x1="9" y1="9" x2="15" y2="15"></line>
+                                </svg>
+                                {{ task.completion_message }}
+                            </div>
+                        </div>
+                        <div class="task-actions" v-if="task.status === 'running'">
+                            <button @click="stopTask(task)" class="btn btn-danger btn-sm">
+                                <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                    <rect x="6" y="6" width="12" height="12"></rect>
+                                </svg>
+                                Stop
+                            </button>
+                        </div>
+                        <div class="task-actions" v-else>
+                            <button @click="reExecuteTask(task)" class="btn btn-secondary btn-sm">
+                                <svg width="14" height="14" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2">
+                                    <polyline points="23 4 23 10 17 10"></polyline>
+                                    <polyline points="1 20 1 14 7 14"></polyline>
+                                    <path d="M3.51 9a9 9 0 0 1 14.85-3.36L23 10M1 14l4.64 4.36A9 9 0 0 0 20.49 15"></path>
+                                </svg>
+                                Re-execute
+                            </button>
+                        </div>
+                    </div>
+                </div>
+                <div class="empty-state" v-else>
+                    <svg width="64" height="64" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="1">
+                        <polyline points="22 12 18 12 15 21 9 3 6 12 2 12"></polyline>
+                    </svg>
+                    <p>No tasks yet</p>
+                    <p class="hint">Enter a task above to get started</p>
+                </div>
+            </section>
+        </main>
+
+        <!-- Toast notifications -->
+        <div class="toast-container">
+            <div
+                class="toast"
+                v-for="toast in toasts"
+                :key="toast.id"
+                :class="toast.type"
+            >
+                {{ toast.message }}
+            </div>
+        </div>
+    </div>
+
+    <script src="/static/js/dashboard.js"></script>
+</body>
+</html>

