MobileAgent
GUI agent models and frameworks that operate phones, desktops, and browsers
What it is
A family of research projects from Alibaba's Tongyi Lab for AI agents that read screenshots and operate graphical interfaces on Android, desktop, and the web. It includes the GUI-Owl vision-language model weights in several sizes and the Mobile-Agent multi-agent framework, which adds planning, reflection, progress tracking, and memory on top of them. Researchers and developers use the code to run agents on real devices, reproduce results on benchmarks such as AndroidWorld and OSWorld, and study training methods like reinforcement learning for GUI automation.