Train and evaluate agents on real computer-use tasks.
Cua has 4K+ environments and 400K+ tasks spanning desktop apps, cross-app workflows, browser surfaces, and generated state variations. Linux, Windows, and macOS images are available.
Pick a domain or scope a cross-app workflow.
The public catalog groups the available inventory by workflow. We share environment-level availability and task specifications during scoping.
LibreOffice, Joplin, Zotero, Thunderbird, Calibre
ERPNext, Odoo, Dolibarr, Nextcloud, Mattermost
Blender, GIMP, Krita, Inkscape, Audacity
VS Code, Bruno, DBeaver, Postman, Insomnia
Avogadro2, IQmol, QMForge, QGIS, R
Open-source HTML5 games, browser puzzles, and simulations
One environment from each core domain.
Start with a compact sample, inspect the task format, then scope the applications and difficulty mix required for your training or evaluation target.
Environment, task, and verification artifacts for each approved scope.
Tell us which applications, operating systems, task families, and verification methods you need. We will return a scoped catalog and delivery plan.