Blaizzy/nativ

What it solves

Nativ provides a unified, native macOS workspace for running AI models locally on Apple silicon. It eliminates the friction of managing separate Python environments and server setups by bundling a server and a polished user interface into a single app, allowing users to chat with models, generate images, and serve local APIs without complex configuration.

How it works

Nativ uses a SwiftUI application that interfaces with NativServerKit, which manages an embedded Python distribution and a bundled mlx-vlm server. This server leverages the MLX runtime to run models directly on Apple's unified memory. It can discover models in the Hugging Face cache and expose them via OpenAI- and Anthropic-compatible local APIs for integration with other tools.

Who it’s for

It is designed for macOS users with Apple silicon who want a private, local AI experience, as well as developers who need a local inference server to power coding agents and other AI-integrated applications.

Highlights

• Local Chat and Vision: Supports streaming conversations, image attachments, and reasoning output.
• Image Generation: Dedicated tab for generating and editing images locally using MLX models.
• Model Management: Browse, download, and switch MLX models from Hugging Face with memory-fit warnings.
• Local API Server: Provides OpenAI- and Anthropic-compatible endpoints for external tools like VS Code, Cursor, and Aider.
• System Monitoring: Live tracking of CPU/GPU load, unified memory pressure, and thermal sensors.
• Extension Platform: Supports modular capabilities, including a built-in audio extension for voice dictation.
• MCP Integration: Connects to Model Context Protocol servers to give the chat model access to external tools.
• Advanced Inference: Controls for sampling, thinking budgets, KV-cache quantization, and speculative decoding.