Nativ is a native macOS app from Blaizzy, maintainer of the mlx-vlm and mlx-audio libraries, that turns a Mac with Apple silicon into a local AI workspace built on Apple's MLX runtime. Under a SwiftUI shell, NativServerKit owns an embedded, relocatable Python distribution and bundles an mlx-vlm server that discovers models already in your Hugging Face cache, honoring HF_HUB_CACHE and HF_HOME, so the app doubles as a chat client, model manager, performance dashboard, and OpenAI- and Anthropic-compatible local inference server reachable at http://127.0.0.1:8080 by default, with host and port configurable from a Developer page. Chat supports streaming text and vision with image attachments, reasoning output, and persistent history, and Command+Shift+N opens independent windows sharing the same server, loaded models, and settings. The Model library browses and downloads compatible Hugging Face models with memory-fit warnings, can store the cache on an external APFS drive, and can preload separate language, image-generation, speech, and embedding models at once, warning if the combination would exceed unified memory. A system monitor tracks CPU, GPU, memory, and thermal sensors, alongside token-usage and decode-speed analytics. It can configure terminal coding agents such as Codex, Claude Code, Aider and Goose, and editors like VS Code, Cursor and Zed, pointed at models served locally. An extension platform lets independently versioned capabilities be installed or removed, starting with an Audio extension for voice dictation. Nativ also connects MCP servers with per-tool consent and runs saved prompts on a schedule. It requires macOS 26 or newer on Apple silicon and enough unified memory for the loaded model; inference runs on-device once a model is downloaded, though downloads and, when building from source, first assembly of the bundled Python runtime need network access. Version 0.3.7, released 2026-09-07, added chat import/export, an external model cache, and an optional SearXNG search provider. It installs from a signed DMG on GitHub Releases or via Homebrew, with Sparkle handling updates.