app.py

"""
LifeFlow AI - Multi-User Safe with Real Streaming
支持多用戶並發 + 真正的串流輸出 + 安全的設定隔離
"""

import sys
from pathlib import Path
import gradio as gr
from datetime import datetime
import time as time_module
import json
import uuid
from typing import Dict, Any, Optional

# ===== 導入配置 =====
from config import DEFAULT_SETTINGS, APP_TITLE

# ===== 導入 UI 組件 =====
from ui.theme import get_enhanced_css
from ui.components.header import create_header, create_top_controls
from ui.components.input_form import create_input_form, toggle_location_inputs
from ui.components.confirmation import create_confirmation_area, create_exit_button
from ui.components.results import create_team_area, create_result_area, create_tabs
from ui.components.modals import create_settings_modal, create_doc_modal

# ===== 導入核心工具 =====
from core.utils import (
    create_agent_stream_output, create_agent_card_enhanced,
    create_task_card, create_summary_card, create_animated_map,
    get_reasoning_html_reversed, create_celebration_animation,
    create_result_visualization
)

# ===== 導入 Agno Agent 組件 =====
from agno.models.google import Gemini
from agno.agent import RunEvent
from agno.run.team import TeamRunEvent

# ===== 導入 Agent 系統 =====
from src.agent.base import UserState, Location, get_context
from src.agent.planner import create_planner_agent
from src.agent.core_team import create_core_team
from src.infra.context import set_session_id, get_session_id
from src.infra.poi_repository import poi_repo
from src.tools import (
    ScoutToolkit, OptimizationToolkit,
    NavigationToolkit, WeatherToolkit, ReaderToolkit
)
from src.infra.config import get_settings
from src.infra.logger import get_logger

logger = get_logger(__name__)


class UserSession:
    """
    單個用戶的會話數據
    每個用戶有獨立的實例，確保資料隔離
    """

    def __init__(self):
        self.session_id: Optional[str] = None
        self.planner_agent = None
        self.core_team = None
        self.user_state: Optional[UserState] = None
        self.task_list: list = []
        self.reasoning_messages: list = []
        self.chat_history: list = []
        self.planning_completed: bool = False

        # 存儲位置信息（用於重新初始化 Agent）
        self.lat: Optional[float] = None
        self.lng: Optional[float] = None

        # [Security Fix] 用戶個人的自定義設定 (API Keys, Model Choice)
        self.custom_settings: Dict[str, Any] = {}

        # 系統預設設定 (從環境變數讀取)
        self.agno_settings = get_settings()

    def to_dict(self) -> Dict[str, Any]:
        """序列化為字典（用於 Gradio State）"""
        return {
            'session_id': self.session_id,
            'task_list': self.task_list,
            'reasoning_messages': self.reasoning_messages,
            'chat_history': self.chat_history,
            'planning_completed': self.planning_completed,
            'lat': self.lat,
            'lng': self.lng,
            # [Security Fix] 保存用戶自定義設定
            'custom_settings': self.custom_settings
        }

    @classmethod
    def from_dict(cls, data: Dict[str, Any]) -> 'UserSession':
        """從字典恢復（用於 Gradio State）"""
        session = cls()
        session.session_id = data.get('session_id')
        session.task_list = data.get('task_list', [])
        session.reasoning_messages = data.get('reasoning_messages', [])
        session.chat_history = data.get('chat_history', [])
        session.planning_completed = data.get('planning_completed', False)
        session.lat = data.get('lat')
        session.lng = data.get('lng')
        # [Security Fix] 恢復用戶自定義設定
        session.custom_settings = data.get('custom_settings', {})
        return session


class LifeFlowAI:
    """LifeFlow AI - Multi-User Safe Version"""

    def __init__(self):
        # [Security Fix] 移除全域 settings，避免用戶間資料汙染
        pass

    def _initialize_agents(self, session: UserSession, lat: float, lng: float):
        """初始化 Agents（每個用戶獨立，並應用用戶設定）"""
        session.lat = lat
        session.lng = lng

        if session.planner_agent is not None:
            logger.debug(f"Agents already initialized for session {session.session_id}")
            return

        # 生成 Session ID
        if session.session_id is None:
            session.session_id = str(uuid.uuid4())
            token = set_session_id(session.session_id)
            logger.info(f"🆔 New Session: {session.session_id}")
        else:
            set_session_id(session.session_id)
            logger.info(f"🔄 Restoring Session: {session.session_id}")

        # 設定用戶狀態
        session.user_state = UserState(location=Location(lat=lat, lng=lng))

        # [Security Fix] 讀取用戶選擇的模型，如果沒有則使用預設
        selected_model_id = session.custom_settings.get('model', 'gemini-2.5-flash')

