| name | start-proxy |
| description | Start or restart the LLM proxy server and all its backend services (llama-server, TTS server) using the start-proxy.sh script with nohup. Trigger on ANY user query requesting a proxy start, restart, or relaunch: 'start proxy', 'restart proxy', 'start the proxy', 'proxy is down', 'restart the proxy', 'restart it', 'restart the server', 'reboot the proxy'. |
Start/Restart Proxy Skill
Start or restart the LLM proxy server and all backend services (llama-server,
TTS server) in a single command.
Prerequisites
The proxy project must be at /home/rgardler/projects/llm/proxy with the
start-proxy.sh script present at proxy/scripts/start-proxy.sh.
Usage
Run from the proxy directory:
cd /home/rgardler/projects/llm/proxy
nohup bash scripts/start-proxy.sh --restart &>/tmp/proxy-startup.log &
Wait for startup to complete before verifying:
sleep 30
curl -s http://localhost:8000/health | python3 -m json.tool
Confirm all three services are healthy:
status: "healthy"llama_server_running: truetts_server_running: truetts_server_healthy: true
If the health check returns "degraded", check the startup log:
tail -30 /tmp/proxy-startup.log
Behaviour
--restart kills any running proxy, llama-server, and TTS processes before
starting fresh.nohup keeps the process running after the terminal session ends.- The script auto-resolves API keys from
~/.pi/agent/auth.json.
- Startup takes 30-60 seconds (model loading for llama-server + TTS server).
Errors
| Symptom | Likely Cause |
|---|
| Port conflict (8080/8081/8000) | Stale process from previous run — rerun with --restart |
| TTS server not starting | Check scripts/start-qwentts.sh path in config.yaml |
| llama-server not starting | Check the binary at path from config.yaml or in PATH |
| GPU contention | TTS and LLM share the same GPU — requests may 503 during concurrent generation |
