This project enables AI agents to interact with macOS applications, allowing users to delegate tasks to their MacBooks. It's designed for AI developers and power users seeking to automate workflows across any Mac app.

How It Works

macOS-use leverages large language models (LLMs) to interpret user prompts and translate them into actionable commands for macOS applications. It achieves this by simulating user interactions like clicks and typing, enabling agents to navigate and control any app on the system. The approach aims for broad compatibility across all Mac applications.

Quick Start & Requirements

• Install via pip: pip install mlx-use
• Alternatively, clone from GitHub: git clone https://github.com/browser-use/macOS-use.git && cd macOS-use
• Requires an API key for supported providers: OpenAI, Anthropic, or Gemini.
• Recommended environment: brew install uv && uv venv && source .venv/bin/activate
• Install locally: uv pip install --editable .
• Example usage: python examples/try.py
• Documentation: https://github.com/browser-use/macOS-use

Highlighted Details

• Enables AI agents to control any macOS application.
• Supports interaction with UI elements through simulated user actions.
• Demonstrations include calculator operations, website logins, and online information retrieval.
• Future vision includes local inference with MLX and MLX-VLM for zero-cost, private execution.

Maintenance & Community

• Project is actively seeking contributors.
• Contact via X/Twitter or Discord for input and feedback.

Licensing & Compatibility

• License details are not explicitly stated in the README.
• Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

This project is under active development and carries significant risks. It can access and use private credentials, stored passwords, and interact with all apps and UI components without explicit restrictions. It may not stop at CAPTCHAs or other bot identification measures, and user discretion is strongly advised due to potential for unsupervised operation. Success rates may vary.