        # [Security Fix] 優先使用用戶提供的 API Key (這裡以 Gemini 為例，若支援其他模型需擴充邏輯)
        # 注意：實際應用中需根據選擇的 Model ID (Claude/Gemini) 來決定使用哪個 Key
        gemini_key = session.custom_settings.get('gemini_api_key') or session.agno_settings.gemini_api_key

        # 初始化模型 (應用設定)
        planner_model = Gemini(
            id=selected_model_id,
            thinking_budget=2048,
            api_key=gemini_key
        )

        main_model = Gemini(
            id=selected_model_id,
            thinking_budget=1024,
            api_key=gemini_key
        )

        lite_model = Gemini(
            id="gemini-2.5-flash-lite",  # 輕量級模型通常固定或由次要選項決定
            api_key=gemini_key
        )

        # 配置模型和工具
        models_dict = {
            "team": main_model,
            "scout": main_model,
            "optimizer": lite_model,
            "navigator": lite_model,
            "weatherman": lite_model,
            "presenter": main_model,
        }

        # [Note] 如果 Toolkit 支援傳入 API Key，應在此處從 session.custom_settings 傳入
        tools_dict = {
            "scout": [ScoutToolkit()],
            "optimizer": [OptimizationToolkit()],
            "navigator": [NavigationToolkit()],
            "weatherman": [WeatherToolkit()],
            "presenter": [ReaderToolkit()],
        }

        planner_kwargs = {
            "additional_context": get_context(session.user_state),
            "timezone_identifier": session.user_state.utc_offset,
            "debug_mode": False,
        }

        team_kwargs = {
            "timezone_identifier": session.user_state.utc_offset,
        }

        # 創建 Agents
        session.planner_agent = create_planner_agent(
            planner_model,
            planner_kwargs,
            session_id=session.session_id
        )

        session.core_team = create_core_team(
            models_dict,
            team_kwargs,
            tools_dict,
            session_id=session.session_id
        )

        logger.info(f"✅ Agents initialized for session {session.session_id} using model {selected_model_id}")

    def step1_analyze_tasks(self, user_input: str, auto_location: bool,
                            lat: float, lon: float, session: UserSession):
        """Step 1: 真正的串流分析"""
        if not user_input.strip():
            yield from self._empty_step1_outputs(session)
            return

        if auto_location:
            lat, lon = 25.033, 121.565

        try:
            self._initialize_agents(session, lat, lon)

            self._add_reasoning(session, "planner", "🚀 Starting analysis...")
            yield self._create_step1_outputs(
                stream_text="🤔 Analyzing your request with AI...",
                session=session,
                agent_status=("planner", "working", "Initializing...")
            )

            time_module.sleep(0.3)

            self._add_reasoning(session, "planner", f"Processing: {user_input[:50]}...")
            yield self._create_step1_outputs(
                stream_text="🤔 Analyzing your request with AI...\n📋 AI is extracting tasks...",
                session=session,
                agent_status=("planner", "working", "Extracting tasks...")
            )

            planner_stream = session.planner_agent.run(
                f"help user to update the task_list, user's message: {user_input}",
                stream=True,
                stream_events=True
            )

            accumulated_response = ""
            displayed_text = "🤔 Analyzing your request with AI...\n📋 AI is extracting tasks...\n\n"
            show_content = True

            for chunk in planner_stream:
                if chunk.event == RunEvent.run_content:
                    content = chunk.content
                    accumulated_response += content

                    if show_content:
                        if "@@@" in accumulated_response:
                            show_content = False
                            remaining = content.split("@@@")[0]
                            if remaining:
                                displayed_text += remaining
                        else:
                            displayed_text += content

                        yield self._create_step1_outputs(
                            stream_text=displayed_text,
                            session=session,
                            agent_status=("planner", "working", "Processing...")
                        )

            json_data = "{" + accumulated_response.split("{", maxsplit=1)[-1]
            json_data = json_data.replace("`", "").replace("@", "").replace("\\", " ").replace("\n", " ")

            session.planner_agent.update_session_state(
                session_id=session.session_id,
                session_state_updates={"task_list": json_data}
            )

            try:
                task_list_data = json.loads(json_data)
                session.task_list = self._convert_task_list_to_ui_format(task_list_data)
            except json.JSONDecodeError as e:
                logger.error(f"Failed to parse task_list: {e}")
                session.task_list = []

            self._add_reasoning(session, "planner", f"✅ Extracted {len(session.task_list)} tasks")

            high_priority = sum(1 for t in session.task_list if t.get("priority") == "HIGH")
            total_time = sum(
                int(t.get("duration", "0").split()[0])
                for t in session.task_list
                if t.get("duration")
            )

            final_text = displayed_text + f"\n✅ Analysis complete! Found {len(session.task_list)} tasks."

