Nativ makes local models a desktop API

Simon Willison points to Nativ, a macOS app that wraps MLX models in a desktop chat interface and a localhost API server, similar in shape to LM Studio. Why it matters: Local inference becomes more useful when it is not just a terminal experiment. A desktop wrapper plus local API gives builders a small bench for private prototypes, latency checks, and model swapping.

Try this: Treat it as a sandbox: check where models are stored, how the localhost API is authenticated, and whether it can reuse existing Hugging Face cache files safely.