Cua's integration with HUD, which Cua's announcement describes as an evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → platform for computer-use agents, lets developers test different model configurations on the same desktop tasks. That includes composite agents, which divide planning and visual targeting between models. The announcement shows how to select a configuration by changing the model string and run either an individual task or a dataset.
The integration also exposes screenshots and action traces during a run. Developers can watch each click, keystroke and screenshot, then try another configuration on the same task. The announcement documents the setup and tracing; it does not report results for any configuration.
Terms in this piece · Glossary
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