            yield self._create_step1_complete_outputs(
                stream_text=final_text,
                session=session,
                high_priority=high_priority,
                total_time=total_time
            )

        except Exception as e:
            logger.error(f"Error in step1: {e}", exc_info=True)
            yield self._create_error_outputs(str(e), session)

    def modify_task_chat(self, user_message: str, session: UserSession):
        """修改任務（帶真正的串流）"""
        if not user_message.strip():
            chat_html = self._generate_chat_history_html(session)
            task_html = self._generate_task_list_html(session)
            yield chat_html, task_html, session.to_dict()
            return

        session.chat_history.append({
            "role": "user",
            "message": user_message,
            "time": datetime.now().strftime("%H:%M:%S")
        })

        yield (
            self._generate_chat_history_html(session),
            self._generate_task_list_html(session),
            session.to_dict()
        )

        try:
            if session.planner_agent is None:
                if session.lat is not None and session.lng is not None:
                    session.chat_history.append({
                        "role": "assistant",
                        "message": "🔄 Restoring AI system...",
                        "time": datetime.now().strftime("%H:%M:%S")
                    })
                    yield (
                        self._generate_chat_history_html(session),
                        self._generate_task_list_html(session),
                        session.to_dict()
                    )
                    self._initialize_agents(session, session.lat, session.lng)
                    session.chat_history.pop()
                else:
                    session.chat_history.append({
                        "role": "assistant",
                        "message": "❌ Error: Please restart the planning process.",
                        "time": datetime.now().strftime("%H:%M:%S")
                    })
                    yield (
                        self._generate_chat_history_html(session),
                        self._generate_task_list_html(session),
                        session.to_dict()
                    )
                    return

            session.chat_history.append({
                "role": "assistant",
                "message": "🤔 AI is thinking...",
                "time": datetime.now().strftime("%H:%M:%S")
            })
            yield (
                self._generate_chat_history_html(session),
                self._generate_task_list_html(session),
                session.to_dict()
            )

            planner_stream = session.planner_agent.run(
                f"help user to update the task_list, user's message: {user_message}",
                stream=True,
                stream_events=True
            )

            accumulated_response = ""
            displayed_thinking = "🤔 AI is thinking...\n\n"
            show_content = True

            for chunk in planner_stream:
                if chunk.event == RunEvent.run_content:
                    content = chunk.content
                    accumulated_response += content

                    if show_content:
                        if "@@@" in accumulated_response:
                            show_content = False
                            content = content.split("@@@")[0]

                        if content:
                            displayed_thinking += content
                            session.chat_history[-1] = {
                                "role": "assistant",
                                "message": displayed_thinking,
                                "time": datetime.now().strftime("%H:%M:%S")
                            }
                            yield (
                                self._generate_chat_history_html(session),
                                self._generate_task_list_html(session),
                                session.to_dict()
                            )

            json_data = "{" + accumulated_response.split("{", maxsplit=1)[-1]
            json_data = json_data.replace("`", "").replace("@", "").replace("\\", " ").replace("\n", " ")

            session.planner_agent.update_session_state(
                session_id=session.session_id,
                session_state_updates={"task_list": json_data}
            )

            task_list_data = json.loads(json_data)
            session.task_list = self._convert_task_list_to_ui_format(task_list_data)

            session.chat_history[-1] = {
                "role": "assistant",
                "message": "✅ Tasks updated based on your request",
                "time": datetime.now().strftime("%H:%M:%S")
            }

            self._add_reasoning(session, "planner", f"Updated: {user_message[:30]}...")

            yield (
                self._generate_chat_history_html(session),
                self._generate_task_list_html(session),
                session.to_dict()
            )

        except Exception as e:
            logger.error(f"Error in modify_task_chat: {e}", exc_info=True)
            if session.chat_history and "thinking" in session.chat_history[-1].get("message", "").lower():
                session.chat_history.pop()
            session.chat_history.append({
                "role": "assistant",
                "message": f"❌ Error: {str(e)}",
                "time": datetime.now().strftime("%H:%M:%S")
            })
            yield (
                self._generate_chat_history_html(session),
                self._generate_task_list_html(session),
                session.to_dict()
            )

    def step2_search_pois(self, session: UserSession):
        """Step 2: Scout 開始工作"""
        self._add_reasoning(session, "team", "🚀 Core Team activated")
        self._add_reasoning(session, "scout", "Searching for POIs...")