dashboard/static/js/dashboard.js

@@ -0,0 +1,404 @@
+/**
+ * AutoGLM Dashboard - Vue.js Application
+ */
+
+const { createApp } = Vue;
+
+createApp({
+    data() {
+        return {
+            devices: [],
+            tasks: [],
+            ws: null,
+            wsConnected: false,
+            refreshing: false,
+            toasts: [],
+            toastIdCounter: 0,
+            reconnectAttempts: 0,
+            maxReconnectAttempts: 5,
+            reconnectDelay: 2000,
+        };
+    },
+
+    computed: {
+        activeTasks() {
+            return this.tasks.filter(t => t.status === 'running');
+        },
+
+        connectedDeviceCount() {
+            return this.devices.filter(d => d.is_connected).length;
+        },
+    },
+
+    methods: {
+        /* API Methods */
+        async loadDevices() {
+            try {
+                const response = await axios.get('/api/devices');
+                this.devices = response.data.map(d => ({
+                    ...d,
+                    taskInput: '',
+                    screenshot: null,
+                }));
+            } catch (error) {
+                this.showToast('Failed to load devices', 'error');
+                console.error('Error loading devices:', error);
+            }
+        },
+
+        async loadTasks() {
+            try {
+                const response = await axios.get('/api/tasks');
+                this.tasks = response.data;
+            } catch (error) {
+                console.error('Error loading tasks:', error);
+            }
+        },
+
+        async refreshDevices() {
+            this.refreshing = true;
+            try {
+                await axios.get('/api/devices/refresh');
+                await this.loadDevices();
+                this.showToast('Devices refreshed', 'success');
+            } catch (error) {
+                this.showToast('Failed to refresh devices', 'error');
+            } finally {
+                this.refreshing = false;
+            }
+        },
+
+        async executeTask(device) {
+            if (!device.taskInput || !device.taskInput.trim()) {
+                return;
+            }
+
+            const task = device.taskInput.trim();
+            device.taskInput = '';
+
+            try {
+                const response = await axios.post('/api/tasks/execute', {
+                    device_id: device.device_id,
+                    task: task,
+                    max_steps: 100,
+                    lang: 'cn',
+                    // Backend will use config from environment
+                });
+
+                this.showToast(`Task started: ${task}`, 'success');
+                await this.loadTasks();
+            } catch (error) {
+                this.showToast('Failed to start task', 'error');
+                console.error('Error executing task:', error);
+            }
+        },
+
+        async stopTask(task) {
+            try {
+                await axios.post(`/api/tasks/${task.task_id}/stop`);
+                this.showToast('Stopping task...', 'info');
+            } catch (error) {
+                this.showToast('Failed to stop task', 'error');
+            }
+        },
+
+        async reExecuteTask(task) {
+            const device = this.devices.find(d => d.device_id === task.device_id);
+            if (!device) {
+                this.showToast('Device not found', 'error');
+                return;
+            }
+
+            if (!device.is_connected) {
+                this.showToast('Device is not connected', 'error');
+                return;
+            }
+
+            if (device.status === 'busy') {
+                this.showToast('Device is busy', 'error');
+                return;
+            }
+
+            try {
+                const response = await axios.post('/api/tasks/execute', {
+                    device_id: task.device_id,
+                    task: task.task,
+                    max_steps: task.max_steps || 100,
+                    lang: 'cn',
+                });
+
+                this.showToast(`Task re-executed: ${task.task}`, 'success');
+                await this.loadTasks();
+            } catch (error) {
+                this.showToast('Failed to re-execute task', 'error');
+                console.error('Error re-executing task:', error);
+            }
+        },
+
+        async captureScreenshot(deviceId) {
+            try {
+                const response = await axios.get(`/api/devices/${deviceId}/screenshot`);
+                const device = this.devices.find(d => d.device_id === deviceId);
+                if (device) {
+                    device.screenshot = response.data.screenshot;
+                }
+            } catch (error) {
+                console.error('Error capturing screenshot:', error);
+            }
+        },
+
+        /* WebSocket Methods */
+        connectWebSocket() {
+            const protocol = location.protocol === 'https:' ? 'wss:' : 'ws:';
+            const wsUrl = `${protocol}//${location.host}/ws`;
+
+            this.ws = new WebSocket(wsUrl);
+
+            this.ws.onopen = () => {
+                this.wsConnected = true;
+                this.reconnectAttempts = 0;
+                console.log('WebSocket connected');
+
+                // Subscribe to all devices updates
+                this.sendWebSocketMessage({
+                    type: 'subscribe',
+                    device_id: '*',  // Subscribe to all devices
+                });
+            };
+
+            this.ws.onmessage = (event) => {
+                try {
+                    const msg = JSON.parse(event.data);
+                    this.handleWSMessage(msg);
+                } catch (error) {
+                    console.error('Error parsing WebSocket message:', error);
+                }
+            };
+
+            this.ws.onclose = () => {
+                this.wsConnected = false;
+                console.log('WebSocket disconnected');
+
+                // Attempt to reconnect
+                if (this.reconnectAttempts < this.maxReconnectAttempts) {
+                    this.reconnectAttempts++;
+                    console.log(`Reconnecting... (${this.reconnectAttempts}/${this.maxReconnectAttempts})`);
+                    setTimeout(() => this.connectWebSocket(), this.reconnectDelay);
+                } else {
+                    this.showToast('WebSocket connection lost', 'error');
+                }
+            };
+
+            this.ws.onerror = (error) => {
+                console.error('WebSocket error:', error);
+            };
+        },
+
+        sendWebSocketMessage(data) {
