botrun-flow-lang 5.11.11__py3-none-any.whl → 5.12.261__py3-none-any.whl

botrun_flow_lang/api/hatch_api.py

@@ -286,13 +286,40 @@ async def get_hatches(
     user_id: str,
     offset: int = Query(0, ge=0),
     limit: int = Query(20, ge=1, le=100),
+    sort_by: str = Query("updated_at", description="Field to sort by (name, updated_at)"),
+    order: str = Query("desc", regex="^(asc|desc)$", description="Sort order: asc or desc"),
     current_user: CurrentUser = Depends(verify_jwt_token),
     hatch_store=Depends(get_hatch_store),
 ):
+    """Get hatches for a user with sorting options.
+
+    Args:
+        user_id: User ID to get hatches for
+        offset: Pagination offset
+        limit: Maximum number of results (1-100)
+        sort_by: Field to sort by - only 'name' or 'updated_at' are supported (default: updated_at)
+        order: Sort order - 'asc' or 'desc' (default: desc for newest first)
+
+    Returns:
+        List of hatches sorted by the specified field
+
+    Raises:
+        HTTPException: 400 if sort_by field is not supported
+    """
     # Verify user permission to access hatches for the specified user_id
     verify_user_permission(current_user, user_id)
 
-    hatches, error = await hatch_store.get_hatches(user_id, offset, limit)
+    # Validate sort_by field - only allow fields with Firestore indexes
+    allowed_sort_fields = ["name", "updated_at"]
+    if sort_by not in allowed_sort_fields:
+        raise HTTPException(
+            status_code=400,
+            detail=f"Invalid sort_by field '{sort_by}'. Allowed fields: {', '.join(allowed_sort_fields)}",
+        )
+
+    hatches, error = await hatch_store.get_hatches(
+        user_id, offset, limit, sort_by, order
+    )
     if error:
         raise HTTPException(status_code=500, detail=error)
     return hatches

botrun_flow_lang/api/line_bot_api.py

@@ -100,6 +100,11 @@ BOTRUN_FRONT_URL = os.getenv("BOTRUN_FRONT_URL", None)
 SUBSIDY_API_TOKEN = os.getenv("SUBSIDY_API_TOKEN", None)
 SUBSIDY_API_URL = os.getenv("SUBSIDY_API_URL", "https://p271-subsidy-ie7vwovclq-de.a.run.app/v1/generateContent")
 
+# BigQuery Token Logging API 相關環境變數
+BIGQUERY_TOKEN_LOG_API_URL = os.getenv("BIGQUERY_TOKEN_LOG_API_URL", "http://localhost:8002/api/v1/logs/text")
+BIGQUERY_TOKEN_LOG_ENABLED = os.getenv("BIGQUERY_TOKEN_LOG_ENABLED", "true").lower() == "true"
+SUBSIDY_LINE_BOT_MODEL_NAME = os.getenv("SUBSIDY_LINE_BOT_MODEL_NAME", "gemini-2.0-flash-thinking-exp")
+
 # 全局變數
 # 用於追蹤正在處理訊息的使用者，避免同一使用者同時發送多條訊息造成處理衝突
 _processing_users = set()
@@ -202,6 +207,111 @@ async def log_to_bigquery(
         )
 
