M

mem

by @runablehqv1.0.0
4.0(4)

AI Agent 持久化记忆存储系统,支持全文搜索、本地存储和快速检索,让 AI 跨会话记住用户偏好和关键信息

memoryvector-searchagent-memoryknowledge-managementsqliteGitHub
安装方式
npx skills add runablehq/memory --skill mem
compare_arrows

Before / After 效果对比

1
使用前

AI 每次对话都需要重新提供上下文,用户偏好和项目决策无法跨会话保留,重复解释同样的问题

使用后

AI 自动存储和检索关键信息,跨会话记住用户偏好、项目决策和代码片段,大幅减少重复沟通

description SKILL.md

mem

mem — Agent Memory Store

A CLI tool for storing and retrieving memories with full-text search. Data is stored locally in ~/.mem/mem.db.

When to Use

  • Remember user preferences, project decisions, important facts

  • Store code snippets, commands, configurations for later recall

  • Search your knowledge base before asking the user for information you may have stored

  • Attach images (screenshots, diagrams) to memories

Commands

Three operators: (none) = recall, + = remember, - = forget.

Recall (search, list, get)

mem                             # list recent memories
mem "deploy"                    # full-text search
mem "database" --tag db         # search filtered by tag
mem 7sjtNVyZrNIa                # get full content by ID
mem --tag prefs                 # list filtered by tag
mem "api" --limit 5 --json     # limit results, JSON output
mem --full                      # show full content for all

Remember

mem + "user prefers dark mode" --tag prefs
mem + "deploy: bun build --compile" --tag deploy
mem + "chose SQLite for simplicity" --tag architecture
mem + --image ./screenshot.png --title "Current UI" --tag ui
echo "long content" | mem + --tag notes

Forget

mem - <id>                      # delete one memory
mem - id1 id2 id3               # delete multiple

Piping

mem "old" --json | jq -r '.[].id' | xargs -I{} mem - {}
echo "long content" | mem + --tag notes

Best Practices

  • Tag consistently — Use lowercase, descriptive tags like prefs, api, deploy, db

  • Search before asking — Check if you've stored relevant information before asking the user

  • Store decisions — When making architectural or design decisions, store the reasoning

  • Keep memories atomic — One concept per memory for better searchability

Output Formats

  • Default: One-line summary per result

  • --full: Complete content inline

  • --json: Structured JSON for parsing

Weekly Installs1.7KRepositoryrunablehq/memoryFirst SeenTodaySecurity AuditsGen Agent Trust HubPassSocketPassSnykPassInstalled onclaude-code634cursor501codex372cline288windsurf221github-copilot203

forum用户评价 (0)

发表评价

效果
易用性
文档
兼容性

暂无评价,来写第一条吧

统计数据

安装量249
评分4.0 / 5.0
版本1.0.0
更新日期2026年3月22日
对比案例1 组

用户评分

4.0(4)
5
0%
4
0%
3
0%
2
0%
1
0%

为此 Skill 评分

0.0

兼容平台

🔧Claude Code

时间线

创建2026年3月22日
最后更新2026年3月22日