Native phone appScreen-control agent

C3 Agent

An on-phone AutoGLM-inspired task loop with ESP32-C3 Bluetooth input.

What this project does

The current C3 Agent keeps screen observation, context and model calls in the iPhone app. It adapts Open-AutoGLM’s action loop and uses a separate visual model for pointer localization and result checks.

Limits and verification

The maintainer reports unresolved pointer-localization failures and no stable acceptance of complete tasks. C3 has no native USB HID; old AirPlay / Python architecture belongs to a legacy tag.

Reviewed sources

Reviewed READMEb2ad856
Hardware
ESP32-C3; USB is for power and flashing
Runs on
Native iPhone app
Screen / observation
ScreenCaptureKit with explicit sharing consent
Input path
Bluetooth HID mouse and keyboard
Setup requirements
iOS 27+, Xcode 27+, signing and ESP-IDF 6.0.1
License information
MIT; Open-AutoGLM components Apache-2.0