 
+async def log_tokens_to_bigquery(
+    user_id: str,
+    display_name: str,
+    log_content: str,
+    model: str,
+    input_tokens: int | None,
+    output_tokens: int | None,
+    total_tokens: int | None,
+    request: Request,
+    session_id: str = "",
+) -> None:
+    """
+    記錄 token 使用量到 BigQuery logging API
+
+    Args:
+        user_id: LINE 使用者 ID
+        display_name: 使用者顯示名稱
+        log_content: 使用者輸入訊息
+        model: 使用的 AI 模型
+        input_tokens: 輸入 token 數量
+        output_tokens: 輸出 token 數量
+        total_tokens: 總 token 數量
+        request: FastAPI Request 物件
+        session_id: Session ID (可選，預設使用 user_id)
+    """
+    # 檢查功能是否啟用
+    if not BIGQUERY_TOKEN_LOG_ENABLED:
+        logging.debug("[Token Logger] BigQuery token logging is disabled")
+        return
+
+    start_time = time.time()
+
+    try:
+        tz = pytz.timezone("Asia/Taipei")
+        current_time = datetime.now(tz)
+
+        # 組裝 payload
+        payload = {
+            "action_details": log_content,
+            "action_type": "call_subsidy_api",
+            "botrun": "subsidy_line_bot",
+            "dataset_name": os.getenv("BOTRUN_LOG_DATASET_NAME", "subsidy_line_bot"),
+            "department": os.getenv("BOTRUN_LOG_DEPARTMENT", "subsidy_line_bot"),
+            "developer": "",
+            "domain_name": "subsidy_line_bot",
+            "input_tokens": input_tokens,
+            "model": model,
+            "output_tokens": output_tokens,
+            "resource_id": json.dumps({
+                "user_id": user_id,
+                "timestamp": current_time.strftime("%Y-%m-%dT%H:%M:%SZ")
+            }),
+            "session_id": session_id or user_id,
+            "total_tokens": total_tokens,
+            "user_agent": request.headers.get("user-agent", "Line Platform"),
+            "user_name": display_name
+        }
+
+        logging.info(
+            f"[Token Logger] Logging tokens for user {display_name} ({user_id}): "
+            f"input={input_tokens}, output={output_tokens}, total={total_tokens}"
+        )
+
+        # 使用 aiohttp 非同步呼叫 API
+        timeout = aiohttp.ClientTimeout(total=10)  # 10 秒超時
+        async with aiohttp.ClientSession(timeout=timeout) as session:
+            async with session.post(
+                BIGQUERY_TOKEN_LOG_API_URL,
+                json=payload,
+                headers={"Content-Type": "application/json"}
+            ) as response:
+                response_text = await response.text()
+
+                if response.status == 200:
+                    elapsed_time = time.time() - start_time
+                    logging.info(
+                        f"[Token Logger] Successfully logged tokens to BigQuery for user "
+                        f"{display_name} ({user_id}), elapsed time: {elapsed_time:.3f}s"
+                    )
+                else:
+                    logging.error(
+                        f"[Token Logger] Failed to log tokens, API returned status {response.status}: {response_text}"
+                    )
+
+    except asyncio.TimeoutError:
+        elapsed_time = time.time() - start_time
+        logging.error(
+            f"[Token Logger] Timeout while logging tokens for user {display_name} ({user_id}), "
+            f"elapsed time: {elapsed_time:.3f}s"
+        )
+    except aiohttp.ClientError as e:
+        elapsed_time = time.time() - start_time
+        logging.error(
+            f"[Token Logger] Network error while logging tokens for user {display_name} ({user_id}): {e}, "
+            f"elapsed time: {elapsed_time:.3f}s"
+        )
+    except Exception as e:
+        elapsed_time = time.time() - start_time
+        logging.error(
+            f"[Token Logger] Unexpected error while logging tokens for user {display_name} ({user_id}): {e}, "
+            f"elapsed time: {elapsed_time:.3f}s"
+        )
+        traceback.print_exc()
+
+
 def get_prompt_from_google_doc(tag_name: str, fallback_prompt: str = ""):
     """
     從 Google 文件中提取指定標籤的內容
@@ -612,10 +722,11 @@ async def subsidy_webhook(request: Request):
                 )
                 responses.append(response)
 
+            # NOTE: 按讚反讚功能已暫時停用（2025-12-03），日後需要可以取消註解以下程式碼
             # 處理使用者藉由按讚反讚按鈕反饋的postback事件
-            elif isinstance(event, PostbackEvent):
-                await handle_feedback(event, line_bot_api)
-                responses.append("feedback_handled")
+            # elif isinstance(event, PostbackEvent):
+            #     await handle_feedback(event, line_bot_api)
+            #     responses.append("feedback_handled")
 
     return {"responses": responses}
 
@@ -736,13 +847,37 @@ async def handle_message(
     _processing_users.add(user_id)
 
     try:
-        reply_text, related_questions = await get_reply_text(
+        reply_text, related_questions, usage_metadata = await get_reply_text(
             user_message, user_id, display_name, request
         )
         logging.info(
             f"[Line Bot Webhook: handle_message] Total response length: {len(reply_text)}"
         )
 
