A SwiftUI control tower for local MLX models on Apple Silicon

What it does Nativ is a native macOS control panel for the mlx-vlm Python server. It discovers MLX models already cached by Hugging Face, lets you download new ones with memory-fit warnings, and wraps inference in a SwiftUI interface for chatting, image generation, and system monitoring. The app also exposes an OpenAI- and Anthropic-compatible local API at a configurable localhost address, so external editors and coding agents can talk to your Mac’s GPU directly.

The interesting bit The heavy lifting is done by a bundled, embedded Python distribution managed by NativServerKit; the SwiftUI layer handles model discovery, chat history, live CPU/GPU/RAM menu-bar graphs, and preloading multiple models while warning if you’re about to exceed unified memory. It is, in effect, a polished native frontend for a Python backend that many developers would otherwise wire up manually from the command line.

Key highlights

• Serves OpenAI-compatible chat, image, audio, and model endpoints, plus Anthropic Messages, from a local mlx-vlm server.
• Includes a system monitor showing per-core CPU load, GPU utilization, unified memory pressure, disk throughput, and SMART health.
• Preloads separate language, image-generation, and speech models simultaneously with memory-budget guardrails.
• Integrates with a laundry list of coding tools—Codex, Claude Code, Aider, Goose, and several editors—by acting as a local API backend.
• Advanced inference knobs like KV-cache quantization, prefix caching, and speculative decoding are exposed in the GUI.

Caveats

• Requires macOS 26 and Apple Silicon; Intel Macs need not apply.
• Model downloads and first-time build dependencies still require network access, so it is not an air-gapped zero-setup experience out of the box.
• Audio-only model support is listed as coming soon.

Verdict Developers who want a native Mac GUI for local MLX inference—and especially those looking to point Cursor, VS Code, or terminal agents at a local API without managing Python environments—should take a look. If you don’t have a recent Apple Silicon Mac or you prefer to orchestrate models entirely from the command line, this is just a very pretty wrapper.