Nativ is a new, 100% open-source application that allows users to run frontier AI models locally on their Mac computers, specifically those with Apple Silicon (M1+). It offers a "real app, real local models, no cloud" experience, eliminating the need for accounts, subscriptions, or cloud services.

Users can interact with open models through a chat interface featuring streaming responses and per-message performance metrics. The curated library includes standout models from Google, Cohere, and Liquid AI, with Nativ recommending optimal models for specific hardware. The application supports multiple modalities, including language, vision, video, code, and audio, enabling tasks like image captioning, video summarization, code autocompletion, and speech generation.

Nativ provides detailed telemetry, showing tokens/sec, memory pressure, and thermal state, appealing to developers. It is optimized for Apple Silicon using MLX-VLM and Metal. The software is free forever and MIT-licensed, with no paywalls or data selling. Integrations allow users to connect existing coding agents to locally running models via a single model server endpoint.

The developers emphasize their commitment to open source, contrasting Nativ with proprietary "local AI" apps. They state that every part of Nativ is open for inspection, forking, or contribution, free from VC roadmaps or enterprise tiers, aiming to empower researchers and hackers.