Nativ is a new open-source Mac app that runs open AI models entirely on your own machine, with no account, no subscription, and no cloud. Built by MLX developer Prince Canuma and released as v0.0.1 on July 20, 2026, it reached the Hacker News front page within hours. It lets you chat with local models, serve them through an OpenAI- or Anthropic-compatible API, and point coding tools like Claude Code and Codex at that endpoint.

What This Enables

Nativ exposes a bundled server at http://127.0.0.1:8080 that speaks both the OpenAI and Anthropic API formats, so you can run a coding agent, a chat app, or a batch script against a model on your laptop instead of a paid cloud API. Because it is built on Apple's MLX framework and the mlx-vlm server, it runs text and vision models on unified memory, and you can attach images to a chat. For a builder, that means a private, zero-cost inference backend you can wire into Claude Code, Codex, Pi, Hermes, or OpenCode in a few minutes.

Why It Matters for Creators

Local model runners are how many builders escape per-token API bills and keep prompts private. Nativ enters a field led by LM Studio and Ollama, but it is Apple-Silicon-native and leans on MLX rather than llama.cpp, which can mean better throughput on M-series memory. The Anthropic-compatible endpoint is the notable part: tools written for Claude, including Claude Code, can target a local model with no rewrite. That lowers the cost of testing open models inside workflows already built around commercial agents.

Key Details

Requirements: a Mac with Apple Silicon and macOS 26 or newer, plus enough unified memory for your chosen model.

Install: download the DMG from the releases page, drag Nativ to Applications, and launch. There is no sign-up.

Features: local chat with streaming and image input, a model library for discovering and managing MLX models, performance analytics for request volume and token usage, and local OpenAI- and Anthropic-compatible APIs.

License and price: MIT, free, funded through sponsorship rather than a paid tier. The source code is on GitHub.

What to Do Next

If you build on Apple Silicon, install Nativ, pull an MLX model from the built-in library, and start the server. Then set your agent's base URL to the local endpoint and its model name to the one you loaded, and compare tokens per second against your current setup before moving any production calls off a paid API. Because this is a v0.0.1 release, keep a cloud fallback for anything time-sensitive while the app matures.