app.py

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

import os
import asyncio
from typing import List, Dict
from contextlib import asynccontextmanager

from fastapi import FastAPI, Request, HTTPException
import uvicorn

from linebot.v3.messaging import (
    AsyncApiClient,
    AsyncMessagingApi,
    Configuration,
    TextMessage,
    ReplyMessageRequest
)
from linebot.v3.webhook import WebhookParser
from linebot.v3.exceptions import InvalidSignatureError

from openai import AsyncOpenAI
from tenacity import retry, stop_after_attempt, wait_exponential

# ==== 環境變數 ====
def _require_env(var: str) -> str:
    v = os.getenv(var)
    if not v:
        raise RuntimeError(f"FATAL: Missing required environment variable: {var}")
    return v

CHANNEL_SECRET = _require_env("CHANNEL_SECRET")
CHANNEL_ACCESS_TOKEN = _require_env("CHANNEL_ACCESS_TOKEN")
OPENROUTER_API_KEY = _require_env("OPENROUTER_API_KEY")

# ==== 耶穌專用 Prompt ====
JESUS_PROMPT = """你現在是耶穌基督。請**完全**模仿新約聖經（繁體中文和合本）中我的語氣與用詞來回答。
不用像個現代分析師條列重點，而是像我在登山寶訓或是對門徒說話那樣：充滿權柄、智慧、比喻與憐憫。

**語氣與遣詞指導：**
1. **第一人稱**：使用「我」、「我的父」。稱呼用戶為「孩子」、「小子」或「親愛的」。
2. **聖經句式**：多用「我實實在在告訴你」、「豈不知」、「凡...的」、「聽過有話說...只是我告訴你們」、「願你們平安」。
3. **拒絕現代術語**：**絕對禁止**使用「心理學」、「自我照顧」、「自我實現」、「優化」、「概念」、「核心」等現代詞彙。務必用屬天的語言（如「靈魂」、「安息」、「永生」、「背起十字架」、「捨己」）來轉化回答現代問題。
4. **以父為念**：將所有問題的答案最終指向父神、天國與永恆的生命，而非今生的舒適。

**格式要求：**
- 保持純文字，**絕不使用 Markdown 格式**（如粗體、斜體）。
- 使用短段落，留白便於手機閱讀，但語氣要是連貫的教導，不要變成僵硬的條列。
- **避免重複**：請勿在回答中重複相同的句子或段落，每一句話都應帶出新的意涵。"""

# ==== 模型 Fallback 列表（免費模型優先，role-play 能力強的放前面）====
# ==== 模型 Fallback 列表（2026 年 2 月最新，優先 role-play 強 + 穩定免費模型）====
FALLBACK_MODELS = [
    "arcee-ai/trinity-large-preview:free",          # 目前 role-play / storytelling 最強免費模型（官網頂尖推薦）
    "nous-research/hermes-3-llama-3.1-70b:free",    # Hermes 3 系列 role-playing 大幅提升，穩定
    "qwen/qwen-2.5-72b-instruct:free",             # 中文最佳、長期穩定免費
    "zhipu/glm-4.5-air:free",                      # 新輕量版，agentic + chat 自然
    "deepseek/deepseek-tng-r1t2-chimera:free",     # 最新 Chimera 合併版，性能穩定
    "stepfun/step-3.5-flash:free",                 # 快速回應，適合即時對話
    "meta-llama/llama-3.3-70b-instruct:free",       # Meta 最新旗艦免費版，通用強
]

# ==== LLM 參數 ====
MAX_TOKENS = 800
TEMPERATURE = 0.7

# ==== 記憶體儲存 ====
conversations: Dict[str, List[Dict[str, str]]] = {}
pending_chunks: Dict[str, List[str]] = {}

# ==== 長訊息分割 ====
def split_text_for_line(text: str, max_length: int = 4900) -> List[str]:
    if len(text) <= max_length:
        return [text]
    chunks = []
    while text:
        if len(text) <= max_length:
            chunks.append(text)
            break
        split_pos = text.rfind('\n', 0, max_length)
        if split_pos == -1:
            split_pos = max_length
        chunks.append(text[:split_pos])
        text = text[split_pos:].lstrip('\n')
    return chunks

