中文版

esp32-iphone-agent（GitHub：https://github.com/onehumanbeing/esp32-iphone-agent ，2026-10-11 创建，今晨抓取时已 4,061★）让 AI 模型直接操作一台真实的 iPhone——而整套硬件只是一块约 7 美元的 ESP32 开发板，不需要电脑、不需要越狱、不需要开发者账号、不需要 Mac。

新版的工作方式是：iPhone 通过 Wi-Fi 屏幕镜像把画面投给 ESP32-S3（它就是 AirPlay 接收器），再经蓝牙传回同一台 iPhone 上的 Agent 应用解码——视觉模型看懂画面后，ESP32 通过 USB HID 模拟鼠标和键盘把操作发回手机。项目给的例子是：让它“设一个周一、周二重复的早上 8 点闹钟”，它会自己离开 Agent 应用、打开时钟、拨动时间滚轮、勾选日期并保存。模型可以是任何 OpenAI 兼容的视觉模型。早期的 v0.1.0 还需要一台 Mac / Linux 电脑跑 UxPlay 和 Python 服务，新版已经把电脑去掉了。

为什么值得了解

• “让 AI 操作真机”此前通常意味着要一台 Mac、开发者账号或越狱——这个项目把门槛压到了一块开发板加一部 iPhone，思路很巧
• 对 iPhone 用户尤其直观：不用模拟器、不用云真机，就是你手里这台手机被 AI 操作，演示的闹钟任务也正是日常会用到的操作
• 提醒一句：这还是个早期项目，需要一块 ESP32-S3 开发板和一番动手配置，别把它当成开箱即用的 App；另外，交给 AI 的操作权限不小，先从闹钟这类无害任务试起

English Version

esp32-iphone-agent (GitHub: https://github.com/onehumanbeing/esp32-iphone-agent , created 2026-10-11, 4,061 stars at this morning’s fetch) lets an AI model operate a real iPhone — with hardware that is just one ESP32 board costing about $7. No computer, no jailbreak, no developer account, no Mac required.

How the new version works: the iPhone mirrors its screen over Wi-Fi to the ESP32-S3 (which acts as the AirPlay receiver), which passes it over Bluetooth back to the Agent app on the same iPhone for decoding — the vision model reads the screen, and the ESP32 sends actions back as USB HID mouse and keyboard input. The project’s example: ask it to “set an alarm for 8:00 AM that repeats every Monday and Tuesday”, and it leaves the Agent app on its own, opens Clock, spins the time wheel, ticks the days and saves. The model can be any OpenAI-compatible vision model. The earlier v0.1.0 still needed a Mac / Linux computer running UxPlay and a Python server; the new version removes the computer entirely.

Why it’s worth knowing

• “Let AI operate a real phone” used to mean a Mac, a developer account or a jailbreak — this project squeezes the requirement down to one dev board plus an iPhone, which is a clever bit of engineering
• Especially intuitive for iPhone users: no emulator, no cloud device farm — it is the phone in your hand being operated by AI, and the alarm demo is exactly the kind of everyday task you’d try first
• One caution: this is still an early project that needs an ESP32-S3 board and some hands-on setup, not an out-of-the-box app; and the control you’re handing to AI is substantial, so start with harmless tasks like the alarm