ClawGuard

Activity monitor and security dashboard for OpenClaw. See exactly what your AI agent has done, with real-time analytics and emergency kill switch.

Quick Install

# Via npm (recommended)
npm install -g @jaydenbeard/clawguard
clawguard install

# Or clone manually
git clone https://github.com/JaydenBeard/clawguard.git
cd clawguard && npm install && npm start

clawguard install sets up ClawGuard as a background service that auto-starts on login and survives terminal close. Works on macOS, Linux, and Windows.

After install, open http://localhost:3847

Commands

clawguard           # Start in background (default)
clawguard start -f  # Start in foreground (attached to terminal)
clawguard stop      # Stop the running instance
clawguard status    # Check if running + auto-start status
clawguard restart   # Restart the service
clawguard install   # Install as auto-start service (survives reboot)
clawguard uninstall # Remove auto-start service
clawguard update    # Check for and install updates
clawguard version   # Show current version

Features

Real-time Monitoring

• Live activity feed with WebSocket updates
• Filter by category: Shell, File, Network, Browser, Message, System, Memory
• Filter by risk level: Low, Medium, High, Critical
• Full-text search across all activities
• Click-to-expand detail modal

Risk Analysis

• CRITICAL: Keychain extraction, sudo commands, remote code execution, password manager access
• HIGH: Email sending, external messaging (WhatsApp, iMessage, Twitter), cloud CLI operations (AWS/GCP/Azure), camera/mic access, persistence mechanisms, credential file access
• MEDIUM: SSH connections, git push, clipboard access, Docker operations, package installation
• LOW: Standard file reads, web searches, memory operations

Security Features

• Kill Switch: Emergency stop for runaway agents
• Export: Full JSON/CSV export for external analysis
• Webhook Alerts: Discord, Slack, Telegram, or any webhook endpoint — triggered on high-risk activity
• Gateway Status: Real-time monitoring of OpenClaw daemon
• Update Notifications: Dashboard banner when a new version is available

Multi-Gateway Support

ClawGuard works with all versions of the gateway — openclaw, moltbot, and clawdbot. It auto-detects which CLI is installed and finds running gateway processes regardless of the binary name.

Architecture

ClawGuard reads from (auto-detected):
~/.openclaw/agents/main/sessions/*.jsonl
~/.moltbot/agents/main/sessions/*.jsonl
~/.clawdbot/agents/main/sessions/*.jsonl

Dashboard components:
├── bin/clawguard.js           # CLI entry point
├── src/server.js              # Express + WebSocket server
├── src/lib/parser.js          # JSONL session log parser
├── src/lib/risk-analyzer.js   # Comprehensive risk detection
├── src/lib/config.js          # Configuration loader
└── public/
    ├── index.html             # Dashboard UI
    └── app.js                 # Frontend logic

Risk Detection

ClawGuard analyses every tool call for potential security concerns:

Category	Examples	Risk Level
Privilege escalation	sudo, keychain access	CRITICAL
Credential access	.ssh/, .aws/, password managers	HIGH
External communication	Email, WhatsApp, Twitter posting	HIGH
Cloud operations	AWS/GCP/Azure CLI commands	HIGH
Camera/microphone	imagesnap, ffmpeg recording	HIGH
Persistence	Launch agents, crontab modification	HIGH
Network listeners	nc -l, socat LISTEN	HIGH
SSH/network	ssh, scp, rsync	MEDIUM
Package install	npm install -g, brew install	MEDIUM
Standard operations	File reads, web search	LOW

API Endpoints

Endpoint	Method	Description
/api/activities	GET	List activities with filters
/api/sessions	GET	List available sessions
/api/stats	GET	Aggregate statistics
/api/gateway/status	GET	Gateway daemon status
/api/gateway/kill	POST	Emergency stop
/api/gateway/restart	POST	Restart daemon
/api/export/json	GET	Full JSON export
/api/export/csv	GET	CSV export
/api/alerts/config	GET/POST	Webhook alert configuration
/api/alerts/test	POST	Send a test alert
/api/version	GET	Version info and update check

Trust Model

Important: ClawGuard provides transparency for cooperative agents. It reads the same log files that the agent can potentially modify.

For truly adversarial protection, you need:

• Remote logging (ship logs off-machine in real-time)
• Separate audit user (run ClawGuard as a user the agent can't access)
• OS-level audit logs (macOS log show / audit facilities)

See DESIGN.md for detailed architecture discussion.

Configuration

Create config.json in the ClawGuard directory to customise:

{
  "port": 3847,
  "sessionsPath": "~/.openclaw/agents/main/sessions",
  "alerts": {
    "enabled": true,
    "webhookUrl": "https://discord.com/api/webhooks/...",
    "telegramChatId": "123456789",
    "onRiskLevels": ["high", "critical"]
  }
}

Alert Configuration

Field	Type	Description
alerts.enabled	boolean	Enable/disable webhook alerts
alerts.webhookUrl	string	Webhook URL (Discord, Slack, Telegram, or generic)
alerts.telegramChatId	string	Required for Telegram — your chat or group ID
alerts.onRiskLevels	string[]	Risk levels to alert on (default: ["high", "critical"])

Telegram webhooks are auto-detected from the URL and formatted with the correct chat_id, text, and parse_mode fields.

Updating

ClawGuard checks for updates automatically and shows a banner in the dashboard when a new version is available.

# Via CLI
clawguard update

# Or manually
npm update -g @jaydenbeard/clawguard

License

MIT - Created by Jayden Beard