Open-AutoGLM

Why we included this project

Open-AutoGLM is an open phone agent that actually operates a phone for you. It combines a vision-language model with a control framework that reads the screen, plans the next action, and executes taps and inputs over ADB or HDC, so a plain-language request like "open the food delivery app and find nearby hotpot" turns into a real sequence of actions. The repo ships the AutoGLM-Phone-9B weights plus a multilingual variant, and it works on Android and HarmonyOS out of the box, with an iOS path through WebDriverAgent. It also handles the awkward parts: a confirmation step for sensitive actions, human takeover for login and captcha screens, and remote ADB over WiFi so you can drive a device from another machine. If you are building phone-use agents, UI automation, or a personal assistant demo, this is a working reference you can run and extend rather than a paper architecture.