ESP32 iPhone Agent

Automate actions on a real iPhone using an AI vision model and an inexpensive microcontroller board without jailbreaking.

Automate actions on a real iPhone using an AI vision model and an inexpensive microcontroller board without jailbreaking.

The gist

ESP32 iPhone Agent is an open-source tool that allows vision AI models to control a physical iPhone like a human. By combining an inexpensive microcontroller board with an iOS companion app, it sees the screen via AirPlay and sends taps and keystrokes over USB. It is designed for small teams and creators who want to automate repetitive tasks on a real iPhone without a permanent computer connection or jailbreak.

What it does

  • Streams the iPhone display to a microcontroller board using built-in AirPlay screen mirroring
  • Controls the phone by simulating a standard USB mouse and keyboard via iOS AssistiveTouch
  • Sends screen captures to any OpenAI-compatible vision AI model to determine the next action
  • Executes actions including tapping buttons, swiping, typing text, and pressing keys
  • Replays and reviews past runs step by step with touch paths drawn directly on the phone

How it works

Architecture Diagram
ESP32 iPhone Agent Architecture Diagram

The microcontroller board plugs into the iPhone and creates a local Wi-Fi hotspot for AirPlay screen mirroring. An app on the phone captures this video feed and sends screen images to a vision AI model over cellular data. The AI decides what action to take next, and the app instructs the board to move the on-screen pointer and tap or type through a USB connection.

Best for

  • Automating routine multi-step workflows directly inside native iPhone apps
  • Testing mobile applications on real iOS devices without cloud device farms
  • Creators and small agencies needing physical phone automation without jailbreaking

Watch out for

  • Requires a Mac with Xcode and developer tools for the initial one-time flashing and app installation
  • Needs a specific $7 ESP32-S3 board with PSRAM and native USB, plus an adapter for Lightning iPhones
  • Consumes API credits from your chosen vision model provider for each visual action step
  • The iPhone must use cellular data during runs because its Wi-Fi connection is dedicated to the board

What you need: Install it on your own computer. Cost: free, open source.