33 upvotes
New framework lets computer-use agents improve via open-world loops
What is CUA-Master: Scaling Computer-use Agents through Open-World Recursive Improvements?
CUA-Master is a training pipeline that lets AI agents that operate computers learn from real interaction data instead of only self-generated examples. It adds an iterative feedback step where each generation of the agent is evaluated in open environments and the results are fed back to refine the next version.
Who it is for
Developers of autonomous software assistants should watch it because it offers a way to get more diverse, controllable, and reliable agent behavior without hand-curating data.
How to try it
Open the Hugging Face paper page for the code and usage instructions