phone-harness lets AI agents drive your physical iPhone

An open-source tool now lets AI agents take full, programmatic control of your physical iPhone without a jailbreak.

Created this week, ShawnPana's phone-harness project has already collected 1,300 GitHub stars. It gives LLMs control over a real iPhone by using macOS Sequoia's built-in iPhone Mirroring app. The tool uses screencapture and Apple's Vision framework OCR for eyes, then posts system CGEvents to tap, swipe, and type.

Why it matters: You don't need Xcode, WebDriverAgent, or jailbreaks to automate mobile workflows. Because it treats the mirror window as its interface, an agent can drive real iOS apps straight from your Mac terminal without installing helper software on the phone itself.

Know this: Setup requires granting your Mac terminal two permissions: Screen Recording to see the window and Accessibility to send taps. It has clear limits. Unlocking your physical phone pauses the session, DRM video renders black, and OCR cannot read unlabeled icons without a vision model. But the core codebase remains tiny at roughly 500 lines of Python.

Your phone just became another scriptable terminal window.