ZJU-REAL/Awesome-GUI-Agents

What it solves

This project compiles the latest research, tools, and benchmarks on GUI agents — AI systems that operate computer interfaces like a human by seeing the screen and clicking or typing. It helps researchers and developers keep track of a fast-moving field without digging through scattered sources.

How it works

The repository organizes papers and resources into themed sections: perception, exploration, planning, and interaction. It provides a chronological list of technical reports, papers on online reinforcement learning, GUI grounding benchmarks, memory mechanisms, and more. It also includes weekly paper summaries, links to open-source frameworks like ClawGUI, and analyses of conference papers (AAAI, ICLR). Entries are updated frequently and are accompanied by direct links to arXiv, GitHub, and model pages.

Who it’s for

Primarily researchers and engineers working on GUI automation, computer-use agents, or multimodal AI systems. It is also useful for anyone wanting an overview of current approaches in building agents that interact with graphical user interfaces.

Highlights

• Curated lists of papers, datasets, and benchmarks, all sorted by topic and date
• Weekly summaries of new research and conference-specific analyses (AAAI 2026, ICLR 2026)
• Direct links to open-source projects, arXiv papers, and Hugging Face models (e.g., GUI-G2, ClawGUI)
• A modular breakdown of GUI agents into four core capabilities: perception, exploration, planning, and interaction