back
loading skill details...
Monitors local llama.cpp servers on macOS, auto-restarts on failure, sends notifications, and supports customizable hooks for self-healing supervision.
# Rock-Solid Watchdog
A self-healing supervisor for local LLM inference servers (llama.cpp).
Monitors health, auto-restarts on failure, escalates to recovery.
## Type
Plugin — runs alongside OpenClaw gateway
## How it works
```bash
# 1. Clone / download
git clone https://github.com/gutchapa/rock-solid-watchdog.git
cd rock-solid-watchdog
# 2. Configure (optional env vars)
export LLAMA_PORT=8085
export LLAMA_SERVER=~/llama.cpp/build/bin/llama-server
# 3. Run
bash rock-solid-watchdog.sh
```
## Features
- Configurable health check interval and retry limits
- macOS notifications on failure (optional)
- Consecutive failure tracking with rate-limited escalation
- Graceful restart with full logging
- Custom hook system (pre-start, post-start, post-fail, escalate)
## Requirements
- bash 4+
- curl
- pkill/pgrep
- macOS (Linux-compatible with notify-send)
## Files
```
rock-solid-watchdog/
├── rock-solid-watchdog.sh # Main script
├── README.md # This file
└── hooks/ # Optional custom hooks
├── 00-notify-post-start.sh.example
└── 00-log-post-fail.sh.example
```
## Env vars
| Variable | Default | Description |
|----------|---------|-------------|
| `LLAMA_SERVER` | `~/llama.cpp/build/bin/llama-server` | Path to server binary |
| `LLAMA_PORT` | `8084` | Watchdog port |
| `LLAMA_HOST` | `127.0.0.1` | Bind address |
| `MAX_CONSECUTIVE_FAILS` | `5` | Auto-restart threshold |
| `ENABLE_LLAMA_WATCH` | `1` | Enable/disable |
| `LOG_FILE` | `~/.openclaw/logs/watchdog.log` | Log path |
| `HOOK_DIR` | `./hooks/` | Custom hook scripts directory |
## License
Apache 2.0
don't have the plugin yet? install it then click "run inline in claude" again.