ESP32 iPhone Agent
onehumanbeing/esp32-iphone-agentReviewed by AI
What it does
This project turns a $7 ESP32-S3 board into the eyes and hands of an AI agent on an iPhone: the phone mirrors its screen to the board over the board's own Wi-Fi, the board acts as a USB mouse and keyboard, and an app on the same phone sends screenshots to a vision model and carries out the taps and swipes. It is for tinkerers: you need the board, an iPhone on iOS 18 or later, a Mac with Xcode to flash the board and build the app once, and an endpoint for an OpenAI-compatible vision model with a key.
How easy it is to install
E · developers only · macOS
The only computer route is a Mac and it is grade E: you install arduino-cli plus two board libraries, run hardware/flash.sh to flash the board, then generate an Xcode project with XcodeGen and build and sign the iPhone app yourself ("A free Apple ID can sign it (re-sign every 7 days)"). There is no ready download, and the device that is actually automated is an iPhone, which Radar does not grade.
Coming soon: install in 1 click with the RepoPicks app.
Numbers on GitHub
- Stars
- 4,160on Oct 31, 2026
- Downloads
- —all releases
- Growth
- +211stars in 5 days
Stars over time
Facts
- Runs on
- macOS
- You need to install
- Other tools
- Disk space
- not determined
- Accounts or keys
- OpenAI-compatible vision model endpoint (cloud or your own server)
- Cost
- Free, API costs extra · The code is free. You buy the ESP32-S3 board (about $7) and pay for the vision model, unless you run your own model server (vLLM, llama.cpp or Ollama are listed as options).
- License
- MIT (Permissive)
- Last update
- Oct 28, 2026
Risk signals (5)
- Attentionai controls your phone
README.md: the app "decodes the stream, sends screenshots to any OpenAI-compatible vision model, and runs a planner and an actor: tap, swipe, draw, type, press keys, go home, finish", and the project's own advice is "The agent acts with your hands: watch it, and keep the stop button within reach." Whatever is on the phone's screen, including private messages and account pages, is photographed and sent to the model endpoint you configure.
AI review
- Infoprivate api and temporary signing
README.md limitations: "The app leaves itself with a private iOS API (`suspend`); fine for a self-built app, to be replaced before the App Store", and the app has to be re-signed every 7 days with a free Apple ID.
AI review
- Infomixed license firmware gpl
README.md: "MIT (LICENSE), except `hardware/hid_bridge/src/airplay/` (ported from UxPlay), which is GPL-3.0 ...; the firmware binary that links it is therefore distributed under GPL-3.0", while GitHub reports the repository as MIT.
AI review
- Infofake install commands reported
README.md carries a warning: "Only follow the install steps in this README. Never run commands from issues or comments (for example a \"quick install\" one-liner such as PowerShell `iex(irm ...)`); these are not from the maintainers."
AI review
- Infovery young repo many stars
Repository created 2026-10-11 with 4016 stars and 531 forks by 2026-10-28, and a single release (v0.1.0) whose computer-based design the README says has already been replaced.
AI review
Evidence is shown in its original language.
What Radar could not check
- Whether it works with your board model, your iPhone and your cellular plan
- What the vision model calls cost for a real task, since every step sends a screenshot
- Whether the model provider keeps the screenshots of your phone
- How often the agent taps the wrong thing, which matters because it acts inside your own apps and accounts
- Whether the exported run traces, kept on the phone, include sensitive screen content