feat: enhance long-term memory with structured chat records

astrbot/builtin_stars/astrbot/long_term_memory.py

@@ -2,6 +2,8 @@
 import random
 import uuid
 from collections import defaultdict
+from dataclasses import dataclass, field
+from typing import Literal
 
 from astrbot import logger
 from astrbot.api import star
@@ -11,9 +13,36 @@
 from astrbot.api.provider import LLMResponse, Provider, ProviderRequest
 from astrbot.core.astrbot_config_mgr import AstrBotConfigManager
 
-"""
-聊天记忆增强
-"""
+
+@dataclass
+class ChatRecord:
+    """单条聊天记录，用于长期记忆存储。"""
+
+    msg_id: str
+    """消息唯一标识（用户消息取 message_obj.message_id，AI 回复取 'ai:<uuid>'）"""
+    role: Literal["user", "assistant"]
+    """角色：user 表示用户消息，assistant 表示 AI 回复"""
+    text: str
+    """格式化后的文本，如 '[昵称/HH:MM:SS]: ...' 或 '[You/HH:MM:SS]: ...'"""
+    created_at: str = field(default_factory=lambda: datetime.datetime.now().isoformat())
+    """创建时间（ISO 格式），用于调试/扩展"""
+
+
+def _get_event_msg_id(event: AstrMessageEvent) -> str:
+    """
+    获取当前事件对应的消息 ID。
+    以保证同一事件链路（handle_message -> on_req_llm -> after_req_llm）使用同一 ID。
+    """
+    msg_id = getattr(event.message_obj, "message_id", None)
+    if msg_id:
+        return str(msg_id)
+    # fallback: 使用 extra 缓存
+    cached = event.get_extra("_ltm_msg_id")
+    if cached:
+        return cached
+    generated = f"ltm:{uuid.uuid4().hex}"
+    event.set_extra("_ltm_msg_id", generated)
+    return generated
 
 
 class LongTermMemory:
@@ -144,7 +173,11 @@ async def handle_message(self, event: AstrMessageEvent):
 
             final_message = "".join(parts)
             logger.debug(f"ltm | {event.unified_msg_origin} | {final_message}")
-            self.session_chats[event.unified_msg_origin].append(final_message)
+
+            msg_id = _get_event_msg_id(event)
+            record = ChatRecord(msg_id=msg_id, role="user", text=final_message)
+            self.session_chats[event.unified_msg_origin].append(record)
+
             if len(self.session_chats[event.unified_msg_origin]) > cfg["max_cnt"]:
                 self.session_chats[event.unified_msg_origin].pop(0)
 
@@ -153,7 +186,14 @@ async def on_req_llm(self, event: AstrMessageEvent, req: ProviderRequest):
         if event.unified_msg_origin not in self.session_chats:
             return
 
-        chats_str = "\n---\n".join(self.session_chats[event.unified_msg_origin])
+        current_msg_id = _get_event_msg_id(event)
+
+        # 构造历史字符串时按 msg_id 过滤当前轮消息
+        history_records: list[ChatRecord] = self.session_chats[event.unified_msg_origin]
+        filtered_texts = [
+            rec.text for rec in history_records if rec.msg_id != current_msg_id
+        ]
+        chats_str = "\n---\n".join(filtered_texts)
 
         cfg = self.cfg(event)
         if cfg["enable_active_reply"]:
@@ -180,7 +220,11 @@ async def after_req_llm(self, event: AstrMessageEvent, llm_resp: LLMResponse):
             logger.debug(
                 f"Recorded AI response: {event.unified_msg_origin} | {final_message}"
             )
-            self.session_chats[event.unified_msg_origin].append(final_message)
+
+            ai_msg_id = f"ai:{uuid.uuid4().hex}"
+            record = ChatRecord(msg_id=ai_msg_id, role="assistant", text=final_message)
+            self.session_chats[event.unified_msg_origin].append(record)
+
             cfg = self.cfg(event)
             if len(self.session_chats[event.unified_msg_origin]) > cfg["max_cnt"]:
                 self.session_chats[event.unified_msg_origin].pop(0)