A $7 ESP32-S3 gives an AI agent full control over a stock, unjailbroken iPhone. No Mac, no Xcode, no Apple Developer Account required. Henry Yuan published the full method on GitHub under the project name esp32-iphone-agent.
The pipeline has three parts. First, AirPlay screen mirroring combined with UxPlay feeds the iPhone display as a JPEG stream via GStreamer to whichever AI agent you choose. Second, the ESP32-S3 presents itself to the iPhone as a USB-HID device, emulating both mouse and keyboard. Third, iOS AssistiveTouch translates those HID inputs into on-screen actions, giving the agent a complete input-output loop over the device.
The architecture matters more than the demo. This is a hardware-level approach that sidesteps every software restriction Apple builds in, and it works on any iOS device with a USB port and AssistiveTouch support. Read the original for the GStreamer pipeline specifics and the HID enumeration details, which is where the real portability of this method lives.
[READ ORIGINAL →]