+            if (this.ws && this.ws.readyState === WebSocket.OPEN) {
+                this.ws.send(JSON.stringify(data));
+            }
+        },
+
+        handleWSMessage(msg) {
+            switch (msg.type) {
+                case 'device_update':
+                    this.updateDevice(msg.data);
+                    break;
+
+                case 'task_started':
+                    this.handleTaskStarted(msg.data);
+                    break;
+
+                case 'task_step':
+                    this.handleTaskStep(msg.data);
+                    break;
+
+                case 'task_completed':
+                    this.handleTaskCompleted(msg.data);
+                    break;
+
+                case 'task_failed':
+                    this.handleTaskFailed(msg.data);
+                    break;
+
+                case 'task_stopped':
+                    this.handleTaskStopped(msg.data);
+                    break;
+
+                case 'screenshot':
+                    this.handleScreenshot(msg.data);
+                    break;
+
+                case 'error':
+                    this.showToast(msg.data.error || 'An error occurred', 'error');
+                    break;
+
+                case 'pong':
+                    // Ping response, ignore
+                    break;
+
+                default:
+                    console.log('Unknown WebSocket message type:', msg.type);
+            }
+        },
+
+        /* Message Handlers */
+        updateDevice(data) {
+            const device = this.devices.find(d => d.device_id === data.device_id);
+            if (device) {
+                Object.assign(device, {
+                    status: data.status,
+                    is_connected: data.is_connected,
+                    model: data.model,
+                    current_app: data.current_app,
+                });
+            }
+        },
+
+        handleTaskStarted(data) {
+            this.tasks.unshift({
+                task_id: data.task_id,
+                device_id: data.device_id,
+                task: data.task,
+                status: 'running',
+                current_step: 0,
+                max_steps: data.max_steps || 100,
+                current_action: null,
+                thinking: null,
+                started_at: new Date(),
+            });
+
+            // Update device status
+            const device = this.devices.find(d => d.device_id === data.device_id);
+            if (device) {
+                device.status = 'busy';
+            }
+        },
+
+        handleTaskStep(data) {
+            const task = this.tasks.find(t => t.task_id === data.task_id);
+            if (task) {
+                task.current_step = data.step;
+                task.current_action = data.action;
+                task.thinking = data.thinking;
+                // Store the completion message
+                if (data.message && data.finished) {
+                    task.completion_message = data.message;
+                }
+
+                if (data.finished) {
+                    task.status = data.success ? 'completed' : 'failed';
+                    this.releaseDevice(data.device_id);
+                }
+            }
+        },
+
+        handleTaskCompleted(data) {
+            const task = this.tasks.find(t => t.task_id === data.task_id);
+            if (task) {
+                task.status = 'completed';
+                task.finished_at = new Date();
+                // Store completion message
+                if (data.message) {
+                    task.completion_message = data.message;
+                }
+                this.releaseDevice(data.device_id);
+                this.showToast(`Task completed: ${task.task}`, 'success');
+            }
+        },
+
+        handleTaskFailed(data) {
+            const task = this.tasks.find(t => t.task_id === data.task_id);
+            if (task) {
+                task.status = 'failed';
+                task.error = data.error;
+                task.finished_at = new Date();
+                this.releaseDevice(data.device_id);
+                this.showToast(`Task failed: ${data.error}`, 'error');
+            }
+        },
+
+        handleTaskStopped(data) {
+            const task = this.tasks.find(t => t.task_id === data.task_id);
+            if (task) {
+                task.status = 'stopped';
+                task.finished_at = new Date();
+                this.releaseDevice(data.device_id);
+                this.showToast('Task stopped', 'info');
+            }
+        },
+
+        handleScreenshot(data) {
+            const device = this.devices.find(d => d.device_id === data.device_id);
+            if (device) {
+                device.screenshot = data.screenshot;
+            }
+        },
+
+        releaseDevice(deviceId) {
+            const device = this.devices.find(d => d.device_id === deviceId);
+            if (device && device.status === 'busy') {
+                device.status = 'online';
+            }
+        },
+
+        /* Utility Methods */
+        formatAppName(packageName) {
+            if (!packageName) return '';
+            // Format package name for display
+            const parts = packageName.split('.');
+            return parts[parts.length - 1] || packageName;
+        },
+
+        showToast(message, type = 'info') {
+            const id = this.toastIdCounter++;
+            this.toasts.push({ id, message, type });
+
+            // Auto-remove after 3 seconds
+            setTimeout(() => {
+                this.toasts = this.toasts.filter(t => t.id !== id);
+            }, 3000);
+        },
+
+        /* Heartbeat */
+        startHeartbeat() {
+            setInterval(() => {
+                if (this.ws && this.ws.readyState === WebSocket.OPEN) {
+                    this.sendWebSocketMessage({ type: 'ping' });
+                }
+            }, 30000); // Ping every 30 seconds
+        },
+    },
+
+    mounted() {
+        // Load initial data
+        this.loadDevices();
+        this.loadTasks();
+
+        // Connect WebSocket
+        this.connectWebSocket();
+
+        // Start heartbeat
+        this.startHeartbeat();
+
+        // Refresh devices periodically
+        setInterval(() => {
+            if (!this.refreshing) {
+                this.loadDevices();
+            }
+        }, 10000); // Every 10 seconds
+
+        // Refresh tasks periodically
+        setInterval(() => {
+            this.loadTasks();
+        }, 5000); // Every 5 seconds
+    },
+
+    beforeUnmount() {
+        // Close WebSocket on unmount
+        if (this.ws) {
+            this.ws.close();
+        }
+    },
+}).mount('#app');

