openclaw-memory-auto

Automatic memory archiving for OpenClaw. Works for everyone.

Features

• Zero Configuration: Works out of the box with smart defaults
• Cross-Platform: Pure TypeScript, runs on Windows/Mac/Linux
• Universal: Language-agnostic keywords, customizable to your workflow
• Non-Intrusive: Runs on startup or cron, no manual work needed
• Respectful: Only archives what matters, not every chat

Installation

npm install openclaw-memory-auto

Quick Start

Add to your openclaw.config.js:

import memoryAuto from 'openclaw-memory-auto';

export default {
  plugins: [memoryAuto]
};

That's it! The plugin will:

• ✅ On every agent startup, archive yesterday's chat
• ✅ Highlight key moments using smart keyword matching
• ✅ Store daily logs in memory/YYYY-MM-DD.md
• ✅ Prepare prompts for AI-assisted MEMORY.md updates

Configuration (Optional)

Customize behavior in openclaw.config.js:

import memoryAuto from 'openclaw-memory-auto';

export default {
  plugins: [memoryAuto],
  
  memoryAuto: {
    // Add your own keywords (case-insensitive)
    keywords: [
      '任务', '部署', 'API', 'token', 'OpenClaw',
      'task', 'deploy', 'config', 'password', 'secret'
    ],
    
    // Custom paths (relative to workspace)
    paths: {
      memoryDir: 'memory',
      logsDir: 'logs'
    },
    
    // Schedule settings
    schedule: {
      archiveHour: 9,      // 9 AM local time
      checkOnStartup: true // always check on startup
    },
    
    // Enable AI refinement (beta)
    refine: {
      enabled: true,
      model: 'stepfun/step-3.5-flash:free',
      prompt: `Extract long-term memories from today's log.`
    }
  }
};

How It Works

1. On Startup: Plugin checks if yesterday's daily log exists
2. If Missing: Scans all transcript files (agents/*/sessions/*.jsonl)
3. Time Filter: Extracts only yesterday's messages (local timezone aware)
4. Highlighting: Matches messages against your keyword list
5. Archive: Writes memory/YYYY-MM-DD.md with summary and highlights
6. Refine (optional): Calls AI to update MEMORY.md with structured memory

File Structure

After running:

workspace/
├── memory/
│   ├── 2026-03-11.md      # Auto-generated daily log
│   └── .2026-03-11.archived  # Marker (prevents re-archive)
├── logs/
│   ├── memory-archive.log
│   └── last_refine_prompt.txt  # For AI refinement
└── MEMORY.md              # Your long-term memory (manually or auto-updated)

Custom Keywords

Add to your config:

memoryAuto: {
  keywords: [
    // Defaults already included - just add yours
    '项目名', '客户', '截止日期',
    'project', 'client', 'deadline'
  ]
}

Keywords are case-insensitive and support Unicode (中文, emoji, etc.).

AI Refinement

Enable in config to automatically update MEMORY.md:

memoryAuto: {
  refine: {
    enabled: true,
    model: 'stepfun/step-3.5-turbo:free',
    prompt: `Extract: skills learned, project milestones, user preferences, important data, inside jokes.`
  }
}

The AI will append structured sections to MEMORY.md daily.

FAQ

Q: Does it archive everything?
A: No. Only messages containing your keywords (default: 30+ common terms). This keeps logs meaningful and privacy-friendly.

Q: Where are transcripts stored?
A: OpenClaw automatically saves sessions to agents/*/sessions/*.jsonl. The plugin reads these.

Q: Can I run on a schedule instead of startup?
A: Yes. Use OpenClaw's heartbeat cron:

agents: {
  defaults: {
    heartbeat: {
      tasks: [{
        name: 'memory-auto-archive',
        cron: '0 9 * * *', // daily 9 AM
        command: 'npx openclaw-memory-auto --archive'
      }]
    }
  }
}

Q: I have multiple agents. Will it work?
A: Yes. Scans all agents/*/sessions/ directories.

Development

git clone https://github.com/openclaw-plugins/memory-auto.git
cd memory-auto
npm install
npm run build

License

MIT - Free for personal and commercial use.

Contributing

PRs welcome! Please read CONTRIBUTING.md first.

Made with 🦞 by the OpenClaw Community