        agent_updates = [
            create_agent_card_enhanced("planner", "complete", "Tasks ready"),
            create_agent_card_enhanced("scout", "working", "Searching..."),
            *[create_agent_card_enhanced(k, "idle", "On standby")
              for k in ["optimizer", "validator", "weather", "traffic"]]
        ]

        return (
            get_reasoning_html_reversed(session.reasoning_messages),
            "🗺️ Scout is searching...",
            *agent_updates,
            session.to_dict()
        )

    def step3_run_core_team(self, session: UserSession):
        """Step 3: 運行 Core Team（帶串流）"""
        try:
            if session.core_team is None or session.planner_agent is None:
                if session.lat is not None and session.lng is not None:
                    self._initialize_agents(session, session.lat, session.lng)
                else:
                    raise ValueError("Cannot restore agents: location not available")

            set_session_id(session.session_id)
            task_list_input = session.planner_agent.get_session_state().get("task_list")

            if not task_list_input:
                raise ValueError("No task list available")

            if isinstance(task_list_input, str):
                task_list_str = task_list_input
            else:
                task_list_str = json.dumps(task_list_input, indent=2, ensure_ascii=False)

            self._add_reasoning(session, "team", "🎯 Multi-agent collaboration started")

            team_stream = session.core_team.run(
                f"Plan this trip: {task_list_str}",
                stream=True,
                stream_events=True,
                session_id=session.session_id
            )

            report_content = ""

            for event in team_stream:
                if event.event in [TeamRunEvent.run_content]:
                    report_content += event.content
                elif event.event == "tool_call":
                    tool_name = event.tool_call.get('function', {}).get('name', 'unknown')
                    self._add_reasoning(session, "team", f"🔧 Tool: {tool_name}")
                elif event.event == TeamRunEvent.run_completed:
                    self._add_reasoning(session, "team", "🎉 Completed")

            report_html = f"## 🎯 Planning Complete\n\n{report_content}..."
            return report_html, session.to_dict()

        except Exception as e:
            logger.error(f"Error in step3: {e}")
            return f"## ❌ Error\n\n{str(e)}", session.to_dict()

    def step4_finalize(self, session: UserSession):
        """Step 4: 完成"""
        try:
            final_ref_id = poi_repo.get_last_id_by_session(session.session_id)
            if not final_ref_id:
                raise ValueError(f"No final result found for session {session.session_id}")

            structured_data = poi_repo.load(final_ref_id)
            timeline = structured_data.get("timeline", [])
            metrics = structured_data.get("metrics", {})
            traffic_summary = structured_data.get("traffic_summary", {})

            timeline_html = self._generate_timeline_html(timeline)
            metrics_html = self._generate_metrics_html(metrics, traffic_summary)

            safe_task_list = session.task_list if session.task_list else []
            result_viz = create_result_visualization(safe_task_list, structured_data)
            map_fig = self._generate_map_from_data(structured_data)

            agent_updates = [
                create_agent_card_enhanced(k, "complete", "✓ Complete")
                for k in ["planner", "scout", "optimizer", "validator", "weather", "traffic"]
            ]

            self._add_reasoning(session, "team", "🎉 All completed")
            session.planning_completed = True

            return (
                timeline_html, metrics_html, result_viz, map_fig,
                gr.update(visible=True), gr.update(visible=False),
                "🎉 Planning completed!",
                *agent_updates,
                session.to_dict()
            )

        except Exception as e:
            logger.error(f"Error in step4: {e}", exc_info=True)
            error_html = f"<div style='color:red'>Error: {str(e)}</div>"
            default_map = create_animated_map()
            agent_updates = [
                create_agent_card_enhanced(k, "idle", "Waiting...")
                for k in ["planner", "scout", "optimizer", "validator", "weather", "traffic"]
            ]
            return (
                error_html, error_html, error_html, default_map,
                gr.update(visible=True), gr.update(visible=False),
                f"Error: {str(e)}",
                *agent_updates,
                session.to_dict()
            )

