swarm-backend / core /memory_core.py
lk080424's picture
Upload folder using huggingface_hub
17fba62 verified
#!/usr/bin/env python3
"""
虫群智能体系统 — 统一记忆核心
合并 enhanced_memory_core / v2 / simple_memory_core 三者功能
SQLite存储 + 重要性评分 + 自动分类 + 过期清理
"""
import hashlib
import json
import logging
import os
import re
import sqlite3
import threading
import time
from collections import Counter
from datetime import datetime, timedelta
from typing import Dict, List, Optional
from core.types import MemoryCategory, MemoryRecord
logger = logging.getLogger(__name__)
# 默认数据库路径
DEFAULT_DB_PATH = "/home/admin/swarm/data/memory.db"
class MemoryCore:
"""统一记忆核心 — 单例"""
_instance = None
def __new__(cls, db_path: str = DEFAULT_DB_PATH):
if cls._instance is None:
cls._instance = super().__new__(cls)
cls._instance._db_path = db_path
cls._instance._lock = threading.Lock()
cls._instance._init_db()
cls._instance._init_categories()
# 启动后台清理
cls._instance._start_cleanup_thread()
return cls._instance
# ============================================================
# 初始化
# ============================================================
def _init_db(self):
"""初始化数据库表和索引"""
# :memory: 模式不需要创建目录
db_dir = os.path.dirname(self._db_path)
if db_dir:
os.makedirs(db_dir, exist_ok=True)
conn = sqlite3.connect(self._db_path)
c = conn.cursor()
c.execute("""
CREATE TABLE IF NOT EXISTS memories (
id INTEGER PRIMARY KEY AUTOINCREMENT,
memory_id TEXT UNIQUE,
user_id TEXT,
conversation_id TEXT,
title TEXT,
user_message TEXT,
ai_response TEXT,
category TEXT DEFAULT 'general',
importance REAL DEFAULT 0.5,
priority INTEGER DEFAULT 1,
access_count INTEGER DEFAULT 0,
tags TEXT,
created_at DATETIME DEFAULT CURRENT_TIMESTAMP,
last_accessed DATETIME,
expires_at DATETIME
)
""")
for idx in ["idx_user", "idx_category", "idx_importance", "idx_expires"]:
try:
c.execute(f"CREATE INDEX IF NOT EXISTS {idx} ON memories({idx.replace('idx_', '')})")
except Exception:
pass
conn.commit()
conn.close()
def _init_categories(self):
"""初始化分类关键词映射"""
self._category_keywords = {
MemoryCategory.GREETING: ["你好", "hello", "hi", "早上好", "晚上好", "您好"],
MemoryCategory.PERSONAL: ["名字", "年龄", "工作", "兴趣", "喜欢", "我", "我的"],
MemoryCategory.INFORMATION: ["什么", "如何", "为什么", "怎么", "解释", "说明", "介绍"],
MemoryCategory.TASK: ["做", "执行", "完成", "处理", "帮助", "需要", "请"],
MemoryCategory.CREATION: ["写", "创作", "生成", "创建", "编写", "设计", "开发"],
}
def _start_cleanup_thread(self):
"""启动后台清理线程"""
t = threading.Thread(target=self._cleanup_worker, daemon=True)
t.start()
# ============================================================
# 存储
# ============================================================
def store(self, user_id: str, conversation_id: str, title: str,
user_message: str, ai_response: str,
category: MemoryCategory = None,
custom_tags: List[str] = None,
retention_days: int = None) -> str:
"""
存储一条记忆,返回 memory_id
自动完成:分类检测、重要性评分、标签生成、过期时间
"""
# 自动分类
if category is None:
category = self._detect_category(user_message)
# 生成memory_id
memory_id = hashlib.md5(
f"{user_id}_{conversation_id}_{datetime.now().timestamp()}".encode()
).hexdigest()[:16]
# 重要性评分
importance = self._calc_importance(user_message, ai_response, category)
# 标签
tags = self._generate_tags(user_message, ai_response)
if custom_tags:
tags = list(set(tags + custom_tags))
# 过期时间
expires_at = None
if retention_days is None:
retention_days = self._default_retention(category)
if retention_days > 0:
expires_at = (datetime.now() + timedelta(days=retention_days)).isoformat()
# 写入数据库
with self._lock:
conn = sqlite3.connect(self._db_path)
c = conn.cursor()
c.execute("""
INSERT INTO memories
(memory_id, user_id, conversation_id, title,
user_message, ai_response, category, importance,
priority, tags, created_at, last_accessed, expires_at)
VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
""", (
memory_id, user_id, conversation_id, title,
user_message, ai_response, category.value, importance,
1, json.dumps(tags, ensure_ascii=False),
datetime.now().isoformat(), datetime.now().isoformat(), expires_at
))
conn.commit()
conn.close()
logger.debug(f"记忆存储: {memory_id} [{category.value}] 重要度={importance:.2f}")
return memory_id
# ============================================================
# 检索
# ============================================================
def retrieve(self, query: str, user_id: str = None,
top_k: int = 10, category: MemoryCategory = None,
min_importance: float = 0.0) -> List[Dict]:
"""检索相关记忆"""
with self._lock:
conn = sqlite3.connect(self._db_path)
c = conn.cursor()
conditions = ["expires_at IS NULL OR expires_at > ?"]
params: list = [datetime.now().isoformat()]
if user_id:
conditions.append("user_id = ?")
params.append(user_id)
if category:
conditions.append("category = ?")
params.append(category.value)
if min_importance > 0:
conditions.append("importance >= ?")
params.append(min_importance)
if query:
conditions.append("(user_message LIKE ? OR ai_response LIKE ?)")
params.extend([f"%{query}%", f"%{query}%"])
where = " AND ".join(conditions)
c.execute(f"""
SELECT memory_id, user_id, conversation_id, title,
user_message, ai_response, category, importance,
access_count, tags, created_at
FROM memories WHERE {where}
ORDER BY importance DESC, last_accessed DESC
LIMIT ?
