Nativ is a native macOS app for running AI models locally on Apple silicon. It bundles an mlx-vlm server on the MLX runtime, finds compatible models already in the Hugging Face cache, and wraps model management, chat, monitoring, and a local API server in one SwiftUI app. Inference runs on the Mac once a model is downloaded.
It can serve as a private chat app, a model manager, a performance dashboard, or a local inference server for other tools. The server exposes OpenAI-compatible chat, Responses, image, audio, embeddings, and model endpoints plus Anthropic-compatible Messages endpoints, and the app configures and launches terminal coding agents (Codex, Claude Code, Pi, Hermes, OpenCode, Aider, Goose, Crush, Qwen Code, and OpenClaw) and editors (VS Code, Cursor, Zed, JetBrains, Cline, and Continue) against the models it serves.
Beyond chat with vision, reasoning output, and response metrics, Nativ includes local image generation and editing, a shared work pane where agents can edit documents and code or operate a browser page beside the chat, MCP server connections with per-tool consent, scheduled prompts, and an extension platform. The first extension provides local voice dictation in any app with transcript history and usage statistics.
Features
- Local chat and vision: streaming conversations with image attachments, reasoning output, response metrics, and persistent history
- Model library: browse and download MLX models from Hugging Face with memory-fit warnings, preload separate language, image, speech, and embedding models
- Local API server: OpenAI-compatible chat, Responses, image, audio, embeddings, and models endpoints plus Anthropic Messages endpoints
- Coding-tool integrations: configure terminal coding agents and editors to use models served by Nativ
- Shared work pane: document, code, HTML, and website tabs beside chat that agents can edit or operate
- Image generation: generate and edit images locally with compatible MLX image models
- Performance analytics: request volume, token usage, time to first token, decode speed, and recent activity
- System monitor: per-core CPU, GPU utilization, memory and swap pressure, disk health, and thermal sensors
- MCP and scheduled tasks: connect MCP servers with per-tool consent and run saved prompts on a schedule
- Advanced inference controls: sampling, thinking budgets, structured output, KV-cache quantization, prefix caching, and speculative decoding
- Extensions and voice dictation: independently versioned extensions, starting with private local dictation in any app
- Menu bar controls: start or stop the server, switch models, and pin live CPU, GPU, and RAM graphs