examples/callback_hooks.py

@@ -0,0 +1,415 @@
+#!/usr/bin/env python3
+"""
+Phone Agent Callback Hooks Examples / Phone Agent 回调钩子示例
+
+Demonstrates how to use callback hooks to control task execution.
+演示如何使用回调钩子来控制任务执行。
+
+Features / 功能:
+- Interrupt tasks during execution / 在执行过程中中断任务
+- Switch to new tasks dynamically / 动态切换到新任务
+- Modify actions before execution / 在执行前修改操作
+- Collect execution statistics / 收集执行统计信息
+
+Configuration / 配置:
+    This script loads settings from .env file (if present).
+    本脚本会从 .env 文件加载配置（如果存在）。
+"""
+
+import threading
+import time
+from collections import defaultdict
+from dotenv import load_dotenv
+
+# Load .env file for configuration
+# 加载 .env 配置文件
+load_dotenv()
+
+from phone_agent import PhoneAgent
+from phone_agent.agent import AgentConfig, StepResult
+from phone_agent.config import get_messages
+from phone_agent.model import ModelConfig
+
+
+def _create_agent_with_callbacks(
+    lang: str = "cn",
+    step_callback=None,
+    before_action_callback=None,
+) -> PhoneAgent:
+    """
+    Helper function to create PhoneAgent with config from environment variables.
+    从环境变量创建配置的 PhoneAgent 辅助函数。
+    """
+    import os
+
+    model_config = ModelConfig(
+        base_url=os.getenv("PHONE_AGENT_BASE_URL", "http://localhost:8000/v1"),
+        model_name=os.getenv("PHONE_AGENT_MODEL", "autoglm-phone-9b"),
+        api_key=os.getenv("PHONE_AGENT_API_KEY", "EMPTY"),
+    )
+
+    agent_config = AgentConfig(
+        max_steps=int(os.getenv("PHONE_AGENT_MAX_STEPS", "100")),
+        device_id=os.getenv("PHONE_AGENT_DEVICE_ID"),
+        lang=lang,
+        step_callback=step_callback,
+        before_action_callback=before_action_callback,
+    )
+
+    return PhoneAgent(model_config=model_config, agent_config=agent_config)
+
+
+# ============================================================================
+# Example 1: Interrupt after max steps / 示例1：达到最大步数后中断
+# ============================================================================
+
+def example_interrupt_after_steps(lang: str = "cn"):
+    """Interrupt task after reaching maximum steps / 达到最大步数后中断任务"""
+    msgs = get_messages(lang)
+
+    max_allowed_steps = 5
+    steps_taken = []
+
+    def step_callback(result: StepResult) -> str | None:
+        """Called after each step / 每步执行后调用"""
+        steps_taken.append(result.step_count)
+        print(f"[Callback] Step {result.step_count}: {result.action}")
+
+        # Interrupt after reaching max steps
+        if result.step_count >= max_allowed_steps:
+            return "stop"
+        return None
+
+    agent = _create_agent_with_callbacks(lang=lang, step_callback=step_callback)
+
+    print(f"\n{'='*50}")
+    print(f"Task will be interrupted after {max_allowed_steps} steps")
+    print(f"{'='*50}\n")
+
+    result = agent.run("打开小红书浏览美食内容")
+
+    print(f"\n{'='*50}")
+    print(f"Task interrupted after {len(steps_taken)} steps")
+    print(f"{'='*50}")
+
+
+# ============================================================================
+# Example 2: Switch tasks dynamically / 示例2：动态切换任务
+# ============================================================================
+
+def example_switch_tasks(lang: str = "cn"):
+    """Switch to new task based on conditions / 根据条件切换到新任务"""
+    msgs = get_messages(lang)
+
+    task_queue = ["打开微信", "打开淘宝", "打开美团"]
+    current_task_index = [0]  # Use list to allow modification in closure
+
+    def step_callback(result: StepResult) -> str | None:
+        """Switch to next task after current one completes"""
+        # Check if current task is done
+        if result.finished and current_task_index[0] < len(task_queue):
+            next_task = task_queue[current_task_index[0]]
+            current_task_index[0] += 1
+            print(f"\n[Callback] Switching to task: {next_task}\n")
+            return next_task
+        return None
+
+    agent = _create_agent_with_callbacks(lang=lang, step_callback=step_callback)
+
+    print(f"\n{'='*50}")
+    print(f"Task queue: {' -> '.join(task_queue)}")
+    print(f"{'='*50}\n")
+
+    # Start with first task
+    result = agent.run(task_queue.pop(0))
+
+
+# ============================================================================
+# Example 3: Interactive task control / 示例3：交互式任务控制
+# ============================================================================
+
+def example_interactive_control(lang: str = "cn"):
+    """Control task execution with user input / 通过用户输入控制任务执行"""
+    import sys
+    from select import select
+
+    msgs = get_messages(lang)
+
+    class TaskController:
+        def __init__(self):
+            self.new_task = None
+            self.should_stop = False
+            self._thread = threading.Thread(target=self._input_listener, daemon=True)
+            self._thread.start()
+
+        def _input_listener(self):
+            """Listen for user input in background"""
+            print("\n[Interactive Control]")
+            print("  Enter 's' + Enter to stop")
+            print("  Enter 'n:<task>' + Enter to switch to new task")
+            print("-" * 40)
+
+            while True:
+                try:
+                    # Non-blocking input check
+                    if select([sys.stdin], [], [], 0.1)[0]:
+                        cmd = sys.stdin.readline().strip()
+                        if cmd.lower() == 's':
+                            self.should_stop = True
+                            print("\n[Control] STOP requested\n")
+                        elif cmd.lower().startswith('n:'):
+                            self.new_task = cmd[2:]
+                            print(f"\n[Control] New task: {self.new_task}\n")
+                except (IOError, EOFError):
+                    break
+
+        def step_callback(self, result: StepResult) -> str | None:
+            """Check control flags after each step"""
+            if self.should_stop:
+                return "stop"
+            if self.new_task:
+                task = self.new_task
+                self.new_task = None
+                return task
+            return None
+
+    controller = TaskController()
+    agent = _create_agent_with_callbacks(lang=lang, step_callback=controller.step_callback)
+
+    print(f"\n{'='*50}")
+    print(f"Interactive Control Mode")
+    print(f"{'='*50}\n")
+
+    result = agent.run("打开抖音浏览视频")
+
+
+# ============================================================================
+# Example 4: Modify actions before execution / 示例4：在执行前修改操作
+# ============================================================================
+
+def example_modify_actions(lang: str = "cn"):
+    """Modify actions before they are executed / 在操作执行前进行修改"""
+    msgs = get_messages(lang)
+
+    action_log = []
+
+    def before_action_callback(action: dict) -> dict | None:
+        """Called before executing each action / 每个操作执行前调用"""
+        action_log.append(action.copy())
+
+        # Example: Prevent launching certain apps
+        if action.get("action") == "Launch":
+            app = action.get("app", "")
+            if "游戏" in app or "Game" in app:
+                print(f"[Callback] Blocked launching: {app}")
+                # Replace with a safe action
