MLX Studio
A native macOS app that wraps a self-hosted vMLX inference server, serving language, vision and image-generation models over an OpenAI-compatible API on Apple Silicon with no Python setup.
MLX Studio is a full-featured macOS app built on Apple's MLX framework that wraps the vMLX inference server, serving large language models, vision models and image-generation models through an OpenAI-compatible HTTP API on Apple Silicon. Models are browsed and downloaded from Hugging Face directly in the app, and a built-in native parallel downloader resumes across restarts.
Features include multi-turn chat with streaming, markdown and code highlighting, sending images to multimodal models, native support for reasoning models such as DeepSeek-R1 and QwQ, function calling with structured output, text-to-image generation, Kokoro-based text-to-speech, Whisper-based transcription, a menu-bar mode, and developer tools for model inspection, conversion and quantization. The serving engine supports continuous batching with PagedAttention, speculative decoding, KV cache quantization, prefix caching and hybrid Mamba/SSM plus transformer architectures.
Conversations are saved and restored through SQLite persistence. Releases are signed and notarized by Apple.