Nativ: Run Local AI Models Natively on Macs

Nativ is a free, open-source macOS app for running MLX models locally on Apple silicon.

It combines AI chat, model discovery, downloads, performance analytics, server controls, and OpenAI- and Anthropic-compatible localhost APIs. Requires macOS 26 or newer.

Start a conversation inside Nativ, inspect its speed and resource use, or connect Claude Code, Codex, Pi, Hermes, or OpenCode to the local server.

The dashboard tracks token usage, time to first token, decode speed, memory pressure, thermal state, and recent requests.

Available unified memory determines which models fit and how much capacity macOS retains for other applications.

Key Features

• Runs local text and vision models through MLX on Apple silicon.
• Streams responses and stores chat history inside the Mac app.
• Finds compatible models already stored in the Hugging Face cache.
• Downloads, switches, inspects, and removes MLX models from one library.
• Tracks token usage, response latency, decode speed, memory pressure, and thermal state.
• Exposes OpenAI-compatible endpoints for chat, responses, models, images, and audio.
• Accepts Anthropic-compatible Messages and token-counting requests.
• Connects Claude Code, Codex, Pi, Hermes, and OpenCode to local models.
• Displays server health, runtime details, live logs, and cache statistics.
• Controls the server and loaded model from the macOS menu bar.

Nativ v.s. Ollama & LM Studio

Ollama and LM Studio manage their own model formats and serve them through their own runtimes. While Nativ runs models through mlx-vlm, Apple’s MLX inference stack, and pulls those models straight from a Mac’s existing Hugging Face cache.

Use Cases

Open a repository through Claude Code or Codex, then select Nativ as the model endpoint. The coding client handles files, commands, and repository tools while model requests go to the MLX model loaded on the Mac.

Load several compatible models in turn and run the same prompt through each one. Nativ records response speed, time to first token, token use, and system pressure, creating a practical basis for choosing a model that fits the hardware.

Attach a screenshot, chart, interface mockup, or photograph to a conversation with a compatible vision model. The response and its performance data appear in the same chat workspace.

Point an existing development client at the local OpenAI- or Anthropic-compatible endpoint. Applications built around those request formats can test local models through Nativ before their backend configuration changes elsewhere.

How to Use Nativ

1. Open the latest release page and download the DMG.

2. Drag Nativ into the Applications folder.

3. Launch the app on an Apple silicon Mac running macOS 26 or newer.

4. Select a compatible model already present in the Hugging Face cache, or open Models and download one.

5. Generate a server API key if local endpoint protection is required.

6. Open Chat, enter a prompt, and wait for the model to load.

7. Open Analytics to review speed, token use, memory pressure, and recent activity.

8. Open Integrations or Developer to configure a coding client or copy an endpoint address.

The default server address is:

http://127.0.0.1:8080

A basic chat-completions request looks like this:

curl http://127.0.0.1:8080/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "your-model-id",
    "messages": [
      {
        "role": "user",
        "content": "Explain the main modules in this project."
      }
    ],
    "stream": false
  }'

Add the following header when the server API key is active:

-H 'Authorization: Bearer your-api-key'

Pros

• Native SwiftUI Mac app
• Integrated MLX model management
• OpenAI-compatible local endpoints
• Anthropic Messages compatibility
• Built-in inference telemetry
• Direct coding-agent integrations
• MIT-licensed source code
• No account or subscription

Cons

• Large local storage requirements
• macOS 26 minimum
• No Intel Mac support
• MLX-focused model compatibility
• Unified memory limits model size

Alternatives & Related Resources

• Osaurus: Free Local AI Agents for Mac
• Run Any LLMs Locally On Mac & Windows – LM Studio AI
• Claude Code Local: Run Claude Code With Local AI on Apple Silicon
• Wayland: Local-First AI Agent Command Center for Claude Code & Codex
• SmallCode: Fast, Free, Local AI Coding Agent for Small LLMs
• The Ultimate Claude Code Resource List: Agents, Skills, Plugins & More

FAQs

Q: Does Nativ’s MIT license cover downloaded AI models?
A: No. The MIT license covers the Nativ application source code. Each downloaded model carries a separate license that controls redistribution, commercial use, attribution, and other usage terms.

Q: Does API compatibility make a local model behave like GPT or Claude?
A: No. API compatibility covers request and response formats supported by the server. Output quality, context handling, tool use, reasoning behavior, and instruction following depend on the selected local model.

Q: Can I use Nativ’s server with tools other than the five listed coding agents?
A: Yes. Any client that supports the OpenAI or Anthropic API format can point at Nativ’s local server by changing its base URL.

Q: Does Nativ send my data to the cloud?
A: No. Inference runs entirely on your Mac. No accounts, no subscriptions, no cloud, and no data to sell. Model downloads and first-time build dependencies require network access. Your conversations and prompts never leave your device.

Q: What models can I run?
A: Nativ works with MLX-format models, including open models from Google, Cohere, and Liquid AI. The model library allows you to discover installed MLX models, browse compatible models on Hugging Face, and download them directly.