    # ===== 輔助方法 =====

    # [Security Fix] 新增：保存設定到 User Session
    def save_settings(self, google_key, weather_key, anthropic_key, model, session_data):
        """保存設定到用戶 Session"""
        session = UserSession.from_dict(session_data)

        session.custom_settings['google_maps_api_key'] = google_key
        session.custom_settings['openweather_api_key'] = weather_key
        session.custom_settings['anthropic_api_key'] = anthropic_key
        session.custom_settings['model'] = model

        return "✅ Settings saved locally!", session.to_dict()

    def _add_reasoning(self, session: UserSession, agent: str, message: str):
        session.reasoning_messages.append({
            "agent": agent,
            "message": message,
            "time": datetime.now().strftime("%H:%M:%S")
        })

    def _create_step1_outputs(self, stream_text: str, session: UserSession, agent_status: tuple):
        agent_key, status, message = agent_status
        return (
            self._create_stream_html(stream_text), "", "",
            get_reasoning_html_reversed(session.reasoning_messages),
            gr.update(visible=False), gr.update(visible=False), "", message,
            create_agent_card_enhanced(agent_key, status, message),
            create_agent_card_enhanced("scout", "idle", "On standby"),
            create_agent_card_enhanced("optimizer", "idle", "On standby"),
            create_agent_card_enhanced("validator", "idle", "On standby"),
            create_agent_card_enhanced("weather", "idle", "On standby"),
            create_agent_card_enhanced("traffic", "idle", "On standby"),
            session.to_dict()
        )

    def _create_step1_complete_outputs(self, stream_text: str, session: UserSession,
                                       high_priority: int, total_time: int):
        summary_html = create_summary_card(len(session.task_list), high_priority, total_time)
        task_html = self._generate_task_list_html(session)
        return (
            self._create_stream_html(stream_text), summary_html, task_html,
            get_reasoning_html_reversed(session.reasoning_messages),
            gr.update(visible=True), gr.update(visible=False),
            self._generate_chat_welcome_html(), "✓ Tasks extracted",
            create_agent_card_enhanced("planner", "complete", "Tasks ready"),
            create_agent_card_enhanced("scout", "idle", "On standby"),
            create_agent_card_enhanced("optimizer", "idle", "On standby"),
            create_agent_card_enhanced("validator", "idle", "On standby"),
            create_agent_card_enhanced("weather", "idle", "On standby"),
            create_agent_card_enhanced("traffic", "idle", "On standby"),
            session.to_dict()
        )

    def _empty_step1_outputs(self, session: UserSession):
        yield (
            create_agent_stream_output(), "", "",
            get_reasoning_html_reversed(),
            gr.update(visible=False), gr.update(visible=False), "",
            "Waiting for input...",
            create_agent_card_enhanced("planner", "idle", "On standby"),
            create_agent_card_enhanced("scout", "idle", "On standby"),
            create_agent_card_enhanced("optimizer", "idle", "On standby"),
            create_agent_card_enhanced("validator", "idle", "On standby"),
            create_agent_card_enhanced("weather", "idle", "On standby"),
            create_agent_card_enhanced("traffic", "idle", "On standby"),
            session.to_dict()
        )

    def _create_error_outputs(self, error: str, session: UserSession):
        return (
            self._create_stream_html(f"❌ Error: {error}"), "", "",
            get_reasoning_html_reversed(session.reasoning_messages),
            gr.update(visible=False), gr.update(visible=False), "",
            f"Error: {error}",
            create_agent_card_enhanced("planner", "idle", "Error occurred"),
            create_agent_card_enhanced("scout", "idle", "On standby"),
            create_agent_card_enhanced("optimizer", "idle", "On standby"),
            create_agent_card_enhanced("validator", "idle", "On standby"),
            create_agent_card_enhanced("weather", "idle", "On standby"),
            create_agent_card_enhanced("traffic", "idle", "On standby"),
            session.to_dict()
        )

    def _create_stream_html(self, text: str) -> str:
        return f"""<div class="stream-container"><div class="stream-text">{text}<span class="stream-cursor"></span></div></div>"""

    def _convert_task_list_to_ui_format(self, task_list_data):
        ui_tasks = []
        if isinstance(task_list_data, dict):
            tasks = task_list_data.get("tasks", [])
        elif isinstance(task_list_data, list):
            tasks = task_list_data
        else:
            return []

        for i, task in enumerate(tasks, 1):
            ui_task = {
                "id": i,
                "title": task.get("description", "Task"),
                "priority": task.get("priority", "MEDIUM"),
                "time": task.get("time_window", "Anytime"),
                "duration": f"{task.get('duration_minutes', 30)} minutes",
                "location": task.get("location_hint", "To be determined"),
                "icon": self._get_task_icon(task.get("category", "other"))
            }
            ui_tasks.append(ui_task)
        return ui_tasks

    def _get_task_icon(self, category: str) -> str:
        icons = {
            "medical": "🏥", "shopping": "🛒", "postal": "📮",
            "food": "🍽️", "entertainment": "🎭", "transportation": "🚗",
            "other": "📋"
        }
        return icons.get(category.lower(), "📋")