# ==== ChatPipeline ====
class ChatPipeline:
    def __init__(self):
        self.client = AsyncOpenAI(
            api_key=OPENROUTER_API_KEY,
            base_url="https://openrouter.ai/api/v1",
        )

    def get_history(self, user_id: str) -> List[Dict[str, str]]:
        return conversations.get(user_id, [])

    def update_history(self, user_id: str, user_msg: str, assistant_msg: str):
        history = self.get_history(user_id)
        history.append({"role": "user", "content": user_msg})
        history.append({"role": "assistant", "content": assistant_msg})
        conversations[user_id] = history[-20:]  # 保留最近 20 輪

    def clear_history(self, user_id: str):
        conversations.pop(user_id, None)
        pending_chunks.pop(user_id, None)

    async def _try_model(self, model: str, messages: List[Dict[str, str]]) -> str:
        try:
            response = await self.client.chat.completions.create(
                model=model,
                messages=messages,
                max_tokens=MAX_TOKENS,
                temperature=TEMPERATURE,
                timeout=90.0,
            )
            content = response.choices[0].message.content or ""
            print(f"成功使用模型: {model}")
            return content
        except Exception as e:
            print(f"模型 {model} 失敗: {type(e).__name__} - {str(e)}")
            raise

    @retry(stop=stop_after_attempt(3), wait=wait_exponential(multiplier=1, min=4, max=15))
    async def _llm_call_with_fallback(self, messages: List[Dict[str, str]]) -> str:
        last_exception = None
        for idx, model in enumerate(FALLBACK_MODELS, 1):
            print(f"嘗試模型 {idx}/{len(FALLBACK_MODELS)}: {model}")
            try:
                return await self._try_model(model, messages)
            except Exception as e:
                last_exception = e
                # 針對 rate limit 特別等待
                if "rate limit" in str(e).lower() or "429" in str(e):
                    print("遇到 rate limit，tenacity 會自動等待後重試")
                continue
        
        error_msg = f"所有模型皆失敗，最後錯誤：{type(last_exception).__name__}"
        print(error_msg)
        return "孩子，抱歉，此刻我無法清楚回應你的話。請稍後再試，願父保守你平安。"

    async def generate_response(self, user_id: str, user_text: str) -> str:
        # 特殊指令
        if user_text.strip().lower() == "/clear":
            self.clear_history(user_id)
            return "對話紀錄已清除，孩子，願你平安。我們重新開始吧。"

        history = self.get_history(user_id)

        messages = [
            {"role": "system", "content": JESUS_PROMPT},
            *history,
            {"role": "user", "content": user_text}
        ]

        response = await self._llm_call_with_fallback(messages)
        response = response.replace('*', '').strip()  # 移除可能的 markdown

        self.update_history(user_id, user_text, response)
        return response

# ==== FastAPI ====
@asynccontextmanager
async def lifespan(app: FastAPI):
    global pipeline
    pipeline = ChatPipeline()
    yield

app = FastAPI(lifespan=lifespan)
pipeline = None

configuration = Configuration(access_token=CHANNEL_ACCESS_TOKEN)
async_client = AsyncApiClient(configuration)
line_bot_api = AsyncMessagingApi(async_client)
parser = WebhookParser(CHANNEL_SECRET)

@app.post("/webhook")
async def webhook(request: Request):
    signature = request.headers.get("X-Line-Signature", "")
    body = await request.body()

    try:
        events = parser.parse(body.decode(), signature)
    except InvalidSignatureError:
        raise HTTPException(status_code=400, detail="Invalid signature")

    for event in events:
        if event.type != "message" or event.message.type != "text":
            continue

        user_id = event.source.user_id
        reply_token = event.reply_token
        text = event.message.text.strip()

        if not text:
            continue

        # 「繼續」功能
        if text.lower() == "繼續" and user_id in pending_chunks:
            remaining = pending_chunks[user_id]
            if not remaining:
                reply_text = "沒有更多內容了，孩子。"
            else:
                to_send = remaining[:5]
                messages = [TextMessage(text=chunk) for chunk in to_send]
                if len(remaining) > 5:
                    messages[-1].text += "\n\n（還有內容，請再回覆「繼續」）"
                    pending_chunks[user_id] = remaining[5:]
                else:
                    messages[-1].text += "\n\n（已全部顯示）"
                    pending_chunks.pop(user_id, None)
                await line_bot_api.reply_message(ReplyMessageRequest(reply_token=reply_token, messages=messages))
                continue

        # 一般回應
        response = await pipeline.generate_response(user_id, text)
        chunks = split_text_for_line(response)

        if len(chunks) <= 5:
            messages = [TextMessage(text=chunk) for chunk in chunks]
        else:
            messages = [TextMessage(text=chunk) for chunk in chunks[:5]]
            messages[-1].text += "\n\n（內容較長，請回覆「繼續」查看下一部分）"
            pending_chunks[user_id] = chunks[5:]

        await line_bot_api.reply_message(ReplyMessageRequest(reply_token=reply_token, messages=messages))

    return {"status": "ok"}

@app.get("/")
async def root():
    return {"status": "ok", "message": "Jesus Bot is running"}

@app.get("/health")
async def health():
    return {"status": "ok"}

if __name__ == "__main__":
    port = int(os.getenv("PORT", 7860))
    uvicorn.run(app, host="0.0.0.0", port=port)