Reproduce “CUA-Lite: An Open Platform for Computer Use Agents” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Zhanhui Zhou, Weichen Zhang, Haoran Liu, Lingjie Chen, Tianneng Shi, Kevin Lin, Zhengyuan Yang, Lijuan Wang and 1 more

CUA-Lite provides shared interfaces for desktop, browser, and mobile computer-use agents, connecting environments, trajectory data, and model-specific harnesses for evaluation, supervised fine-tuning, and reinforcement learning. Its Docker-based Linux desktop sandbox used 0.6 GB resident memory versus OSWorld’s 3.8 GB; scores differed by 2.7 points on average across 13 models. Training gains were shown on offline browser mirrors, not live websites.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.