Kybernesis for OpenClaw

Infinite memory for your OpenClaw AI assistant

Overview

Kybernesis gives your OpenClaw agent persistent, intelligent memory that works across every platform—Telegram, WhatsApp, Slack, Discord, and more.

The Problem

OpenClaw's built-in memory relies on tool calls. The AI has to decide to access memory, and models don't do this consistently. Result: your AI forgets things it should know.

The Solution

Kybernesis uses auto-recall—relevant memories are automatically injected into context before every turn. No tool calls needed. Your AI just knows.

Key features:

Quick Start

1. Install the Plugin

openclaw plugins install @kybernesis/openclaw-kybernesis

If you encounter module errors after installation, install dependencies:

cd ~/.openclaw/extensions/openclaw-kybernesis
npm install --omit=dev

2. Get Your API Key

1. Go to kybernesis.ai
2. Create an account or sign in
3. Navigate to Settings → API Keys
4. Click Generate New Key
5. Copy your key (starts with kb_...)

3. Configure the Plugin

Set your API key via environment variable:

export KYBERNESIS_API_KEY=kb_your_key_here

Or configure in your OpenClaw config file (~/.openclaw/openclaw.json):

{
  "plugins": {
    "entries": {
      "openclaw-kybernesis": {
        "enabled": true,
        "config": {
          "apiKey": "kb_your_key_here"
        }
      }
    }
  }
}

4. Verify Connection

Run the status command:

/kyber_status

You should see:

🧠 Kybernesis Status
━━━━━━━━━━━━━━━━━━━━━
Connection: ✅ OK
Endpoint: https://api.kybernesis.ai

📊 Memory Stats
Total: 0 memories
├─ 🔥 Hot: 0 (0%)
├─ 🌡️ Warm: 0 (0%)
└─ 🗄️ Archive: 0 (0%)

⚙️ Settings
Auto-recall: ✅ enabled
Auto-capture: ✅ enabled

You're ready to go!

How It Works

Auto-Recall (Before Every Turn)

When you send a message:

1. Your message is used as a search query
2. Kybernesis searches your entire memory (semantic + metadata)
3. Top matches are injected into the AI's context
4. The AI responds with full knowledge of relevant memories

You don't have to ask the AI to remember—it automatically has context.

Auto-Capture (After Every Turn)

After each conversation turn:

1. Your message + AI response are combined
2. A title is generated from your message
3. The conversation is saved to your memory
4. Tags are automatically applied

Your conversations become searchable knowledge, automatically.

Sleep Agent (Background Processing)

While you're away, the Sleep Agent:

• Auto-tags new memories based on content
• Discovers relationships between memories
• Adjusts priority based on access patterns
• Summarizes long content for faster retrieval
• Tiers memories (hot → warm → archive)

Your memory gets smarter without you doing anything.

Commands

/remember — Save Something Manually

Save specific information to memory:

/remember The project deadline is March 15th tags:project,deadlines

Syntax:

/remember <content> [tags:tag1,tag2,tag3]

Examples:

/remember My preferred coding style is functional TypeScript
/remember API key for Stripe is in 1Password tags:credentials,stripe
/remember Meeting notes: decided to use PostgreSQL for the database tags:meetings,architecture

/recall — Search Manually

Search your memories:

/recall database architecture decisions limit:5

Syntax:

/recall <query> [limit:N]

Examples:

/recall project deadlines
/recall what did I decide about authentication limit:3
/recall TypeScript best practices

/kyber_status — Check Status

View connection status, memory stats, and settings:

/kyber_status

Shows:

• Connection health
• Total memories by tier (hot/warm/archive)
• Source breakdown (chat/upload/connector)
• Current plugin settings

Configuration Options

Configure via your OpenClaw config file (~/.openclaw/openclaw.json):

Example: Custom Configuration

{
  "plugins": {
    "entries": {
      "openclaw-kybernesis": {
        "enabled": true,
        "config": {
          "apiKey": "kb_xxxxx",
          "mcpEndpoint": "https://api.kybernesis.ai",
          "autoRecall": true,
          "autoCapture": true,
          "maxRecallResults": 15,
          "captureMinLength": 50,
          "captureTags": "openclaw,personal",
          "includeTierInfo": true,
          "debug": false
        }
      }
    }
  }
}