    def _generate_task_list_html(self, session: UserSession) -> str:
        if not session.task_list:
            return "<p>No tasks available</p>"
        html = ""
        for task in session.task_list:
            html += create_task_card(
                task["id"], task["title"], task["priority"],
                task["time"], task["duration"], task["location"],
                task.get("icon", "📋")
            )
        return html

    def _generate_chat_history_html(self, session: UserSession) -> str:
        if not session.chat_history:
            return self._generate_chat_welcome_html()
        html = '<div class="chat-history">'
        for msg in session.chat_history:
            role_class = "user" if msg["role"] == "user" else "assistant"
            icon = "👤" if msg["role"] == "user" else "🤖"
            html += f"""
            <div class="chat-message {role_class}">
                <span class="chat-icon">{icon}</span>
                <div class="chat-content">
                    <div class="chat-text">{msg["message"]}</div>
                    <div class="chat-time">{msg["time"]}</div>
                </div>
            </div>
            """
        html += '</div>'
        return html

    def _generate_chat_welcome_html(self) -> str:
        return """
        <div class="chat-welcome">
            <h3>💬 Chat with LifeFlow AI</h3>
            <p>Modify your tasks by chatting here.</p>
        </div>
        """

    def _generate_timeline_html(self, timeline):
        """
        生成右側 Timeline Tab 的 HTML
        使用 structured_data['timeline'] 中的豐富數據
        """
        if not timeline:
            return "<div>No timeline data</div>"

        html = '<div class="timeline-container">'
        for i, stop in enumerate(timeline):
            time = stop.get("time", "N/A")
            location = stop.get("location", "Unknown Location")
            weather = stop.get("weather", "")
            aqi = stop.get("aqi", {}).get("label", "")
            travel_time = stop.get("travel_time_from_prev", "")

            # 決定標記樣式
            marker_class = "start" if i == 0 else "stop"

            html += f"""
            <div class="timeline-item">
                <div class="timeline-marker {marker_class}">{i}</div>
                <div class="timeline-content">
                    <div class="timeline-header">
                        <span class="time-badge">{time}</span>
                        <h4>{location}</h4>
                    </div>
                    <div class="timeline-details">
                        {f'<span class="weather-tag">🌤️ {weather}</span>' if weather else ''}
                        {f'<span class="aqi-tag">{aqi}</span>' if aqi else ''}
                    </div>
                    {f'<p class="travel-info">🚗 Drive: {travel_time}</p>' if i > 0 else ''}
                </div>
            </div>
            """
        html += '</div>'

        # 補充少許 CSS 以確保美觀
        html += """
        <style>
        .timeline-header { display: flex; align-items: center; gap: 10px; margin-bottom: 5px; }
        .time-badge { background: #e3f2fd; color: #1976d2; padding: 2px 8px; border-radius: 4px; font-weight: bold; font-size: 12px; }
        .timeline-details { display: flex; gap: 8px; flex-wrap: wrap; margin-bottom: 5px; }
        .weather-tag, .aqi-tag { font-size: 11px; padding: 2px 6px; background: #f5f5f5; border-radius: 4px; border: 1px solid #e0e0e0; }
        .travel-info { font-size: 12px; color: #666; margin: 0; display: flex; align-items: center; gap: 5px; }
        </style>
        """
        return html

    def _generate_metrics_html(self, metrics, traffic_summary):
        """生成左側的簡單 Metrics"""
        if traffic_summary is None: traffic_summary = {}
        if metrics is None: metrics = {}

        total_distance = traffic_summary.get("total_distance_km", 0)
        total_duration = traffic_summary.get("total_duration_min", 0)
        efficiency = metrics.get("route_efficiency_pct", 90)

        # 根據效率改變顏色
        eff_color = "#50C878" if efficiency >= 80 else "#F5A623"

        return f"""
        <div class="metrics-container">
            <div class="metric-card">
                <h3>📏 Distance</h3>
                <p class="metric-value">{total_distance:.1f} km</p>
            </div>
            <div class="metric-card">
                <h3>⏱️ Duration</h3>
                <p class="metric-value">{total_duration:.0f} min</p>
            </div>
            <div class="metric-card" style="border-bottom: 3px solid {eff_color}">
                <h3>⚡ Efficiency</h3>
                <p class="metric-value" style="color: {eff_color}">{efficiency:.0f}%</p>
            </div>
        </div>
        """

    def _generate_map_from_data(self, structured_data):
        """
        生成 Plotly 地圖，包含真實道路軌跡 (Polyline)
        """
        import plotly.graph_objects as go
        from core.utils import decode_polyline  # 確保導入了解碼函數

        try:
            timeline = structured_data.get("timeline", [])
            precise_result = structured_data.get("precise_traffic_result", {})
            legs = precise_result.get("legs", [])

            if not timeline:
                return create_animated_map()

            fig = go.Figure()

