feat: 优化 FAQ 处理和系统稳定性

- 添加本地 FAQ 库快速路径（问候语等社交响应）
- 修复 Chatwoot 重启循环问题（PID 文件清理）
- 添加 LLM 响应缓存（Redis 缓存，提升性能）
- 添加智能推理模式（根据查询复杂度自动启用）
- 添加订单卡片消息功能（Chatwoot 富媒体）
- 增加 LLM 超时时间至 60 秒

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>

agent/agents/customer_service.py

@@ -8,6 +8,7 @@ from core.state import AgentState, ConversationState, add_tool_call, set_respons
 from core.llm import get_llm_client, Message
 from prompts import get_prompt
 from utils.logger import get_logger
+from utils.faq_library import get_faq_library
 
 logger = get_logger(__name__)
 
@@ -36,6 +37,32 @@ async def customer_service_agent(state: AgentState) -> AgentState:
     if state["tool_results"]:
         return await _generate_response_from_results(state)
 
+    # ========== FAST PATH: Check if FAQ was already matched at router ==========
+    # Router already checked FAQ and stored response if found
+    if "faq_response" in state and state["faq_response"]:
+        logger.info(
+            "Using FAQ response from router",
+            conversation_id=state["conversation_id"],
+            response_length=len(state["faq_response"])
+        )
+        return set_response(state, state["faq_response"])
+    # =========================================================================
+
+    # ========== FAST PATH: Check local FAQ library first (backup) ==========
+    # This provides instant response for common questions without API calls
+    # This is a fallback in case FAQ wasn't matched at router level
+    faq_library = get_faq_library()
+    faq_response = faq_library.find_match(state["current_message"])
+
+    if faq_response:
+        logger.info(
+            "FAQ match found, returning instant response",
+            conversation_id=state["conversation_id"],
+            response_length=len(faq_response)
+        )
+        return set_response(state, faq_response)
+    # ============================================================
+
     # Get detected language
     locale = state.get("detected_language", "en")
 

agent/agents/router.py

@@ -9,6 +9,7 @@ from core.llm import get_llm_client, Message
 from core.language_detector import get_cached_or_detect
 from prompts import get_prompt
 from utils.logger import get_logger
+from utils.faq_library import get_faq_library
 
 logger = get_logger(__name__)
 
@@ -34,6 +35,28 @@ async def classify_intent(state: AgentState) -> AgentState:
     state["state"] = ConversationState.CLASSIFYING.value
     state["step_count"] += 1
 
+    # ========== FAST PATH: Check FAQ first BEFORE calling LLM ==========
+    # This avoids slow LLM calls for common questions
+    import re
+    clean_message = re.sub(r'<[^>]+>', '', state["current_message"])
+    clean_message = ' '.join(clean_message.split())
+
+    faq_library = get_faq_library()
+    faq_response = faq_library.find_match(clean_message)
+
+    if faq_response:
+        logger.info(
+            "FAQ matched at router level, skipping LLM classification",
+            conversation_id=state["conversation_id"],
+            message=clean_message[:50]
+        )
+        # Set to customer service intent and store FAQ response
+        state["intent"] = Intent.CUSTOMER_SERVICE.value
+        state["intent_confidence"] = 1.0  # High confidence for FAQ matches
+        state["faq_response"] = faq_response  # Store FAQ response for later use
+        return state
+    # ==============================================================
+
     # Detect language
     detected_locale = get_cached_or_detect(state, state["current_message"])
     confidence = 0.85  # Default confidence for language detection