+                return {"action": "Note", "message": f"Skipped {app}"}
+
+        # Example: Add delay after Tap actions
+        if action.get("action") == "Tap":
+            print(f"[Callback] Tap at {action.get('element')}")
+
+        # Return None to proceed with original action
+        return None
+
+    agent = _create_agent_with_callbacks(lang=lang, before_action_callback=before_action_callback)
+
+    print(f"\n{'='*50}")
+    print(f"Actions will be logged and filtered")
+    print(f"{'='*50}\n")
+
+    result = agent.run("打开微信查看朋友圈")
+
+    print(f"\n{'='*50}")
+    print(f"Total actions executed: {len(action_log)}")
+    print(f"{'='*50}")
+
+
+# ============================================================================
+# Example 5: Collect statistics / 示例5：收集统计信息
+# ============================================================================
+
+def example_collect_statistics(lang: str = "cn"):
+    """Collect execution statistics / 收集执行统计信息"""
+    msgs = get_messages(lang)
+
+    stats = {
+        "actions": defaultdict(int),
+        "total_steps": 0,
+        "errors": 0,
+    }
+
+    def before_action_callback(action: dict) -> dict | None:
+        """Count action types"""
+        action_type = action.get("action", "Unknown")
+        stats["actions"][action_type] += 1
+        return None
+
+    def step_callback(result: StepResult) -> str | None:
+        """Collect step statistics"""
+        stats["total_steps"] = result.step_count
+        if not result.success:
+            stats["errors"] += 1
+        return None
+
+    agent = _create_agent_with_callbacks(
+        lang=lang,
+        before_action_callback=before_action_callback,
+        step_callback=step_callback,
+    )
+
+    print(f"\n{'='*50}")
+    print(f"Collecting execution statistics...")
+    print(f"{'='*50}\n")
+
+    result = agent.run("打开美团搜索附近的餐厅")
+
+    print(f"\n{'='*50}")
+    print(f"Execution Statistics:")
+    print(f"  Total steps: {stats['total_steps']}")
+    print(f"  Errors: {stats['errors']}")
+    print(f"\n  Action breakdown:")
+    for action_type, count in sorted(stats["actions"].items()):
+        print(f"    {action_type}: {count}")
+    print(f"{'='*50}")
+
+
+# ============================================================================
+# Example 6: Task queue with priority / 示例6：优先级任务队列
+# ============================================================================
+
+def example_priority_task_queue(lang: str = "cn"):
+    """Implement priority-based task switching / 实现基于优先级的任务切换"""
+    from queue import PriorityQueue
+    from dataclasses import dataclass, field
+
+    @dataclass(order=True)
+    class PrioritizedTask:
+        priority: int
+        task: str = field(compare=False)
+
+    task_queue = PriorityQueue()
+    task_queue.put(PrioritizedTask(priority=2, task="打开微信发消息"))
+    task_queue.put(PrioritizedTask(priority=1, task="打开淘宝"))  # Higher priority
+    task_queue.put(PrioritizedTask(priority=3, task="打开美团"))
+
+    def step_callback(result: StepResult) -> str | None:
+        """Switch to higher priority task when available"""
+        if result.finished and not task_queue.empty():
+            next_task = task_queue.get()
+            print(f"\n[Callback] Switching to priority {next_task.priority} task: {next_task.task}\n")
+            return next_task.task
+        return None
+
+    agent = _create_agent_with_callbacks(lang=lang, step_callback=step_callback)
+
+    print(f"\n{'='*50}")
+    print(f"Priority Task Queue Mode")
+    print(f"{'='*50}\n")
+
+    # Start with initial task
+    result = agent.run("打开抖音")
+
+
+# ============================================================================
+# Example 7: Timeout control / 示例7：超时控制
+# ============================================================================
+
+def example_timeout_control(lang: str = "cn"):
+    """Interrupt task if it takes too long / 任务耗时过长则中断"""
+    msgs = get_messages(lang)
+
+    start_time = [time.time()]
+    timeout_seconds = 30
+
+    def step_callback(result: StepResult) -> str | None:
+        """Check if task has timed out"""
+        elapsed = time.time() - start_time[0]
+        print(f"[Callback] Elapsed time: {elapsed:.1f}s")
+
+        if elapsed > timeout_seconds:
+            print(f"\n[Callback] Task timeout after {elapsed:.1f}s\n")
+            return "stop"
+        return None
+
+    agent = _create_agent_with_callbacks(lang=lang, step_callback=step_callback)
+
+    print(f"\n{'='*50}")
+    print(f"Task will timeout after {timeout_seconds} seconds")
+    print(f"{'='*50}\n")
+
+    result = agent.run("打开小红书浏览内容")
+
+
+# ============================================================================
+# Example 8: Conditional action blocking / 示例8：条件性操作拦截
+# ============================================================================
+
+def example_conditional_blocking(lang: str = "cn"):
+    """Block actions based on conditions / 根据条件拦截操作"""
+    msgs = get_messages(lang)
+
+    blocked_apps = ["王者荣耀", "和平精英", "抖音"]  # Apps to block
+
+    def before_action_callback(action: dict) -> dict | None:
+        """Block launching certain apps"""
+        if action.get("action") == "Launch":
+            app = action.get("app", "")
+            if any(blocked in app for blocked in blocked_apps):
+                print(f"[Callback] BLOCKED app launch: {app}")
+                # Return finish action to stop
+                return {"action": "finish", "message": f"App {app} is blocked"}
+        return None
+
+    agent = _create_agent_with_callbacks(lang=lang, before_action_callback=before_action_callback)
+
+    print(f"\n{'='*50}")
+    print(f"Blocked apps: {', '.join(blocked_apps)}")
+    print(f"{'='*50}\n")
+
+    result = agent.run("打开抖音刷视频")
+
+
+# ============================================================================
+# Main / 主程序
+# ============================================================================
+
+if __name__ == "__main__":
+    import argparse
+
+    parser = argparse.ArgumentParser(description="Phone Agent Callback Hooks Examples")
+    parser.add_argument(
+        "--example", "-e",
+        type=int,
+        default=1,
+        choices=range(1, 9),
+        help="Example number to run (1-8)",
+    )
+    parser.add_argument(
+        "--lang",
+        type=str,
+        default="cn",
+        choices=["cn", "en"],
+        help="Language for UI messages (cn=Chinese, en=English)",
+    )
+    args = parser.parse_args()
+
+    examples = {
+        1: ("Interrupt after max steps", example_interrupt_after_steps),
+        2: ("Switch tasks dynamically", example_switch_tasks),
+        3: ("Interactive task control", example_interactive_control),
+        4: ("Modify actions before execution", example_modify_actions),
+        5: ("Collect statistics", example_collect_statistics),
+        6: ("Priority task queue", example_priority_task_queue),
+        7: ("Timeout control", example_timeout_control),
+        8: ("Conditional action blocking", example_conditional_blocking),
+    }
+
+    name, func = examples[args.example]
+    print(f"\n{'='*50}")
+    print(f"Example {args.example}: {name}")
+    print(f"{'='*50}")
+
+    func(args.lang)
+
+    print(f"\n✓ Example {args.example} completed\n")

