Self-hosted open-source chat interface for llama.cpp — zero API keys, runs GGUF models entirely offline with a clean web UI. Docker one-liner.