Nativ for Mac: MLX chat, image generation and scheduling

Nativ is presented as an open source Mac app for local AI built around MLX VLM. The walkthrough covers chat, model management and tasks beyond text inference. Users can select a model and its reasoning level, while the developer section exposes OpenAI-compatible and Anthropic-compatible endpoints.

The speaker reports decoding and prefill results virtually identical to the standalone MLX VLM server. This is his comparison, rather than a general performance guarantee. He also shows how to launch OpenCode against Nativ's server, choosing a working folder and using the supplied commands for a coding assistant connection.

Model management scans configured folders and uses Hugging Face for discovery. Models need to be in MLX format. An image generation example produces a downloadable image; the artifacts section stores generated images and documents. A separate vision example extracts text from a selected screenshot region.

Most extension tools are opt-in, according to the speaker. He configures web reading with Tavily and enables a research kit that combines skills and tools. The scheduler demo selects a model and that kit to fetch an RSS feed each day at 5:09 p.m. Nativ creates an ongoing chat session for the task, and the completed run returns a summary.