examples/run_with_callbacks.py

@@ -0,0 +1,199 @@
+#!/usr/bin/env python3
+"""
+带回调钩子的命令行工具 / Command-line tool with callback hooks
+
+支持通过命令行参数配置回调钩子，实现任务中断和切换。
+
+Configuration / 配置:
+    Loads settings from .env file (if present).
+    从 .env 文件加载配置（如果存在）。
+"""
+
+import argparse
+import os
+import sys
+from dotenv import load_dotenv
+
+# Load .env file for configuration
+# 加载 .env 配置文件
+load_dotenv()
+
+from phone_agent import PhoneAgent, AgentConfig
+from phone_agent.model import ModelConfig
+from phone_agent.config import get_messages
+
+
+def create_step_callback(max_steps: int | None = None, lang: str = "cn"):
+    """创建步数限制回调"""
+    if max_steps is None:
+        return None
+
+    def callback(result):
+        if result.step_count >= max_steps:
+            return "stop"
+        return None
+
+    return callback
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Phone Agent with Callback Hooks",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+        epilog="""
+Examples:
+    # 基本用法（与 main.py 相同）
+    python run_with_callbacks.py "打开微信"
+
+    # 限制最大执行步数
+    python run_with_callbacks.py "打开微信" --max-steps 5
+
+    # 使用智谱 API
+    python run_with_callbacks.py "打开微信" \\
+        --base-url https://open.bigmodel.cn/api/paas/v4 \\
+        --model autoglm-phone \\
+        --apikey your-key
+
+    # 交互模式（输入新任务可切换）
+    python run_with_callbacks.py --interactive
+        """
+    )
+
+    # 模型配置
+    parser.add_argument("--base-url", default=os.getenv("PHONE_AGENT_BASE_URL", "http://localhost:8000/v1"), help="模型 API 地址")
+    parser.add_argument("--model", default=os.getenv("PHONE_AGENT_MODEL", "autoglm-phone-9b"), help="模型名称")
+    parser.add_argument("--apikey", default=os.getenv("PHONE_AGENT_API_KEY", "EMPTY"), help="API 密钥")
+
+    # Agent 配置
+    parser.add_argument("--max-steps", type=int, default=int(os.getenv("PHONE_AGENT_MAX_STEPS", "100")), help="最大执行步数")
+    parser.add_argument("--device-id", default=os.getenv("PHONE_AGENT_DEVICE_ID"), help="设备 ID")
+    parser.add_argument("--lang", choices=["cn", "en"], default=os.getenv("PHONE_AGENT_LANG", "cn"), help="语言")
+
+    # 回调配置
+    parser.add_argument("--callback-max-steps", type=int, help="回调强制中断的步数")
+    parser.add_argument("--interactive", action="store_true", help="交互模式（支持动态切换任务）")
+
+    # 任务参数
+    parser.add_argument("task", nargs="?", help="要执行的任务")
+
+    args = parser.parse_args()
+
+    # 创建模型配置
+    model_config = ModelConfig(
+        base_url=args.base_url,
+        model_name=args.model,
+        api_key=args.apikey,
+        lang=args.lang,
+    )
+
+    # 创建回调
+    step_callback = create_step_callback(args.callback_max_steps, args.lang)
+
+    # 交互模式回调
+    if args.interactive:
+        import threading
+        import queue
+
+        task_queue = queue.Queue()
+        input_ready = threading.Event()
+
+        def input_listener():
+            """后台监听用户输入"""
+            print("\n[交互模式] 在任务执行时可以输入：")
+            print("  's' + Enter - 停止当前任务")
+            print("  'n:新任务' + Enter - 切换到新任务")
+            print("-" * 50)
+
+            while True:
+                try:
+                    cmd = input().strip()
+                    if cmd.lower() == 's':
+                        task_queue.put("stop")
+                    elif cmd.lower().startswith('n:'):
+                        task_queue.put(cmd[2:])
+                except (EOFError, KeyboardInterrupt):
+                    break
+
+        threading.Thread(target=input_listener, daemon=True).start()
+
+        def interactive_callback(result):
+            """交互式回调"""
+            try:
+                cmd = task_queue.get_nowait()
+                if cmd == "stop":
+                    return "stop"
+                return cmd
+            except queue.Empty:
+                return None
+
+        step_callback = interactive_callback
+
+    # 创建 Agent 配置
+    agent_config = AgentConfig(
+        max_steps=args.max_steps,
+        device_id=args.device_id,
+        lang=args.lang,
+        step_callback=step_callback,
+    )
+
+    # 创建 Agent
+    agent = PhoneAgent(model_config=model_config, agent_config=agent_config)
+
+    # 打印配置信息
+    msgs = get_messages(args.lang)
+    print("=" * 50)
+    print("Phone Agent with Callback Hooks")
+    print("=" * 50)
+    print(f"Model: {args.model}")
+    print(f"Base URL: {args.base_url}")
+    print(f"Max Steps: {args.max_steps}")
+    if args.callback_max_steps:
+        print(f"Callback Max Steps: {args.callback_max_steps} (强制中断)")
+    print("=" * 50)
+
+    # 执行任务
+    if args.task:
+        print(f"\nTask: {args.task}\n")
+        result = agent.run(args.task)
+        print(f"\nResult: {result}")
+    elif args.interactive:
+        # 交互模式循环
+        print("\n输入任务 (或 'quit' 退出):\n")
+        while True:
+            try:
+                task = input("> ").strip()
+                if task.lower() in ("quit", "exit", "q"):
+                    break
+                if not task:
+                    continue
+
+                print(f"\n执行: {task}\n")
+                result = agent.run(task)
+                print(f"\n结果: {result}\n")
+                agent.reset()
+
+            except KeyboardInterrupt:
+                print("\n\nGoodbye!")
+                break
+    else:
+        # 默认交互模式
+        print("\n输入任务 (或 'quit' 退出):\n")
+        while True:
+            try:
+                task = input("> ").strip()
+                if task.lower() in ("quit", "exit", "q"):
+                    break
+                if not task:
+                    continue
+
+                result = agent.run(task)
+                print(f"\n结果: {result}\n")
+                agent.reset()
+
+            except KeyboardInterrupt:
+                print("\n\nGoodbye!")
+                break
+
+
+if __name__ == "__main__":
+    main()

main.py

@@ -11,6 +11,9 @@ Environment Variables:
     PHONE_AGENT_API_KEY: API key for model authentication (default: EMPTY)
     PHONE_AGENT_MAX_STEPS: Maximum steps per task (default: 100)
     PHONE_AGENT_DEVICE_ID: ADB device ID for multi-device setups
+
+Configuration Files:
+    .env: Automatically loaded from current directory or project root
 """
 
 import argparse
@@ -20,8 +23,12 @@ import subprocess
 import sys
 from urllib.parse import urlparse
 
+from dotenv import load_dotenv
 from openai import OpenAI
 
+# Load .env file if exists (search in current directory and project root)
+load_dotenv()
+
 from phone_agent import PhoneAgent
 from phone_agent.agent import AgentConfig
 from phone_agent.agent_ios import IOSAgentConfig, IOSPhoneAgent

phone_agent/__init__.py

@@ -5,8 +5,14 @@ This package provides tools for automating Android and iOS phone interactions
 using AI models for visual understanding and decision making.
 """
 
-from phone_agent.agent import PhoneAgent
-from phone_agent.agent_ios import IOSPhoneAgent
+from phone_agent.agent import AgentConfig, PhoneAgent, StepResult
+from phone_agent.agent_ios import IOSAgentConfig, IOSPhoneAgent
 
 __version__ = "0.1.0"
-__all__ = ["PhoneAgent", "IOSPhoneAgent"]
+__all__ = [
+    "PhoneAgent",
+    "IOSPhoneAgent",
+    "AgentConfig",
+    "IOSAgentConfig",
+    "StepResult",
+]

phone_agent/agent.py

@@ -22,6 +22,10 @@ class AgentConfig:
     lang: str = "cn"
     system_prompt: str | None = None
     verbose: bool = True
+    step_callback: Callable[["StepResult"], str | None] | None = None
+    """Callback after each step. Return 'stop' to interrupt, or a new task string to switch."""
+    before_action_callback: Callable[[dict[str, Any]], dict[str, Any] | None] | None = None
+    """Callback before executing action. Return modified action dict, or None to proceed as-is."""
 
     def __post_init__(self):
         if self.system_prompt is None:
@@ -37,6 +41,7 @@ class StepResult:
     action: dict[str, Any] | None
     thinking: str
     message: str | None = None
+    step_count: int = 0
 
 
 class PhoneAgent:
@@ -52,12 +57,29 @@ class PhoneAgent:
         confirmation_callback: Optional callback for sensitive action confirmation.
         takeover_callback: Optional callback for takeover requests.
 
