Projects / oMLX

oMLX

Desktop & Productivity

An Apple Silicon model-inference server with a macOS menu-bar app, admin/chat interface and OpenAI-compatible endpoint.

Use with conditions
Last updated2026-10-02
Open source licenseApache-2.0
PlatformsmacOS 15+ on Apple Silicon
DeploymentSelf-hosted service

About this project

oMLX is a model-inference server for Apple Silicon with a macOS menu-bar application. It manages models, memory use and cached context, and exposes an OpenAI-compatible endpoint for client tools. A built-in administration and chat interface can be accessed through the locally running server.

The documented app path is to download a DMG and move it into Applications. Homebrew and source installation are alternatives. The supported system is Apple Silicon macOS 15.0 or later; the source path uses Python 3.11 through 3.13. Initial setup selects a model directory and obtains the model weights needed for inference.

The repository is Apache-2.0, while model-weight licenses remain separate. Available memory and storage depend on the selected model, and access beyond the local machine needs authentication. The README's benchmark figures are maintainer measurements, not a reproduced result for the reader's device, model or workload.

Decision guide
01

Why it is worth a look

  • Run local models on Apple Silicon
  • Manage the inference server from a macOS menu-bar app
02

Before you use it

  • Apple Silicon macOS only
  • Model weights, memory and disk requirements vary
  • Model licenses are independent of Apache-2.0
  • Authenticate network access and treat maintainer benchmarks as workload-specific
Manual setupModerate
With an AI agentModerate
DeploymentSelf-hosted service
Additional requirements1 confirmed; others not verified

Apple Silicon macOS only; Model weights, memory and disk requirements vary; Model licenses are independent of Apache-2.0; Authenticate network access and treat maintainer benchmarks as workload-specific; Confirmed requirements: An always-on device or service; Some requirements have not been verified. Check the official setup instructions.

03

What it can do

  • Run local models on Apple Silicon
  • Manage the inference server from a macOS menu-bar app
  • Chat UI and OpenAI-compatible local endpoint

Who it fits

Good for

Managing local model inference on an Apple Silicon Mac; Connecting a compatible client to a locally hosted endpoint

Not a fit

Windows, Linux or Intel Mac use; Assuming a small model benchmark applies to every model

Getting started

  1. Check Apple Silicon and macOS 15 or later, then download the official DMG
  2. Select a model directory and obtain an appropriately licensed model that fits your memory and storage
  3. Test the built-in local chat or endpoint, and configure authentication before permitting network access