异步慢召回（deepsearch）

class AsyncMemoryManager:
    def __init__(self, actor_id: str):
        # ... 初始化 ...
        self.pending_memory: Optional[str] = None
        self.is_searching = False
        self.lock = threading.Lock()

    def start_deepsearch(self, query: str):
        # 启动后台线程进行 DeepSearch
        thread = threading.Thread(target=self._deepsearch_worker, args=(query,), daemon=True)
        thread.start()

    def _deepsearch_worker(self, query: str):
        try:
            # 执行耗时的 deepsearch
            result = self.client.deepsearch(
                query=query,
                actor_id=self.actor_id,
                summary_by_model=True
            )
            formatted = self._format_deepsearch_result(result)
            
            with self.lock:
                # 将结果存入缓存，供主线程下一轮读取
                if formatted:
                    self.pending_memory = formatted
        finally:
            self.is_searching = False



def main():
    user_id = "user_a"
    memory_manager = AsyncMemoryManager(actor_id=user_id)
    
    while True:
        user_input = input(f"\\n请输入您的问题: ").strip()
        
        # 1. 获取上一轮后台挖掘到的记忆 (如果有)
        pending_memory = memory_manager.get_and_clear_pending_memory()
        if pending_memory:
            print(f"\\n[提示] 使用上轮 DeepSearch 召回的记忆进行回复...")
        
        # 2. 立即回复用户 (带上记忆Context)
        # 此时 DeepSearch 还在后台可能没跑完，但不影响这次回复（除非有 pending）
        chat_with_crewai_streaming(
            user_message=user_input,
            memory_context=pending_memory
        )
        
        # 3. 触发新一轮的后台挖掘，为后续聊天做准备
        memory_manager.start_deepsearch(user_input)