Self-host llamactl

llamactl is a self-hosted control plane that lets you run and manage local LLM inference servers. It handles llama.cpp, MLX on Apple Silicon, and vLLM through one web dashboard, exposing a single OpenAI-compatible API that directs requests to specific model instances. You get a built-in HuggingFace downloader for GGUF and Safetensors models, a llama.cpp router that loads models only when needed, automatic idle shutdown and least-recently-used cleanup, grouped instance limits, and the ability to run instances on remote machines. The dashboard and management API use API keys, while inference endpoints use separate keys; you can also add OpenID Connect login via Authentik, Keycloak, or Authelia. Built with Go and React, it defaults to port 8080 and uses a local SQLite database with no extra services required. Install via pre-built binaries, Docker images with GPU support, or compile from source using Go 1.24+ and Node.js 22+.