            # 1. 繪製路線 (Polyline) - 藍色路徑
            # 從 legs 中提取 polyline 並解碼
            route_lats = []
            route_lons = []

            for leg in legs:
                poly_str = leg.get("polyline")
                if poly_str:
                    decoded = decode_polyline(poly_str)
                    # 解碼後是 list of (lat, lng)
                    leg_lats = [coord[0] for coord in decoded]
                    leg_lons = [coord[1] for coord in decoded]

                    # 為了讓線段連續，我們可以將其串接 (plotly 支援 nan 分隔，或直接單一 trace)
                    # 這裡簡單起見，我們將每個 leg 分開加，或者合併。合併效能較好。
                    route_lats.extend(leg_lats)
                    route_lons.extend(leg_lons)

            if route_lats:
                fig.add_trace(go.Scattermapbox(
                    lat=route_lats,
                    lon=route_lons,
                    mode='lines',
                    line=dict(width=4, color='#4A90E2'),
                    name='Route',
                    hoverinfo='none'  # 路徑本身不需 hover
                ))

            # 2. 繪製站點 (Markers)
            lats, lons, hover_texts, colors, sizes = [], [], [], [], []

            for i, stop in enumerate(timeline):
                coords = stop.get("coordinates", {})
                lat = coords.get("lat")
                lng = coords.get("lng")

                if lat and lng:
                    lats.append(lat)
                    lons.append(lng)

                    # 構建 Hover Text
                    name = stop.get("location", "Stop")
                    time = stop.get("time", "")
                    weather = stop.get("weather", "")
                    text = f"<b>{i}. {name}</b><br>🕒 {time}<br>🌤️ {weather}"
                    hover_texts.append(text)

                    # 樣式：起點綠色，終點紅色，中間藍色
                    if i == 0:
                        colors.append('#50C878')  # Start
                        sizes.append(15)
                    elif i == len(timeline) - 1:
                        colors.append('#FF6B6B')  # End
                        sizes.append(15)
                    else:
                        colors.append('#F5A623')  # Middle
                        sizes.append(12)

            fig.add_trace(go.Scattermapbox(
                lat=lats, lon=lons,
                mode='markers+text',
                marker=dict(size=sizes, color=colors, allowoverlap=True),
                text=[str(i) for i in range(len(lats))],  # 顯示序號
                textposition="top center",
                textfont=dict(size=14, color='black', family="Arial Black"),
                hovertext=hover_texts,
                hoverinfo='text',
                name='Stops'
            ))

            # 設定地圖中心與縮放
            if lats:
                center_lat = sum(lats) / len(lats)
                center_lon = sum(lons) / len(lons)
            else:
                center_lat, center_lon = 25.033, 121.565

            fig.update_layout(
                mapbox=dict(
                    style='open-street-map',  # 或 'carto-positron' 看起來更乾淨
                    center=dict(lat=center_lat, lon=center_lon),
                    zoom=12
                ),
                margin=dict(l=0, r=0, t=0, b=0),
                height=500,
                showlegend=False
            )

            return fig
        except Exception as e:
            logger.error(f"Error generating map: {e}", exc_info=True)
            return create_animated_map()

    def build_interface(self):
        with gr.Blocks(title=APP_TITLE) as demo:
            gr.HTML(get_enhanced_css())
            create_header()

            theme_btn, settings_btn, doc_btn = create_top_controls()

            # 🔥 每個 Browser Tab 都有獨立的 UserSession
            session_state = gr.State(value=UserSession().to_dict())

            with gr.Row():
                with gr.Column(scale=2, min_width=400):
                    (input_area, agent_stream_output, user_input, auto_location,
                     location_inputs, lat_input, lon_input, analyze_btn) = create_input_form(
                        create_agent_stream_output()
                    )

                    (task_confirm_area, task_summary_display,
                     task_list_display, exit_btn_inline, ready_plan_btn) = create_confirmation_area()

                    team_area, agent_displays = create_team_area(create_agent_card_enhanced)
                    (result_area, result_display, timeline_display, metrics_display) = create_result_area(
                        create_animated_map)

                with gr.Column(scale=3, min_width=500):
                    status_bar = gr.Textbox(label="📊 Status", value="Waiting for input...", interactive=False,
                                            max_lines=1)
                    (tabs, report_tab, map_tab, report_output, map_output, reasoning_output,
                     chat_input_area, chat_history_output, chat_input, chat_send) = create_tabs(
                        create_animated_map, get_reasoning_html_reversed()
                    )

