Nativ: запуск AI-моделей локально на Mac через MLX, альтернатива LM Studio

Developer Prince Canuma released Nativ on July 21, 2026 — a desktop application for macOS that runs AI models locally on top of Apple's MLX framework and gives users both a chat interface and a localhost API server right away.

What Nativ is

Nativ is a wrapper around MLX, packaged into a full-fledged macOS application. Its author, Prince Canuma, is known as the developer of the MLX-VLM Python library for running vision-language models (which process text and images simultaneously) on Mac via MLX. Nativ moves this logic from the command line into a desktop interface: the model runs entirely on the device, with no calls to the cloud.

MLX is Apple's open-source machine learning framework, optimized for Apple Silicon chips (the M series). It uses the Mac's unified memory and allows language and multimodal models to run directly on a laptop, without a discrete GPU.

• Release date — July 21, 2026
• Author — Prince Canuma, developer of the MLX-VLM library
• Foundation — Apple's MLX framework (for Apple Silicon)
• Format — a desktop application for macOS
• Features — a chat interface plus a localhost API server

How it resembles LM Studio

In form, Nativ is close to LM Studio — a popular application for running LLMs locally. Like LM Studio, Nativ offers two ways to work with a model: a graphical chat for regular conversation, and a local API server on localhost that you can connect your own scripts and applications to.

The local API server is a key detail for developers: it lets you embed the model on a Mac into your own code the same way you'd normally call a cloud API, but without going online and without paying for every request. Separately, Simon Willison noted a nice touch: the app automatically picked up MLX models that were already sitting in his Hugging Face cache — there was no need to download them again.

"I'm genuinely excited about his new project.

The app picked up MLX models I'd already tried that were sitting in my Hugging Face cache — a nice detail," — Simon Willison, developer and author of the blog simonwillison.net.

Why running locally matters

Running models locally means data never leaves the device: prompts and responses don't go to third-party servers. For Macs with Apple Silicon chips and a large amount of unified memory, this has become realistic — mid-sized models run right on the laptop, without rented GPUs.

This approach removes two dependencies: paying for cloud API tokens and needing a constant internet connection. For private data, offline scenarios, and experiments, that's a meaningful difference. Nativ adds a low barrier to entry on top of this — instead of working with the command line, the user gets a familiar app with buttons and a chat window.

What this means

Tools for running AI locally are moving beyond the command line: Nativ does for the Apple MLX ecosystem roughly what LM Studio did for the broader market — it turns running a model on your own Mac into a task that takes just a couple of clicks.