🦞 openclaw-watchdog

A standalone watchdog service for OpenClaw that monitors the gateway, auto-recovers from crashes, and sends Discord alerts when things go sideways.

What It Does

Health monitoring — Polls the gateway health endpoint at a configurable interval
Auto-restart — Restarts the gateway on first failure detection
Config rollback — Reverts to last-known-good config if a simple restart doesn't work
Discord SOS alerts — Sends a detailed alert with SSH instructions when all recovery fails
Recovery notifications — Lets you know when the gateway comes back online
No spam — Only alerts once per incident

Recovery Flow

Health check fails
       │
       ▼
  ┌─────────────┐
  │ Stage 1:    │──→ Simple restart
  │ Restart     │
  └──────┬──────┘
         │ still failing
         ▼
  ┌─────────────┐
  │ Stage 2:    │──→ Revert config to last-known-good + restart
  │ Rollback    │
  └──────┬──────┘
         │ still failing
         ▼
  ┌─────────────┐
  │ Stage 3:    │──→ Discord SOS with hostname & SSH instructions
  │ SOS Alert   │
  └─────────────┘

When the gateway recovers at any point, a ✅ recovery message is sent.

Prerequisites

OpenClaw installed with openclaw CLI available
Linux with systemd
curl and jq
Discord bot token (optional, for alerts)

Quick Install

git clone https://github.com/jlgrimes/openclaw-watchdog.git
cd openclaw-watchdog
chmod +x setup.sh watchdog.sh
./setup.sh

The installer will:

Copy watchdog.sh to ~/.openclaw/
Create a config file at ~/.openclaw/watchdog.env
Set up and start a systemd service

Configuration

All settings live in ~/.openclaw/watchdog.env (created by the installer):

Variable	Default	Description
`HEALTH_URL`	`http://localhost:3000/health`	Gateway health endpoint
`CHECK_INTERVAL`	`30`	Seconds between health checks
`FAIL_THRESHOLD`	`3`	Consecutive failures before SOS
`OPENCLAW_CONFIG_PATH`	`~/.openclaw/config.yaml`	Path to OpenClaw config
`DISCORD_CHANNEL_ID`	(none)	Discord channel for alerts
`DISCORD_BOT_TOKEN`	(none)	Discord bot token for alerts

After editing, restart the service:

sudo systemctl restart openclaw-watchdog

Manual Usage

You can also run the watchdog directly:

export DISCORD_BOT_TOKEN="your-token"
export DISCORD_CHANNEL_ID="your-channel-id"
./watchdog.sh

Managing the Service

sudo systemctl status openclaw-watchdog    # Check status
sudo systemctl restart openclaw-watchdog   # Restart
sudo systemctl stop openclaw-watchdog      # Stop
tail -f ~/.openclaw/watchdog.log           # View logs

License

MIT

Credits

Built by Jared Grimes and Claw 🦞 (an OpenClaw AI assistant)

3.1 KiB Raw Permalink Blame History