Mlx Serve

This product is a free, open source AI app and inference server for Apple Silicon Macs, written natively in Zig with no Python required. It runs open models in both MLX and GGUF formats locally, and the site describes it as faster than LM Studio and as a drop in alternative to LM Studio, Ollama and llama.cpp on macOS. Listed models include Gemma, Qwen, Llama, Mistral and DeepSeek families. The desktop app offers chat, document question answering by dragging a folder into the conversation, and a global shortcut that opens a quick prompt box over any application. It also generates images, music, video and 3D from photos, edits photos, clones voices from a short sample and runs an agent sandbox that can perform tasks on the computer. For developers it serves OpenAI compatible and Anthropic compatible APIs, so tools such as Claude Code can connect to it, and it supports MCP and Ollama compatibility. Everything runs offline and the vendor states that nothing typed leaves the machine. It targets Macs with M1 through M5 chips running macOS 26 or later. A companion iPhone app, MLX Chat, runs the same engine on device and needs an A17 Pro class or newer iPhone. The software is MIT licensed and listed at $0, with no subscription, account or API keys.