Nativ

Nativ is a free, open-source macOS app for running AI models directly on an Apple Silicon Mac. It combines a native chat interface with model management, performance analytics, and a local inference server that other applications can connect to.

Chat and manage models on your Mac

• Local chat: Stream responses, attach images, view reasoning output when supported, and keep conversation history.
• Model library: Discover compatible MLX models, download them from Hugging Face, reuse cached models, and check memory-fit warnings before loading.
• Images and audio: Generate or edit images with supported models. An optional audio extension adds on-device dictation and speech transcription.
• Live monitoring: See tokens per second, response latency, model activity, and CPU, GPU, and unified memory usage.

Connect coding agents to local AI

Nativ bundles an MLX-VLM server with OpenAI- and Anthropic-compatible API endpoints. Integrations help connect tools such as Codex, Claude Code, OpenCode, VS Code, and Cursor to a model running on your Mac. It also supports Model Context Protocol (MCP) tools with approval before tool execution, plus scheduled prompts.

Compatibility and setup

Nativ requires macOS 26 or later and an Apple Silicon Mac; Intel Macs are not supported. The memory needed depends on the model. Model downloads require an internet connection, but inference runs locally once a model is installed. Certain gated Hugging Face models need an access token, and optional voice features may require microphone or Accessibility permissions.

Nativ is distributed under the MIT license, with no Nativ account or subscription required. Install it from GitHub Releases or through Homebrew.