Tools (For Advanced Use)

The plugin exposes four tools your AI can use directly:

kybernesis_store

Save content to memory:

Tool: kybernesis_store
Args: {
  "content": "The user prefers dark mode in all applications",
  "title": "User Preferences",
  "tags": ["preferences", "settings"]
}

kybernesis_search

Search memories:

Tool: kybernesis_search
Args: {
  "query": "user preferences",
  "limit": 5
}

kybernesis_forget

Delete a memory:

Tool: kybernesis_forget
Args: {
  "memoryId": "mem_abc123"
}

kybernesis_context

Get memory statistics:

Tool: kybernesis_context
Args: {}

Memory Tiering Explained

Memories automatically move between tiers based on usage:

🔥 Hot Tier

• Actively used memories
• Fast retrieval priority
• Recently accessed (within 3 days)
• High relationship density

🌡️ Warm Tier

• Occasionally accessed
• Moderate priority
• Accessed within 21 days
• Has manual tags

🗄️ Archive Tier

• Rarely accessed (30+ days)
• Still searchable
• Low priority score
• Minimal connections

Tiering is automatic—you don't need to manage it. The Sleep Agent handles promotion and demotion based on your actual usage patterns.

Cross-Platform Sync

Your memory syncs across all OpenClaw platforms:

• Telegram → memories available in WhatsApp
• WhatsApp → memories available in Slack
• Slack → memories available in Discord
• Discord → memories available in Telegram

Start a conversation anywhere. Continue it anywhere else. Same context, always.

Troubleshooting

"Connection failed" error

1. Verify your API key is correct (starts with kb_)
2. Check your internet connection
3. Ensure https://api.kybernesis.ai is accessible
4. Try regenerating your API key

"Cannot find module" error after installation

Install dependencies manually:

cd ~/.openclaw/extensions/openclaw-kybernesis
npm install --omit=dev

Memories not being recalled

1. Run /kyber_status to verify auto-recall is enabled
2. Check that you have memories stored (total > 0)
3. Try /recall with a direct query to test search
4. Increase maxRecallResults in your config if needed

Conversations not being captured

1. Verify auto-capture is enabled in /kyber_status
2. Check message length meets minimum (default: 50 chars)
3. Greetings and short commands are filtered out by design

High latency

1. Check your network connection
2. Reduce maxRecallResults in your config for faster responses
3. Consider your distance from the API server

Debug mode

Enable verbose logging in your OpenClaw config:

{
  "plugins": {
    "entries": {
      "openclaw-kybernesis": {
        "config": {
          "debug": true
        }
      }
    }
  }
}

This will show detailed logs for every recall and capture operation.

Privacy & Security

• Your data stays yours: Memories are stored in your Kybernesis workspace
• API key authentication: All requests require your personal API key
• Encrypted in transit: All communication uses HTTPS
• No training on your data: Your memories are never used to train models

FAQ

Q: Does this replace OpenClaw's built-in memory?

It enhances it. OpenClaw's tool-based memory still works, but Kybernesis adds auto-recall so your AI doesn't have to "decide" to remember.

Q: How many memories can I store?

Unlimited. Kybernesis scales to thousands of memories with sub-100ms retrieval.

Q: Will this slow down my AI?

Retrieval takes <100ms. Most users don't notice any delay.

Q: Can I use this with multiple OpenClaw instances?

Yes. All instances using the same API key share the same memory.

Q: How do I delete all my data?

Contact support@kybernesis.ai or use the Kybernesis dashboard to delete your workspace.

Support

• Documentation: kybernesis.ai/docs
• Discord: Join our community
• Email: support@kybernesis.ai
• GitHub Issues: Report bugs

Changelog

v0.2.7 (Current)

• Fixed MCP API communication (SSE response parsing)
• Added proper Accept header for MCP protocol compliance

v0.2.6

• Improved auto-recall relevance scoring
• Added tier info display option
• Fixed capture title generation for questions
• Better error handling for network issues

v0.2.0

• Added Sleep Agent integration
• Memory tiering support
• Cross-platform sync

v0.1.0

• Initial release
• Basic store/search/forget tools
• Auto-recall and auto-capture hooks