Give an LLM direct control of your Mac desktop

You can now give an AI model direct control over your Mac without losing your mouse cursor.

Browser-use released macos-harness, an experimental open-source tool that reached 708 GitHub stars this week. Instead of building custom integrations for every app, it gives LLMs six basic Python primitives—like click, see, and script—to drive macOS directly. The agent writes missing logic on the fly while running inside a single persistent Python process.

Why it matters: Traditional desktop AI agents steal your cursor or force app windows into the foreground while they work. This project captures background app windows quietly, sends inputs straight to an app's process ID, and uses a virtual click-through pointer so the model works alongside you without interrupting your workflow.

It requires macOS and Python 3.12. You can install it via uv or prompt tools like Claude Code to set it up automatically. Note that anonymous telemetry runs by default, but macos-harness telemetry disable turns it off instantly.

Finally, an AI assistant that doesn't wrestle you for the mouse.