Reproducible single-box local-LLM stack for an RTX 4090 (GamingPC), served OpenAI-compatible over Tailscale - View it on GitHub