Back home

Nativ

x+web 0.72
Visit source

Intro

Nativ is the flagship app of Neywa Labs — a free, open-source, native macOS app for running AI models locally on Apple Silicon. Neywa Labs was founded by Prince Canuma, the open-source author behind mlx-vlm and mlx-audio, and officially launched with Nativ on 2026-10-27.

Product page: neywalabs.ai/products/nativ. Source: Blaizzy/nativ on GitHub (~1.5k stars). Tagline: Local AI, native to your Mac — no subscriptions, no per-token bills, and your data stays on your device.

Explain

Key features

  1. Local chat and vision: streaming chat, image attachments, reasoning output, response metrics and persistent history.
  2. Model library: discovers MLX models in your Hugging Face cache, lets you browse/download compatible models, and recommends models that fit your hardware (open models from Google, Cohere, Liquid AI and others).
  3. Performance analytics: live tokens/sec, time-to-first-token, memory pressure and thermal state.
  4. Local APIs: OpenAI-compatible chat, Responses, image, audio and model endpoints, plus Anthropic Messages endpoints.
  5. Coding-tool integrations: point Codex, Claude Code, Pi, Hermes and OpenCode at models served by Nativ.
  6. Advanced inference controls: sampling, thinking budgets, structured output, KV-cache quantization, prefix caching and speculative decoding.
  7. Under the hood: an embedded mlx-vlm server, a SwiftUI app and menu-bar controls.

Why this radar term

The official launch of Neywa Labs and Nativ became an X trending news story. Nativ had existed as an open-source preview, but this is the company's and product's official debut; it was not previously tracked.

Usage

Install

  1. Requirements: an Apple Silicon Mac, macOS 26 or newer, and enough unified memory for your chosen model
  2. Install with brew install --cask nativ, or download the DMG from GitHub Releases and drag it to Applications
  3. First launch: pick an installed language model or use load-on-demand; optionally generate an API key to protect management endpoints
  4. Open Models to download or switch models, then start chatting
  5. Connect coding agents: configure Codex, Claude Code, OpenCode and others to use the local endpoint

Good fit

  • Privacy-minded Mac users who want AI offline (even on a plane)
  • Developers who want local models behind their coding agents to cut API costs

Caveats

  • Model downloads and first-time setup need network access
  • What you can run is bounded by your Mac's unified memory