+        # 記錄 token 使用量到 BigQuery (非阻塞式)
+        if usage_metadata:
+            # 把 user_message 跟 reply_text 合併成 json 格式紀錄
+            log_content = json.dumps(
+                {
+                    "user_message": user_message,
+                    "reply_text": reply_text
+                },
+                ensure_ascii=False
+            )
+            asyncio.create_task(
+                log_tokens_to_bigquery(
+                    user_id=user_id,
+                    display_name=display_name,
+                    log_content=log_content,
+                    model=usage_metadata.get("model", SUBSIDY_LINE_BOT_MODEL_NAME),
+                    input_tokens=usage_metadata.get("promptTokenCount", None),
+                    output_tokens=usage_metadata.get("candidatesTokenCount", None),
+                    total_tokens=usage_metadata.get("totalTokenCount", None),
+                    request=request,
+                    session_id=user_id,
+                )
+            )
+
         # 將長訊息分段，每段不超過 LINE_MAX_MESSAGE_LENGTH
         message_chunks = []
         remaining_text = reply_text
@@ -829,34 +964,40 @@ async def handle_message(
                 )
             )
 
+        # NOTE: 按讚反讚功能已暫時停用（2025-12-03），日後需要可以取消註解以下程式碼
         # 以 Quick Reply 作為按讚反讚按鈕
-        quick_reply = QuickReply(
-            items=[
-                QuickReplyItem(
-                    action=PostbackAction(
-                        label="津好康，真是棒👍🏻",
-                        data="實用",
-                        display_text="津好康，真是棒👍🏻",
-                    )
-                ),
-                QuickReplyItem(
-                    action=PostbackAction(
-                        label="津可惜，不太實用😖",
-                        data="不實用",
-                        display_text="津可惜，不太實用😖",
-                    )
-                ),
-            ]
-        )
+        # quick_reply = QuickReply(
+        #     items=[
+        #         QuickReplyItem(
+        #             action=PostbackAction(
+        #                 label="津好康，真是棒👍🏻",
+        #                 data="實用",
+        #                 display_text="津好康，真是棒👍🏻",
+        #             )
+        #         ),
+        #         QuickReplyItem(
+        #             action=PostbackAction(
+        #                 label="津可惜，不太實用😖",
+        #                 data="不實用",
+        #                 display_text="津可惜，不太實用😖",
+        #             )
+        #         ),
+        #     ]
+        # )
 
         if question_bubble:
             messages.append(FlexMessage(alt_text="相關問題", contents=question_bubble))
 
-        messages[-1].quick_reply = quick_reply
-
+        # messages[-1].quick_reply = quick_reply
+        logging.info(
+            f"[Line Bot Webhook: handle_message] start reply_message"
+        )
         await line_bot_api.reply_message(
             ReplyMessageRequest(reply_token=event.reply_token, messages=messages)
         )
+        logging.info(
+            f"[Line Bot Webhook: handle_message] end reply_message"
+        )
     except Exception as e:
         traceback.print_exc()
         logging.error(
@@ -934,7 +1075,7 @@ async def get_reply_text(
     user_id: str,
     display_name: str,
     request: Request,
-) -> tuple[str, list]:
+) -> tuple[str, list, dict]:
     """
     使用外部 API 處理使用者訊息並回傳回覆內容
 
@@ -945,20 +1086,30 @@ async def get_reply_text(
         request (Request): FastAPI request 物件，用於記錄到 BigQuery
 
     Returns:
-        tuple[str, list]: 包含回覆訊息和相關問題的元組
+        tuple[str, list, dict]: 包含回覆訊息、相關問題和 token 使用量的元組
     """
     start_time = time.time()
 
     try:
         # 取得系統指令
-        system_instruction = get_subsidy_api_system_prompt()
+        # 暫時不需要，因為現在是直接呼叫 cbh 的 api, prompt 在它裡面
+        # system_instruction = get_subsidy_api_system_prompt()
 
