maclocal-api (afm)

Overview

AFM turns an Apple silicon Mac into a private, OpenAI-compatible AI server. It runs Hugging Face MLX models or Apple's on-device Foundation Model, so you can connect the clients and SDKs you already use without a cloud account or API key. It is a native Swift executable and is built with multi-turn, tool-using agents in mind.

Features

• Chat, streaming, tools, structured output, reasoning and logprobs
• OpenAI Responses and Anthropic Messages compatibility
• MLX runtime for open models and VLMs, plus Apple Foundation Models with .fmadapter adapters
• Gateway mode that combines Ollama, LM Studio, Jan and other local servers into one model list
• Web UI and a full-screen terminal chat (--tui)
• Built-in evaluation suites for local models

Requirements

• An Apple silicon Mac
• macOS 26 or later
• MLX model weights download from Hugging Face on first use

Install

1. Install with Homebrew (or pip install macafm)

   brew install scouzi1966/afm/afm

2. Start a small MLX model and open the Web UI

   afm mlx -m Qwen3-0.6B-4bit -w

Example

Run afm --tui for a private terminal chat with Apple's on-device model.