esp32-iphone-agent

Free: The iOS app and firmware are MIT, except the AirPlay receiver code in hardware/hid_bridge/src/airplay/ (ported from UxPlay), which is GPL-3.0, so the flashed firmware binary is GPL-3.0. No paid tier, no account. The app you build from the main branch has no analytics and no developer server (the log uploader ships with no destination). The project site's privacy policy describes a coming store build with opt-in PostHog usage stats and an optional free "Handsy line" model run by the developer, which keeps each step's screenshot and your task; neither is in the main branch yet.

Self-hosted: Hardware is one Seeed Studio XIAO ESP32S3 (plain version, 8 MB PSRAM; not Sense, not Plus), $7.49 on Seeed's store; any ESP32-S3 with native USB and PSRAM should work. The iPhone needs iOS 18 or later and cellular data; Lightning iPhones need a Lightning to USB Camera Adapter. Flash the board with arduino-cli (ESP32 core 3.3.x, NimBLE-Arduino 2.5.x) or the browser installer at handsy.tech (firmware v0.2.1, built from a commit not published on GitHub). Build the app in Xcode on a Mac and sign it yourself: a free Apple ID works with a 7-day re-sign. The one-tap hotspot join needs the Hotspot Configuration capability (a paid Apple developer account); without it you type the hotspot password by hand.

Paid: No paid tier. The running cost is the vision model: any OpenAI-compatible API that accepts images, billed by whichever provider you point it at, or your own server (vLLM, SGLang, llama.cpp, Ollama). Every step sends a screenshot.