+    Callbacks in agent_config:
+        step_callback: Called after each step with StepResult.
+            - Return 'stop' to interrupt the task
+            - Return a new task string to switch tasks
+            - Return None to continue normally
+
+        before_action_callback: Called before executing an action with the action dict.
+            - Return modified action dict to override
+            - Return None to execute the original action
+
     Example:
-        >>> from phone_agent import PhoneAgent
+        >>> from phone_agent import PhoneAgent, AgentConfig
         >>> from phone_agent.model import ModelConfig
         >>>
+        >>> # With callback
+        >>> def on_step(result):
+        ...     if result.step_count > 10:
+        ...         return "stop"  # Interrupt after 10 steps
+        ...     return None
+        >>>
         >>> model_config = ModelConfig(base_url="http://localhost:8000/v1")
-        >>> agent = PhoneAgent(model_config)
+        >>> agent_config = AgentConfig(step_callback=on_step)
+        >>> agent = PhoneAgent(model_config, agent_config)
         >>> agent.run("Open WeChat and send a message to John")
     """
 
@@ -184,6 +206,7 @@ class PhoneAgent:
                 action=None,
                 thinking="",
                 message=f"Model error: {e}",
+                step_count=self._step_count,
             )
 
         # Parse action from response
@@ -204,6 +227,16 @@ class PhoneAgent:
         # Remove image from context to save space
         self._context[-1] = MessageBuilder.remove_images_from_message(self._context[-1])
 
+        # Before action callback - allow modifying or intercepting action
+        if self.agent_config.before_action_callback is not None:
+            try:
+                modified_action = self.agent_config.before_action_callback(action)
+                if modified_action is not None:
+                    action = modified_action
+            except Exception as e:
+                if self.agent_config.verbose:
+                    print(f"Warning: before_action_callback error: {e}")
+
         # Execute action
         try:
             result = self.action_handler.execute(
@@ -234,14 +267,38 @@ class PhoneAgent:
             )
             print("=" * 50 + "\n")
 
-        return StepResult(
+        # Build step result
+        step_result = StepResult(
             success=result.success,
             finished=finished,
             action=action,
             thinking=response.thinking,
             message=result.message or action.get("message"),
+            step_count=self._step_count,
         )
 
+        # Step callback - allow interrupting or switching tasks
+        if self.agent_config.step_callback is not None and not finished:
+            try:
+                callback_result = self.agent_config.step_callback(step_result)
+                if callback_result == "stop":
+                    # Interrupt the task
+                    if self.agent_config.verbose:
+                        print("\n⏹ Task interrupted by callback\n")
+                    step_result.finished = True
+                    return step_result
+                elif isinstance(callback_result, str):
+                    # Switch to new task
+                    if self.agent_config.verbose:
+                        print(f"\n🔄 Switching to new task: {callback_result}\n")
+                    self.reset()
+                    return self._execute_step(callback_result, is_first=True)
+            except Exception as e:
+                if self.agent_config.verbose:
+                    print(f"Warning: step_callback error: {e}")
+
+        return step_result
+
     @property
     def context(self) -> list[dict[str, Any]]:
         """Get the current conversation context."""

phone_agent/agent_ios.py

@@ -24,6 +24,10 @@ class IOSAgentConfig:
     lang: str = "cn"
     system_prompt: str | None = None
     verbose: bool = True
+    step_callback: Callable[[Any], str | None] | None = None
+    """Callback after each step. Return 'stop' to interrupt, or a new task string to switch."""
+    before_action_callback: Callable[[dict[str, Any]], dict[str, Any] | None] | None = None
+    """Callback before executing action. Return modified action dict, or None to proceed as-is."""
 
     def __post_init__(self):
         if self.system_prompt is None:
@@ -39,6 +43,7 @@ class StepResult:
     action: dict[str, Any] | None
     thinking: str
     message: str | None = None
+    step_count: int = 0
 
 
 class IOSPhoneAgent:
@@ -54,12 +59,28 @@ class IOSPhoneAgent:
         confirmation_callback: Optional callback for sensitive action confirmation.
         takeover_callback: Optional callback for takeover requests.
 
+    Callbacks in agent_config:
+        step_callback: Called after each step with StepResult.
+            - Return 'stop' to interrupt the task
+            - Return a new task string to switch tasks
+            - Return None to continue normally
+
+        before_action_callback: Called before executing an action with the action dict.
+            - Return modified action dict to override
+            - Return None to execute the original action
+
     Example:
         >>> from phone_agent.agent_ios import IOSPhoneAgent, IOSAgentConfig
         >>> from phone_agent.model import ModelConfig
         >>>
+        >>> # With callback
+        >>> def on_step(result):
+        ...     if result.step_count > 10:
+        ...         return "stop"  # Interrupt after 10 steps
+        ...     return None
+        >>>
         >>> model_config = ModelConfig(base_url="http://localhost:8000/v1")
-        >>> agent_config = IOSAgentConfig(wda_url="http://localhost:8100")
+        >>> agent_config = IOSAgentConfig(wda_url="http://localhost:8100", step_callback=on_step)
         >>> agent = IOSPhoneAgent(model_config, agent_config)
         >>> agent.run("Open Safari and search for Apple")
     """
@@ -203,6 +224,7 @@ class IOSPhoneAgent:
                 action=None,
                 thinking="",
                 message=f"Model error: {e}",
+                step_count=self._step_count,
             )
 
