Nativ

Local AI, native to your Mac.

Chat, serve, monitor, and connect MLX models from one macOS app.

Nativ is a native macOS workspace for running AI models locally on Apple silicon. It bundles an mlx-vlm server, finds compatible models in your Hugging Face cache, and wraps the whole experience in a polished SwiftUI app.

Use Nativ as a private chat app, a model manager, a performance dashboard, or an OpenAI- and Anthropic-compatible local inference server for the tools you already use.

Inference runs on your Mac after a model has been downloaded. Model downloads and first-time build dependencies still require network access.

Support for dedicated audio-only and image-generation-only models is coming soon.

flowchart LR
    A["Nativ · SwiftUI app"] --> B["NativServerKit"]
    B --> C["Bundled mlx-vlm server"]
    C --> D["MLX runtime"]
    D --> E["Local models · Apple unified memory"]
    F["Apps and coding agents"] -->|"localhost API"| C

NativServerKit owns the embedded Python distribution and server lifecycle. The app adds model discovery, chat, analytics, configuration, integrations, logs, menu bar controls, and software updates around that runtime.

To run the app:

• A Mac with Apple silicon.
• macOS 26 or newer.
• Enough unified memory for the model you choose.

To build from source, you will also need:

• Xcode with the macOS 26 SDK.
• xcodegen.
• Python 3.
• Network access to GitHub Releases and PyPI while the embedded Python bundle is first assembled or refreshed.

Download the latest DMG from GitHub Releases, drag Nativ to Applications, and launch it. Nativ uses Sparkle for subsequent in-app updates.

On first launch:

1. Choose an installed language model, or continue with load-on-demand.
2. Optionally generate an API key to protect the server's management endpoints.
3. Open Models to download or select a compatible model.
4. Start chatting, inspect analytics, or connect one of the supported coding tools.

brew install xcodegen
make xcode-generate
make xcode-build
open build/XcodeDerivedData/Build/Products/Debug/Nativ.app

The first build can take a while because NativServerKit creates a relocatable Python runtime and installs the pinned mlx-vlm server dependencies into the framework resources. Later builds reuse the bundle until an input changes.

By default, the app exposes its server at http://127.0.0.1:8080. The Developer page lists every available endpoint and lets you copy URLs directly.

For example, with a model selected:

curl http://127.0.0.1:8080/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "your-model-id",
    "messages": [{"role": "user", "content": "Why is the sky blue?"}],
    "stream": false
  }'

If you enabled a server API key, also send it as a Bearer token:

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

The server includes:

• OpenAI-compatible /v1/chat/completions, /v1/responses, /v1/models, image, and audio routes.
• Anthropic-compatible /v1/messages and token-counting routes.
• /health, /metrics, cache statistics, cache reset, and model unload endpoints.

Sources/
├── Nativ/                       # SwiftUI application
│   ├── Features/
│   │   ├── Chat/
│   │   ├── Dashboard/
│   │   ├── Developer/
│   │   ├── ImageGeneration/
│   │   ├── Integrations/
│   │   └── Models/
│   ├── Assets.xcassets/
│   ├── ModelProviderIcons/
│   └── Utilities/
└── NativServerKit/              # Embedded server and Swift clients
PythonDistribution/
├── Launcher/                    # Relocatable server launcher
├── Requirements/                # Pinned Python dependencies
└── Scripts/                     # Bundle assembly and verification
Configuration/                   # App metadata and signing settings
Design/                          # Brand source files and README artwork
scripts/                         # Archive, signing, notarization, and release tools
project.yml                      # XcodeGen project definition

Generate and build the Xcode project:

make xcode-generate
make xcode-build

Verify that the bundled executable can launch and print mlx_vlm.server help:

make xcode-smoke

Exercise the long-running process lifecycle and /metrics readiness:

make xcode-lifecycle-smoke

To generate a few real requests and compare metrics before and after:

scripts/run_metrics_queries.py

The first request may take longer while its model downloads and loads.

Built for fast, local inference on Apple silicon.