mlx-serve — run any LLM on your Mac
A native Zig inference server for Apple Silicon that serves MLX and GGUF models behind OpenAI- and Anthropic-compatible APIs, alongside image, video, speech and 3D generation.
mlx-serve is a native server, written in Zig with no Python dependency, that runs large language models on Apple Silicon in either MLX format or any GGUF checkpoint from Hugging Face, including Qwen, Llama, Mistral and Gemma. It exposes both OpenAI-compatible and Anthropic-compatible HTTP APIs from the same local endpoint, so tools such as Claude Code, the OpenAI SDK, Continue, Cursor or Open WebUI can point at it without changes, and it also speaks the Ollama API for pull, chat and generate calls.
Beyond text, the same server generates images, video, music and speech, including voice cloning, and 3D models, all natively on MLX. It ships as MLX Core, a signed macOS menu-bar app with a model browser, chat, an agent mode with MCP tool calling, and a settings window for every server flag, or it can be installed headless through Homebrew or run entirely from the command line for scripting.