vmlx を Homebrew Cask でインストール

プロジェクトリンク

• ホームページ
• リポジトリ
• パッケージマネージャページ

コマンドとエイリアス

ローカルパッケージデータベースに実行可能エイリアスは見つかりませんでした。

プロジェクトの歴史

vMLX was created by Jinho Jang/JANGQ AI as an Apple-Silicon inference and model-serving project. It developed into the engine behind a desktop MLX Studio experience, combining chat, model-session management, image generation, conversion tools, and API reference functions.

採用の歴史

vMLX is distributed both as a macOS desktop application and a Python package. Its compatibility with OpenAI and Anthropic client protocols lets existing SDKs and agent tools use locally hosted Apple-Silicon models by changing the endpoint rather than adopting a project-specific API.

使われ方

Desktop users download the macOS application, load local models, and manage chat or server sessions from the UI and menu bar. Command-line users install the Python package and invoke commands such as `vmlx serve`, `vmlx convert`, `vmlx info`, `vmlx doctor`, and `vmlx bench`. Existing clients can use the local HTTP endpoint with OpenAI- or Anthropic-compatible SDKs.

パッケージ好きにとっての重要性

The project packages an unusually broad local-AI stack—desktop UI, Python inference engine, model conversion, quantization, API serving, and developer tools—around Apple's MLX ecosystem. Package enthusiasts care about it as a Mac-native alternative to general-purpose local model servers and as a way to exploit unified memory, Metal acceleration, and aggressively quantized models.

タイムライン

• 2026: Official project documentation described vMLX as both a Python engine and the backend of a multi-mode MLX Studio desktop application.

Related projects

• Apple MLX and the mlx-lm/mlx-vlm ecosystem provide core Apple-Silicon model primitives.
• JANG supplies the project's adaptive mixed-precision quantization format and runtime integration.
• MLX Studio is the desktop product surface built around vMLX.

ソース

• https://github.com/jjang-ai/mlxstudio#readme
• https://github.com/jjang-ai/vmlx#readme
• https://mlx.studio/