Local Inference
Blazing-fast in-process execution powered by Apple's native MLX Swift framework. Your prompts and data never leave your Mac.
Native application for macOS
LocalModel Bridge
Local LLM
& Shortcuts
Run local AI models (Llama, Qwen) 100% offline on your Mac. Designed exclusively for Apple Silicon, this menu-bar utility exposes a blazing-fast local LLM inference engine system-wide via Shortcuts, macOS Services, or the built-in floating Chat window.
01
Features
Local generative AI, integrated into macOS.
Shortcuts Integration
Trigger and chain AI queries directly within Apple's native Shortcuts app to automate your workflows.
macOS Services
Select text in Safari, Xcode, or Mail, right-click, and replace it instantly with the model's output (rewrite, summarize, translate).
Floating Chat
A standalone floating chat window with multi-turn conversation history to quickly chat with your loaded models.
Smart Memory Control
Automated model unloading after inactivity timeouts and Metal cache purging to release unified memory back to macOS.
Hugging Face Profiles
Enter any Hugging Face Hub ID to download a model automatically, or point to your own local weight directories.
02
Bring generative AI
Automations
Bring generative AI
into your workflows.
LocalModel Bridge exposes system-wide AppIntents for seamless control over your offline inference.
Query Local Model
Send a prompt to your local model and capture the generated text directly inside your shortcuts.
Load Model
Pre-warm system memory by loading a model profile in the background before repetitive tasks.
Unload Model
Instantly release your Mac's unified memory by unloading the model as soon as automation tasks end.
- 01 CLI Automations Query models directly from the Terminal using simple terminal CLI scripts.
- 02 Services Menu Rephrase, translate, or correct highlighted text in Safari, Xcode, or Mail in one click.
- 03 Local API Server Run a local OpenAI-compatible HTTP server (localhost:8080) for third-party developer integrations (e.g. Cursor).
03
Data Privacy
Absolute offline privacy by design.
All prompts, processed documents, and chats never leave your computer. Inferences run locally and offline.
Local Inference
Model operations run in-process on Apple Silicon cores using Metal shaders. No external cloud servers involved.
App Sandbox Security
Network access is strictly confined to Hugging Face model downloads initiated by the user and local host APIs.
No User Analytics
No usage data, prompts, or selected models are collected, monitored, or transmitted by the application.
Security Guarantee: Runs fully offline without accounts, and requires no internet connection after downloading a model.
Read the full privacy policy →
04
Support
Questions about local inference?
We are here to help you configure model files, set up Apple Shortcuts, or integrate the local API server.
Visit support & FAQ →