🤖 A $7 chip gives AI real hands on a real iPhone — no jailbreak, no computer

Tools & Scripts
, ,
1

:robot: Your iPhone just hired a robot intern — for the price of a coffee

ESP32 iPhone Agent (Handsy): a $7 chip gives any vision model eyes and hands on a real iPhone — no jailbreak, no computer, no developer account · open source, 4.8K :star:

a 7 dollar chip gives AI hands on an iPhone
a 7 dollar chip gives AI hands on an iPhone1200×900 233 KB

give it a goal — walk away — come back to a Duolingo streak, a set alarm, or a text that sent itself

board ($7) → flash it in your browser → free app → done: handsy.tech/install · GitHub

① :eye: the eyes

  • :television: the board opens its own Wi-Fi hotspot and appears in Screen Mirroring — your screen streams to it over AirPlay
  • :eyes: it watches like you do: pixels, not APIs — so any app a person can use, it can use

② :raised_hand: the hands

  • :computer_mouse: the same board pretends to be a USB mouse + keyboard — AssistiveTouch turns the pointer into taps, swipes and typing
  • :compass: it even corrects itself: before a precise tap it wiggles the pointer, finds it, and fixes the miss

③ :brain: the brain

  • :robot: any OpenAI-compatible vision model — or the free built-in line (Jev, a “reflex” model that answers in ~70 ms)
  • :desert_island: the Dynamic Island narrates what it’s doing, with pause/stop right there
  • :speaking_head: it even hears the phone — AirPlay audio becomes words, so listening exercises work

④ :repeat_button: the replay

  • :clapper_board: every run replays step-by-step with the taps drawn on top — you can mark steps right/wrong and resume where it stopped
  • :high_voltage: save a run as an intent (like “Daily 7 AM alarm”) and replay it without AI when the screen still matches
  • :receipt: verified fun so far: unattended Duolingo lessons, drawing a smiley, texting Mom, Maps navigation — game auto-battling is not yet a thing (honest!)

▸ the $7-AI-grows-a-body arcade

  • :dog: ESP-Hi robot dog — a ~$7 brain with an emoji face; add servos and it walks
  • :bullseye: AI plays real pinball — ~$30 of electronics that watch the ball and physically flip
  • :desktop_computer: AirKey — the same chip becomes AI hands on your PC: “open Chrome” → real keystrokes
  • :camera: Lynx AI camera — ~$43 pan/tilt camera the AI aims, patrols and questions
  • :hibiscus: SenseCAP Watcher — $69: became a flower-shop QR cashier; at grandma’s it alerts on falls
  • :hatching_chick: Nanoid 2.0 — a CC0 “AI Tamagotchi” with moods, mischief and sleep modes

▸ one minimized window (click to open)

  • :satellite_antenna: no cable? no problem — the community’s iPhone Agent Harness drops the wire: phone stays on home Wi-Fi, a tiny $2–7 Bluetooth bridge sits anywhere in the room
  • :shopping_cart: a real shopping run — the AI Passport fork ordered coconut water on a delivery app: search → add → pop-up → paid, in 28 steps
  • :handshake: coding AIs can grab the phone — the esp32agent CLI lets Claude Code, Codex or Cursor treat your iPhone as a tool
  • :person_in_manual_wheelchair: the quiet superpower — hands-free phone control for people who can’t tap reliably; eye-gaze + agent combos already show 29–60% faster text entry in ALS studies

▸ the fine print, two lines

:mobile_phone: iPhone XS or newer, iOS 18+ · portrait only · ~3–5 sec per step · the phone runs on cellular while the board owns the Wi-Fi · App Store version still in review (free TestFlight now)

:locked_with_key: real talk — the free line sees unblurred screenshots: never for banking, passwords or private chats; screens can carry prompt-injection traps; unplug the board for sensitive stuff

Your phone always had a brain. Now it has hands.