Nativ

What is Nativ?

Nativ is a free, open-source macOS app for running AI models on Apple Silicon Macs. Built on MLX-VLM, it runs chat, vision, code, and audio models without any cloud service. It also exposes a local server so coding tools like Claude Code and Codex can use your downloaded models instead.

Top Features:

• Curated library: recommends models from Google, Cohere, Z.ai, and Liquid AI for your hardware.
• Live telemetry: shows live tokens per second, memory pressure, and thermal state.
• Local endpoint: serves OpenAI and Anthropic compatible APIs for Codex, Claude Code, and OpenCode.

Use Cases:

• Private coding: point your coding agent at a local model to keep code private.
• Model testing: try open models and measure their speed and memory on your Mac.
• Offline work: chat, caption images, and transcribe audio without sending data anywhere.

Who Can Use Nativ?

• Developers: run local models behind familiar coding tools without paying for API calls.
• ML researchers: benchmark open models on Apple Silicon with detailed performance metrics.
• Privacy teams: keep sensitive prompts and documents on company hardware at all times.

Pricing

• Open source (free): the full macOS app under the MIT license, with no accounts or credits.
• Models (free): partner models download at no cost, though they need plenty of disk space.

Pros and Cons

Pros:

• Truly free: no subscription, paywall, or enterprise tier sits behind the download.
• Fully open: every part of the desktop app is MIT licensed and open to forks.
• Apple Silicon tuned: built on MLX and tuned for M-series unified memory and Metal.

Cons:

• Mac only: it runs only on Apple Silicon Macs, leaving Windows and Linux users out.
• Large downloads: some models need nearly 20 GB of disk space before you start.
• Early software: the app is very new, so expect rough edges and frequent updates.

FAQs:

1) Does it need an internet connection?

Only for downloading models, after which everything runs locally on your Mac.

2) Which Macs are supported?

It needs an Apple Silicon Mac with an M1 chip or anything newer.

3) Can I use it with Claude Code?

Yes, it serves a local endpoint that Claude Code, Codex, and OpenCode can call.

4) Is it really free?

Yes, it is MIT licensed with no accounts, subscriptions, or credits to buy.

5) What kinds of models can it run?

It runs language, vision, video, code, and audio models from open labs.