         # Parse action from response
@@ -228,6 +250,16 @@ class IOSPhoneAgent:
         # Remove image from context to save space
         self._context[-1] = MessageBuilder.remove_images_from_message(self._context[-1])
 
+        # Before action callback - allow modifying or intercepting action
+        if self.agent_config.before_action_callback is not None:
+            try:
+                modified_action = self.agent_config.before_action_callback(action)
+                if modified_action is not None:
+                    action = modified_action
+            except Exception as e:
+                if self.agent_config.verbose:
+                    print(f"Warning: before_action_callback error: {e}")
+
         # Execute action
         try:
             result = self.action_handler.execute(
@@ -258,14 +290,38 @@ class IOSPhoneAgent:
             )
             print("=" * 50 + "\n")
 
-        return StepResult(
+        # Build step result
+        step_result = StepResult(
             success=result.success,
             finished=finished,
             action=action,
             thinking=response.thinking,
             message=result.message or action.get("message"),
+            step_count=self._step_count,
         )
 
+        # Step callback - allow interrupting or switching tasks
+        if self.agent_config.step_callback is not None and not finished:
+            try:
+                callback_result = self.agent_config.step_callback(step_result)
+                if callback_result == "stop":
+                    # Interrupt the task
+                    if self.agent_config.verbose:
+                        print("\n⏹ Task interrupted by callback\n")
+                    step_result.finished = True
+                    return step_result
+                elif isinstance(callback_result, str):
+                    # Switch to new task
+                    if self.agent_config.verbose:
+                        print(f"\n🔄 Switching to new task: {callback_result}\n")
+                    self.reset()
+                    return self._execute_step(callback_result, is_first=True)
+            except Exception as e:
+                if self.agent_config.verbose:
+                    print(f"Warning: step_callback error: {e}")
+
+        return step_result
+
     @property
     def context(self) -> list[dict[str, Any]]:
         """Get the current conversation context."""

requirements.txt

@@ -1,9 +1,18 @@
 Pillow>=12.0.0
 openai>=2.9.0
+python-dotenv>=1.0.0
 
 # For iOS Support
 requests>=2.31.0
 
+# For Web Dashboard
+fastapi>=0.104.0
+uvicorn[standard]>=0.24.0
+websockets>=12.0
+pydantic>=2.5.0
+python-multipart>=0.0.6
+aiofiles>=23.2.0
+
 # For Model Deployment
 
 ## After installing sglang or vLLM, please run pip install -U transformers again to upgrade to 5.0.0rc0.

scripts/run_dashboard.bat

@@ -0,0 +1,28 @@
+@echo off
+REM AutoGLM Dashboard - Run Script for Windows
+
+REM Change to project root
+cd /d "%~dp0.."
+
+REM Load .env file if it exists (use set command to parse)
+if exist .env (
+    for /f "tokens=*" %%a in ('type .env ^| findstr /v "^#"') do set %%a
+)
+
+REM Set defaults
+if not defined DASHBOARD_HOST set DASHBOARD_HOST=0.0.0.0
+if not defined DASHBOARD_PORT set DASHBOARD_PORT=8080
+if not defined DASHBOARD_DEBUG set DASHBOARD_DEBUG=false
+
+REM Print configuration
+echo ==========================================
+echo   AutoGLM Dashboard
+echo ==========================================
+echo Host: %DASHBOARD_HOST%
+echo Port: %DASHBOARD_PORT%
+echo Debug: %DASHBOARD_DEBUG%
+echo ==========================================
+echo.
+
+REM Run the dashboard
+python -m uvicorn dashboard.main:app --host %DASHBOARD_HOST% --port %DASHBOARD_PORT% --reload

scripts/run_dashboard.sh

@@ -0,0 +1,34 @@
+#!/bin/bash
+# AutoGLM Dashboard - Run Script
+
+# Get the directory where this script is located
+SCRIPT_DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
+PROJECT_ROOT="$(dirname "$SCRIPT_DIR")"
+
+# Change to project root
+cd "$PROJECT_ROOT"
+
+# Load .env file if it exists
+if [ -f .env ]; then
+    export $(grep -v '^#' .env | xargs)
+fi
+
+# Set defaults
+DASHBOARD_HOST=${DASHBOARD_HOST:-0.0.0.0}
+DASHBOARD_PORT=${DASHBOARD_PORT:-8080}
+DASHBOARD_DEBUG=${DASHBOARD_DEBUG:-false}
+
+# Print configuration
+echo "=========================================="
+echo "  AutoGLM Dashboard"
+echo "=========================================="
+echo "Host: $DASHBOARD_HOST"
+echo "Port: $DASHBOARD_PORT"
+echo "Debug: $DASHBOARD_DEBUG"
+echo "=========================================="
+
+# Run the dashboard
+python -m uvicorn dashboard.main:app \
+    --host "$DASHBOARD_HOST" \
+    --port "$DASHBOARD_PORT" \
+    --reload

setup.py

@@ -32,6 +32,8 @@ setup(
     install_requires=[
         "Pillow>=12.0.0",
         "openai>=2.9.0",
+        "python-dotenv>=1.0.0",
+        "requests>=2.31.0",
     ],
     extras_require={
         "dev": [
@@ -40,10 +42,19 @@ setup(
             "mypy>=1.0.0",
             "ruff>=0.1.0",
         ],
+        "dashboard": [
+            "fastapi>=0.104.0",
+            "uvicorn[standard]>=0.24.0",
+            "websockets>=12.0",
+            "pydantic>=2.5.0",
+            "python-multipart>=0.0.6",
+            "aiofiles>=23.2.0",
+        ],
     },
     entry_points={
         "console_scripts": [
             "phone-agent=main:main",
+            "autoglm-dashboard=dashboard.main:main",
         ],
     },
 )