            (settings_modal, google_maps_key, openweather_key, anthropic_key,
             model_choice, close_settings_btn, save_settings_btn,
             settings_status) = create_settings_modal()

            doc_modal, close_doc_btn = create_doc_modal()

            # ===== Event Handlers =====

            auto_location.change(fn=toggle_location_inputs, inputs=[auto_location], outputs=[location_inputs])

            def analyze_wrapper(ui, al, lat, lon, s):
                session = UserSession.from_dict(s)
                yield from self.step1_analyze_tasks(ui, al, lat, lon, session)

            analyze_btn.click(
                fn=analyze_wrapper,
                inputs=[user_input, auto_location, lat_input, lon_input, session_state],
                outputs=[
                    agent_stream_output, task_summary_display, task_list_display,
                    reasoning_output, task_confirm_area, chat_input_area,
                    chat_history_output, status_bar, *agent_displays, session_state
                ]
            ).then(
                fn=lambda: (gr.update(visible=False), gr.update(visible=True), gr.update(visible=True)),
                outputs=[input_area, task_confirm_area, chat_input_area]
            )

            def chat_wrapper(msg, s):
                session = UserSession.from_dict(s)
                yield from self.modify_task_chat(msg, session)

            chat_send.click(
                fn=chat_wrapper,
                inputs=[chat_input, session_state],
                outputs=[chat_history_output, task_list_display, session_state]
            ).then(fn=lambda: "", outputs=[chat_input])

            exit_btn_inline.click(
                fn=lambda: (
                    gr.update(visible=True), gr.update(visible=False),
                    gr.update(visible=False), gr.update(visible=False),
                    gr.update(visible=False), gr.update(visible=False),
                    gr.update(visible=False), "",
                    create_agent_stream_output(),
                    self._generate_chat_welcome_html(),
                    "Ready...",
                    UserSession().to_dict()
                ),
                outputs=[
                    input_area, task_confirm_area, chat_input_area, result_area,
                    team_area, report_tab, map_tab, user_input,
                    agent_stream_output, chat_history_output, status_bar, session_state
                ]
            )

            ready_plan_btn.click(
                fn=lambda: (gr.update(visible=False), gr.update(visible=False),
                            gr.update(visible=True), gr.update(selected="ai_conversation_tab")),
                outputs=[task_confirm_area, chat_input_area, team_area, tabs]
            ).then(
                fn=lambda s: self.step2_search_pois(UserSession.from_dict(s)),
                inputs=[session_state],
                outputs=[reasoning_output, status_bar, *agent_displays, session_state]
            ).then(
                fn=lambda s: self.step3_run_core_team(UserSession.from_dict(s)),
                inputs=[session_state],
                outputs=[report_output, session_state]
            ).then(
                fn=lambda: (gr.update(visible=True), gr.update(visible=True), gr.update(selected="report_tab")),
                outputs=[report_tab, map_tab, tabs]
            ).then(
                fn=lambda s: self.step4_finalize(UserSession.from_dict(s)),
                inputs=[session_state],
                outputs=[
                    timeline_display, metrics_display, result_display,
                    map_output, map_tab, team_area, status_bar, *agent_displays, session_state
                ]
            ).then(fn=lambda: gr.update(visible=True), outputs=[result_area])

            # Settings
            settings_btn.click(fn=lambda: gr.update(visible=True), outputs=[settings_modal])
            close_settings_btn.click(fn=lambda: gr.update(visible=False), outputs=[settings_modal])

            # [Security Fix] 儲存設定到 session_state
            save_settings_btn.click(
                fn=self.save_settings,
                inputs=[google_maps_key, openweather_key, anthropic_key, model_choice, session_state],
                outputs=[settings_status, session_state]
            )

            # Theme
            theme_btn.click(fn=None, js="""
                () => {
                    const container = document.querySelector('.gradio-container');
                    if (container) {
                        container.classList.toggle('theme-dark');
                        const isDark = container.classList.contains('theme-dark');
                        localStorage.setItem('lifeflow-theme', isDark ? 'dark' : 'light');
                    }
                }
            """)

            doc_btn.click(fn=lambda: gr.update(visible=True), outputs=[doc_modal])
            close_doc_btn.click(fn=lambda: gr.update(visible=False), outputs=[doc_modal])

        return demo


def main():
    app = LifeFlowAI()
    demo = app.build_interface()
    demo.launch(server_name="0.0.0.0", server_port=7860, share=True, show_error=True)


if __name__ == "__main__":
    main()