Browser agentScreen-control agent

iPhone Agent Harness

Safari hosts the agent; a local WebRTC link sends actions to a Bluetooth ESP32 bridge.

What this project does

Screen observation and model interaction run in Safari. The browser validates plans and sends actions over local WebRTC; the board supplies Bluetooth input. Its clipboard helper uses Apple Shortcuts for longer text.

Limits and verification

Board support and OS requirements are the maintainer’s claims. Browser-local storage does not mean screenshots stay on the phone: configured model endpoints or an optional relay can receive context.

Reviewed sources

Reviewed README41c42d5
Hardware
ESP32 C2, C3, C5, C6, S3 or classic; board requirements vary
Runs on
Safari on iPhone / iPad
Screen / observation
Browser screen sharing
Input path
Bluetooth mouse and keyboard via WebRTC bridge
Setup requirements
iOS / iPadOS 27+; desktop Chrome or Edge for browser flashing
License information
MIT