Apps · PROJECT NOTES
WorldGUI
Desktop and web workflow benchmark data with varied starting states, plus a public Windows agent implementation for planning and self-correction.
What can you do with it?
Compare how agents recover when documents, settings or interfaces are already partway through a workflow. Inspect instructions, videos, project-file references and pre-actions across ten office, system, development and media applications.
What should you know before using it?
The current metadata has 611 task-condition records: 111 meta records and 500 augmented starting-state variants, rather than 611 independent workflow goals. Public agent code and data do not establish a packaged reset-and-score harness for the full corpus; the agent's self-critique is not independent benchmark ground truth. Setup uses an installed Windows desktop, model APIs and, for the documented parser, Google Cloud Vision. Application licenses and state preparation remain operator responsibilities. The dataset card declares Apache-2.0; no repository-wide code license was found.
What is verified here?
Runtime: The documented GUI-Thinker/WorldGUI-Agent setup targets Windows and Python 3.11, with model credentials and optional parser services. A frontend portability statement does not establish Android or iOS benchmark tasks.
License note: The WorldGUI-Bench dataset card declares Apache-2.0. No repository-wide code LICENSE was present at the inspected WorldGUI revision.
This entry was reviewed against original project material. We have not independently run the software or reproduced its results. AI-development claims are author-reported; check upstream for current versions, setup requirements, and licenses.
- Resource type
- Dataset + benchmark
- Provenance
- Development provenance not established
- Reviewed
- Platforms
- Windows, Web