""", params + [top_k])
rows = c.fetchall()
conn.close()
results = []
for row in rows:
results.append({
"memory_id": row[0],
"user_id": row[1],
"conversation_id": row[2],
"title": row[3],
"user_message": row[4],
"ai_response": row[5],
"category": row[6],
"importance": row[7],
"access_count": row[8],
"tags": json.loads(row[9]) if row[9] else [],
"created_at": row[10],
})
# 更新访问计数
if results:
self._increment_access([r["memory_id"] for r in results])
return results
def get_relevant_context(self, query: str, user_id: str = None,
top_k: int = 5) -> str:
"""获取与查询相关的上下文文本(供模型使用)"""
memories = self.retrieve(query, user_id=user_id, top_k=top_k)
if not memories:
return ""
parts = []
for m in memories:
parts.append(f"用户: {m['user_message']}\n助手: {m['ai_response']}")
return "\n---\n".join(parts)
# ============================================================
# 统计与维护
# ============================================================
def get_stats(self) -> Dict:
"""获取记忆统计"""
conn = sqlite3.connect(self._db_path)
c = conn.cursor()
c.execute("SELECT COUNT(*) FROM memories")
total = c.fetchone()[0]
c.execute("SELECT category, COUNT(*) FROM memories GROUP BY category")
cat_dist = dict(c.fetchall())
c.execute("SELECT AVG(importance) FROM memories")
avg_imp = c.fetchone()[0] or 0.0
conn.close()
return {
"total_memories": total,
"category_distribution": cat_dist,
"avg_importance": round(avg_imp, 3),
"db_path": self._db_path,
}
def cleanup_expired(self) -> int:
"""清理过期记忆"""
conn = sqlite3.connect(self._db_path)
c = conn.cursor()
c.execute("DELETE FROM memories WHERE expires_at <= ?",
(datetime.now().isoformat(),))
deleted = c.rowcount
conn.commit()
conn.close()
if deleted:
logger.info(f"清理过期记忆: {deleted}条")
return deleted
# ============================================================
# 内部方法
# ============================================================
def _detect_category(self, text: str) -> MemoryCategory:
"""自动检测记忆分类"""
text_lower = text.lower()
for cat, keywords in self._category_keywords.items():
if any(kw in text_lower for kw in keywords):
return cat
return MemoryCategory.GENERAL
def _calc_importance(self, user_msg: str, ai_resp: str,
category: MemoryCategory) -> float:
"""计算重要性评分"""
# 类别权重
cat_weights = {
MemoryCategory.PERSONAL: 1.0,
MemoryCategory.CREATION: 0.9,
MemoryCategory.TASK: 0.8,
MemoryCategory.INFORMATION: 0.6,
MemoryCategory.GENERAL: 0.4,
MemoryCategory.GREETING: 0.3,
}
cat_score = cat_weights.get(category, 0.4)
# 长度因素
total_len = len(user_msg) + len(ai_resp)
len_score = min(total_len / 500.0, 1.0)
# 关键词因素
key_words = ["重要", "关键", "核心", "必须", "需要"]
kw_score = min(sum(1 for w in key_words if w in user_msg) * 0.15, 0.3)
# 加权
importance = cat_score * 0.4 + len_score * 0.3 + kw_score * 0.3
return round(min(max(importance, 0.0), 1.0), 3)
def _generate_tags(self, user_msg: str, ai_resp: str) -> List[str]:
"""自动生成标签"""
tags = []
text = (user_msg + " " + ai_resp).lower()
tag_rules = {
"question": ["什么", "如何", "为什么", "怎么"],
"creation": ["写", "创作", "生成", "创建"],
"technical": ["技术", "开发", "编程", "算法", "系统"],
"business": ["商业", "市场", "营销", "客户"],
"positive": ["好", "棒", "优秀", "喜欢", "满意"],
"negative": ["不好", "糟糕", "失望", "问题", "错误"],
}
for tag, keywords in tag_rules.items():
if any(kw in text for kw in keywords):
tags.append(tag)
return list(set(tags))
def _default_retention(self, category: MemoryCategory) -> int:
"""各类别默认保留天数"""
retentions = {
MemoryCategory.GREETING: 30,
MemoryCategory.INFORMATION: 90,
MemoryCategory.TASK: 180,
MemoryCategory.CREATION: 365,
MemoryCategory.PERSONAL: 730,
MemoryCategory.GENERAL: 60,
}
return retentions.get(category, 60)
def _increment_access(self, memory_ids: List[str]):
"""更新访问计数"""
conn = sqlite3.connect(self._db_path)
c = conn.cursor()
now = datetime.now().isoformat()
for mid in memory_ids:
c.execute("UPDATE memories SET access_count = access_count + 1, last_accessed = ? WHERE memory_id = ?",
(now, mid))
conn.commit()
conn.close()
def _cleanup_worker(self):
"""后台清理线程 — 每小时检查一次"""
while True:
try:
self.cleanup_expired()
except Exception as e:
logger.error(f"记忆清理异常: {e}")
time.sleep(3600)