ESP32 iPhone Agent Gives AI iPhone Control

The cleverness is in exploiting iOS’s supported screen-mirroring and accessibility paths to give vision models hands on a stock phone. It is a compelling open-source computer-use experiment, though the $7 board is only the entry point.

• –AirPlay carries the screen to the board, while BLE forwards video to the iPhone app and USB HID turns model actions into pointer and keyboard input.
• –Any compatible cloud or self-hosted vision endpoint can provide the model, keeping the project relatively vendor-neutral.
• –A Mac with Xcode remains necessary for flashing and app installation; free Apple ID signing expires after seven days.
• –BLE bandwidth limits the mirror to roughly 480×1040 resolution and five frames per second, with several seconds per action.
• –Portrait-only apps, ASCII text entry, inaccessible screens, cellular-data requirements, and a private iOS API keep this closer to a developer prototype than a consumer product. [Project documentation](https://esp32-iphone-agent.com/how-it-works/)