Open-Source Tool

A board that costs about $7 lets an AI agent use a real iPhone. Small app studios could hire it to test every release.

A board that costs about $7 now lets an AI agent see and tap a real iPhone, with no jailbreak and no computer once it's set up. Henry Yuan's open-source ESP32 iPhone Agent turns a cheap real-device tester into a weekend project, and small app studios ship releases without one.

Published Oct 21, 2026

The Wow

What someone built

The ESP32 iPhone Agent mirrors an iPhone's screen to a small ESP32-S3 board through Screen Mirroring and uses the same board as a USB mouse and keyboard with AssistiveTouch on. A companion app sends screenshots to any OpenAI-compatible vision model; a planner and an actor carry out the steps, and every run is kept with its taps and swipes. The README shows it setting a repeating alarm in the Clock app. The repo had about 3,460 GitHub stars on Oct 21, 2026.

Open-Source Tool

License
MIT, except the AirPlay receiver code ported from UxPlay, which is GPL-3.0, so the firmware binary is GPL-3.0 (per the README)
What you need
  • An ESP32-S3 board with native USB and PSRAM; the README was tested on a Seeed Studio XIAO ESP32S3 (about $7)
  • An iPhone on iOS 18 or later with cellular data (Lightning models need a Lightning to USB Camera Adapter)
  • A Mac with Xcode, once, to flash the board and build the app (a free Apple ID can sign it)
  • Any OpenAI-compatible vision model API, cloud or self-hosted
Install
Flash the board with the repo's flash script, build the iOS app in Xcode, then follow the four in-app setup steps (pair, mirror, pointer, model), per the README.
A business you could build on it
Real-iPhone release checks for small app studios

Why it's remarkable

It gives an AI model eyes and hands on the most locked-down consumer device there is, using nothing but ordinary accessibility features, and leaves a replayable trace of every run.

The Opportunity

The business it reveals

Plain-English release checks on a real phone. Indie iOS developers and small agencies often ship updates without anyone walking through the app on a device, because UI tests take time and device clouds cost money every month.

Who would pay

Indie iOS developers and small agencies shipping client apps through TestFlight. Real-device testing is already a paid category (device clouds and QA freelancers).

The Flip

2 ways you could turn this into a business

Fresh step-by-step help guides for iPhone apps

Who pays
Support and docs teams at small software companies with an iOS app
How it would earn
A fee per guide, plus a monthly refresh when the app changes (our hypothesis)
Why it might work
Every run keeps screenshots and actions that can be exported, which is the raw material of a how-to article.
What already exists
Guide-capture tools such as Scribe and Tango focus on desktop and web; many teams still take phone screenshots by hand.

Why we'd try "Real-iPhone release checks for small app studios" first: The buyer already knows the pain, the hardware costs almost nothing, and a free five-task trial on their own build is an easy yes.

Build It

How you could build it

A per-release check: a task list from the release notes, real-iPhone runs, exported replays, and a one-page result.

First step: Order a XIAO ESP32S3, follow the README to set it up on your own iPhone, and run three tasks in an app you use every day. Time each one.

Agents and tools

  • ESP32 iPhone Agent
  • XIAO ESP32S3 board
  • A spare iPhone on iOS 18+
  • A vision model endpoint
  • Your own AI agent to write tasks and reports

Setup

  1. Flash the board and build the app once on a Mac.
  2. Dedicate a spare iPhone with cellular data to testing.
  3. Write a reusable checklist of core journeys (sign up, log in, main action, settings).
  4. Do two practice runs on public apps to learn the timing and failure modes.
  5. Make a one-page report template that links each replay.

What the agent does

  • Turns release notes into plain-English tasks
  • Runs each task on the phone
  • Keeps a replay of every step
  • Drafts the failure summary

What you do

  • Agrees the tasks with the client
  • Watches runs and stops them when needed
  • Decides which failures are real bugs
  • Sends the report and handles test accounts

Your first customer: An indie developer you know, offered five free tasks on their current TestFlight build.

Test demand first: Offer three indie developers a free five-task check. Success: at least one finds a real bug in the replays and asks for the next release. Failure: none of them do.

Your starter instructions

Paste this into the AI agent you use (ChatGPT, Claude, Gemini, Grok Bot or another) as your first message. Copying it starts work in your own agent; it doesn't create or deploy a business by itself.

Get a launch plan

Running it

What stays human: Writing the tasks with the client; watching the run and keeping the stop button close; deciding which failures are real bugs; writing the report; handling test accounts.

Startup cost (estimate)
About $7 for the board, plus an iPhone and a vision-model plan (According to the README, the tested board costs about $7. You need an iPhone on iOS 18+ and a Mac with Xcode once for setup. Model costs per run are not measured)
Running cost
Model calls per task and minutes of supervision per release are unknown until measured; the README says each step takes a few seconds, mostly model time.
How much the agent runs
People run it; AI helps with tasks.

Pricing your own version? Use the free startup-cost worksheet.

The evidence

What the source shows: The project's README, limitations and license (MIT/GPL-3.0). We read them; we haven't run a client test.

What's ours: Per-release smoke tests for small studios are our application of the kit.

Worth planning around

  • Slow by design: the mirrored screen is small (480×1040, 5 fps or less) and each step takes a few seconds.
  • iOS doesn't mirror passcode entry, Apple Pay or DRM video, so those flows can't be checked.
  • The companion app uses a private iOS API, fine for a self-built app but not for the App Store.
  • Mature tools (device clouds, XCUITest, Maestro) go deeper; this wins on setup effort.

Sources

Sources reopened Oct 21, 2026. Spot a mistake? Tell us and we'll correct it in public.

Keep reading

The Daily Paycheck: one remarkable AI business at a time, and exactly how you could build around it. Free; unsubscribe anytime.

Get new AI business discoveries by email.