Skip to main content
Glama
cyhano

mem

by cyhano
OverviewSchemaRelated ServersScoreDiscussions
TypeScript
Local

mem

本地中文语义记忆工具 —— 给 AI coding agent（ZCode / 任意支持 MCP 的客户端）提供「按需召回、不全量注入」的事实型长期记忆。

为什么用它

  • 中文语义召回好：embedding 走硅基流动 BAAI/bge-m3（1024 维），比本地 MiniLM 对中文友好得多（实测"换说法"查询相似度 0.71 vs 0.42）。

  • 不全量注入：不像 MEMORY.md 每轮全量塞进上下文，而是 agent 用到时才 mem_search，只把命中的几条注入。

  • 自动召唤（hook）：会话开始注入概览、每次提问自动召回、回合结束提醒沉淀 —— 三个 hook 形成闭环。

  • 能整理：支持 update（改过时内容）、merge（合并重复）、link（Obsidian 式双链）。

  • 零运维：单 SQLite 文件 + 单入口脚本，无外部服务。

Related MCP server: Memory Palace

安装

cd ~/code
git clone https://github.com/cyhano/mem.git
cd mem
pnpm install
pnpm build

需要 Node.js ≥ 18（推荐 20+）和 pnpm。

配置

写 ~/.config/mem/config.json（或设环境变量 MEM_CONFIG_PATH 指向它）：

{
  "dbPath": "/Users/you/.local/share/mem/knowledge.db",
  "embedding": {
    "provider": "siliconflow",
    "baseUrl": "https://api.siliconflow.cn/v1",
    "apiKey": "你的硅基流动 key（或环境变量 SILICONFLOW_API_KEY）",
    "model": "BAAI/bge-m3",
    "dim": 1024
  }
}

apiKey 也支持通过环境变量 SILICONFLOW_API_KEY 注入，避免明文落盘。

CLI 用法

node dist/cli.js save "标题" "正文" [--kind fact|decision|config|command] [--project 项目名]
node dist/cli.js search "自然语言查询" [--top 5] [--project 项目名]
node dist/cli.js list [--project 项目名]
node dist/cli.js count
node dist/cli.js delete <id>

# 整理
node dist/cli.js update <id> [--title xx] [--content xx] [--kind xx] [--project xx|全局]
node dist/cli.js merge <目标id> <源id>          # 源并入目标后删除源

# 链接（Obsidian 式单向链 + 反向链接）
node dist/cli.js link <A的id> <B的id>          # A → B
node dist/cli.js unlink <A的id> <B的id>
node dist/cli.js links <id>                    # 看出链 + 反链

project 语义：不传 = 全局记忆（任何项目都能搜到）；显式传 --project = 归属某项目。全局单库，search/list 可加 --project 过滤。

接入 ZCode

MCP

在 ~/.zcode/cli/config.json 的 mcp.servers 里加：

{
  "mcp": {
    "servers": {
      "mem": {
        "type": "stdio",
        "command": "/绝对路径/node",
        "args": ["/绝对路径/mem/dist/mcp.js"]
      }
    }
  }
}

提供四个工具：mem_save、mem_search、mem_list、mem_delete（以及 mem_update、mem_merge）。

hook（自动召回/沉淀）

在 ~/.zcode/cli/config.json 的 hooks.events 里注册（三选一或全要）：

{
  "hooks": {
    "enabled": true,
    "events": {
      "SessionStart": [
        { "hooks": [{ "type": "process", "command": "/绝对路径/node",
          "args": ["/绝对路径/mem/hooks/auto-status.mjs"], "timeoutMs": 15000 }] }
      ],
      "UserPromptSubmit": [
        { "hooks": [{ "type": "process", "command": "/绝对路径/node",
          "args": ["/绝对路径/mem/hooks/auto-recall.mjs"], "timeoutMs": 20000 }] }
      ],
      "Stop": [
        { "hooks": [{ "type": "command",
          "command": "python3 /绝对路径/mem/hooks/auto-capture.py", "timeout": 10 }] }
      ]
    }
  }
}

注意：hook 的 matcher 字段不要写成空字符串 ""，要么省略、要么给 ≥1 字符 —— 空串会导致整个配置解析失败。

三个 hook 的作用：

事件

脚本

作用

SessionStart

auto-status.mjs

会话开始注入记忆概览（总数 + 最近标题）

UserPromptSubmit

auto-recall.mjs

每次提问自动召回相关记忆注入上下文

Stop

auto-capture.py

回合结束检测运维/配置类产出，引导存记忆

架构

src/
├── config.ts    # 配置加载 + 默认值
├── embed.ts     # embedding（硅基流动 /v1/embeddings）+ 余弦相似度
├── store.ts     # SQLite 存储（memories 表 + links 表）
├── core.ts      # save/search/update/merge/link 核心逻辑
├── server.ts    # MCP server（stdio）
├── cli.ts       # 命令行
└── hook-*.ts    # hook 专用输出端
hooks/           # 三个 hook 脚本

  • 存储：better-sqlite3 单文件，memories 表存正文 + blob 向量，links 表存单向链接。

  • 召回：查询向量化后与所有记忆暴力算余弦相似度，取 topK（<1 万条时足够快）。

  • 链接：单向存（from → to），反链自动算，对齐 Obsidian 双链语义；删记忆级联清理链接。

许可证

MIT

Tool Schema Changelog

Recent tool additions, removals, and schema changes observed during successful MCP inspections. Dates show when Glama detected each change.

No tool schema history has been recorded yet.

This server cannot be installed

RayrunBuild, deploy, and operate MCP servers without becoming an infrastructure team.ray.runAd

Maintenance

ActivityMaintained
ResponsivenessSyncing

Resources

Unclaimed servers have limited discoverability.

Looking for Admin?

If you are the server author, to access and configure the admin panel.

Related MCP Connectors

Related MCP Servers

  • Memory MCP

    burgessadrien
    F
    license
    Not graded
    quality
    D
    maintenance
    Provides persistent AI agent memory using a local vector database for long-term semantic storage and short-term session scratchpads. It enables low-latency memory operations including search, storage, and bulk management without external cloud dependencies.
    -

  • Memory Palace

    jeffpierce
    A
    license
    Not graded
    quality
    D
    maintenance
    Persistent semantic memory for AI agents, enabling storage, semantic search, knowledge graph connections, and inter-instance messaging across conversations using local models via Ollama.
    46
    MIT

  • memocean-mcp

    ChannelLabAI
    A
    license
    Not graded
    quality
    B
    maintenance
    Local-first AI Memory MCP server for AI agents to search and manage knowledge stored in Obsidian vaults, optimized for CJK languages with zero AI dependency.
    2
    MIT

  • context-vault

    fellanH
    F
    license
    Not graded
    quality
    B
    maintenance
    Persistent memory for AI agents enabling saving, searching, and managing knowledge across sessions with local markdown files.
    2
    -

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/cyhano/mem'

If you have feedback or need assistance with the MCP directory API, please join our Discord server