What it does

Rapid-MLX is an open-source (Apache 2.0) server that runs language models locally on Apple Silicon, built on MLX. It exposes OpenAI-style and Anthropic Messages endpoints, so a client that accepts an OpenAI- or Anthropic-compatible endpoint can usually use it without an adapter, and it is aimed at reliable tool calling for coding agents. The quick start is a few commands: rapid-mlx chat opens a terminal chat, rapid-mlx serve starts a local HTTP server, and rapid-mlx launch can point coding agents such as Claude Code at it. Installation is through Homebrew or a guided installer script, and a one-click Rapid-MLX Desktop app for M-series Macs bundles the same engine. Beyond text, the project documents local vision models, image generation and editing, text-to-video and image-to-video, and audio features including speech synthesis, transcription and voice cloning. Optional audio features ship as extras. The project publishes its own speed comparisons against mlx-lm, oMLX and LM Studio, and lets users share benchmark results from their own Macs. Windows and Linux desktop builds are not available yet.

Capabilities

OpenAI-compatible and Anthropic Messages API endpoints
Tool-call parsing for coding agents, with continuous batching
One-command setup for agents such as Claude Code
Local image generation and editing, video generation and audio models
macOS desktop app for chatting and managing models
Community benchmark sharing for Apple Silicon Macs

When to use it

Serving open models from a Mac without sending data to a cloud API
Pointing a coding agent or script at a local model through a standard API
Comparing local inference speed across Apple Silicon machines
Trying local image, video or speech models on an M-series Mac