Phone Harness

AI agents that need to use mobile apps face a wall: most apps have no API, and controlling a real phone interface requires jailbreaking or custom app installs. Developers building agents that need to act inside mobile apps currently lack a straightforward way to give those agents real input and real screen perception without heavy device setup. This leaves teams unable to automate phone-based workflows at scale or test agent behavior against actual app interfaces.

Phone Harness is an open source (MIT licensed) harness that lets an AI agent interact with a phone the way a person would: tapping, typing and scrolling through apps, with no jailbreak, no API and no app installation required. The agent perceives the screen via vision OCR or the phone's accessibility tree, then issues real input back to the device. It connects to iPhones through iPhone Mirroring (even while backgrounded) and to Android over adb, and works either with your own connected phone or with hosted cloud phones that Phone Harness provides and bills by the minute. For heavier or persistent workloads, it also offers dedicated real phones reserved around the clock, with multiple profiles, configurable concurrency, and choice of proxies.