         # 調用外部 API
         api_response = await call_subsidy_api(
             user_message=line_user_message,
             user_id=user_id,
             display_name=display_name,
-            system_instruction=system_instruction
+            # system_instruction=system_instruction
+        )
+
+        # 提取 token 使用量資訊
+        usage_metadata = api_response.get("usageMetadata", {})
+        logging.info(
+            f"[Line Bot Webhook: get_reply_text] Token usage: "
+            f"input={usage_metadata.get('promptTokenCount', 0)}, "
+            f"output={usage_metadata.get('candidatesTokenCount', 0)}, "
+            f"total={usage_metadata.get('totalTokenCount', 0)}"
         )
 
         # 從 API 回應中提取文字內容
@@ -1003,7 +1154,7 @@ async def get_reply_text(
             f"[Line Bot Webhook: get_reply_text] total took {time.time() - start_time:.3f}s"
         )
 
-        return full_response, related_questions
+        return full_response, related_questions, usage_metadata
 
     except Exception as e:
         import traceback
@@ -1012,7 +1163,7 @@ async def get_reply_text(
 
         # 返回錯誤訊息
         error_message = "抱歉，處理您的訊息時遇到問題，請稍後再試。"
-        return error_message, []
+        return error_message, [], {}
 
 
 async def handle_feedback(

botrun_flow_lang/langgraph_agents/agents/agent_runner.py

@@ -71,15 +71,19 @@ async def langgraph_runner(
             # 設定新的 recursion_limit 為 (multiplier + 1) * MAX_RECURSION_LIMIT
             config["recursion_limit"] = (multiplier + 1) * MAX_RECURSION_LIMIT
 
-    async for event in graph.astream_events(
-        invoke_state,
-        config,
-        version="v2",
-    ):
-        # state = await graph.aget_state(config)
-        # print(state.config)
-
-        yield event
+    try:
+        async for event in graph.astream_events(
+                invoke_state,
+                config,
+                version="v2",
+        ):
+            yield event
+    except Exception as e:
+        # 捕獲 SSE 流讀取錯誤（如 httpcore.ReadError）
+        import logging
+        logging.error(f"Error reading SSE stream: {e}", exc_info=True)
+        # 產生錯誤 event 讓調用者知道
+        yield {"error": f"SSE stream error: {str(e)}"}
 
 
 # graph 是 CompiledStateGraph，不傳入型別的原因是，loading import 需要 0.5秒

botrun_flow_lang/langgraph_agents/agents/langgraph_react_agent.py

@@ -142,11 +142,42 @@ def get_react_agent_model_name(model_name: str = ""):
 
 ANTHROPIC_MAX_TOKENS = 64000
 GEMINI_MAX_TOKENS = 32000
+TAIDE_MAX_TOKENS = 8192
 
 
 def get_react_agent_model(model_name: str = ""):
     final_model_name = get_react_agent_model_name(model_name).strip()
 
+    # 處理 taide/ 前綴的模型
+    if final_model_name.startswith("taide/"):
+        taide_api_key = os.getenv("TAIDE_API_KEY", "")
+        taide_base_url = os.getenv("TAIDE_BASE_URL", "")
+
+        if not taide_api_key or not taide_base_url:
+            raise ValueError(
+                f"Model name starts with 'taide/' but TAIDE_API_KEY or TAIDE_BASE_URL not set. "
+                f"Both environment variables are required for: {final_model_name}"
+            )
+
+        # 取得 taide/ 後面的模型名稱
+        taide_model_name = final_model_name[len("taide/"):]
+
+        if not taide_model_name:
+            raise ValueError(
+                f"Invalid taide model format: {final_model_name}. "
+                "Expected format: taide/<model_name>"
+            )
+
+        model = ChatOpenAI(
+            openai_api_key=taide_api_key,
+            openai_api_base=taide_base_url,
+            model_name=taide_model_name,
+            temperature=0,
+            max_tokens=TAIDE_MAX_TOKENS,
+        )
+        logger.info(f"model ChatOpenAI (TAIDE) {taide_model_name} @ {taide_base_url}")
+        return model
+
     # 處理 vertexai/ 前綴的模型
     if final_model_name.startswith("vertex-ai/"):
         vertex_project = os.getenv("VERTEX_AI_LANGCHAIN_PROJECT", "")