# AgentNet

22.6K human-annotated desktop computer-use tasks on Windows, macOS and Ubuntu, released with OpenCUA for training and evaluating vision-language-action models.

- Catalog URL: https://awesomecua.github.io/resources/url-964d70c40bfc5ca8e1ee/
- Original source: https://huggingface.co/datasets/xlangai/AgentNet
- Kind: dataset
- Review: curated
- Scope: core
- Environments: desktop
- Topics: data-generation
- Evidence checked: 2026-10-16T06:38:26Z
- ID: url:964d70c40bfc5ca8e1ee

## Limitations

Inclusion does not establish reproducibility or endorsement.

- Record JSON: https://awesomecua.github.io/resources/url-964d70c40bfc5ca8e1ee/record.json

## Structured profile

- formats: JSONL, image archives
- modalities: screenshots, instructions, actions, generated reasoning
- splits: not recorded
- supervision: computer-use trajectories, synthesized reasoning

### Count observations

- ≈ 22,600 tasks — AgentNet collection (reported; source: https://huggingface.co/datasets/xlangai/AgentNet/raw/main/README.md)

The card distinguishes human task instructions from generated summaries, reasoning, reflections, and quality fields. Generated labels are not independent human quality acceptance records.

## License declarations

- code: not recorded
- data: mit — https://huggingface.co/datasets/xlangai/AgentNet
- document: not recorded
- weights: not recorded

## Primary evidence

- https://huggingface.co/api/datasets/xlangai/AgentNet (retrieved 2026-10-16T06:38:26Z; SHA-256 f23c588ea9fb27210f9d605e164a32527a24298ca516df98272ff7ca7110eac6)
- https://huggingface.co/datasets/xlangai/AgentNet/raw/main/README.md (retrieved 2026-10-16T11:40:46Z; SHA-256 327a77856df259db0a0c2405478f75a0d0f2bf17c51806abc7c509ad82c3ad57)
- https://huggingface.co/api/datasets/xlangai/AgentNet (retrieved 2026-10-16T06:38:26Z; SHA-256 f23c588ea9fb27210f9d605e164a32527a24298ca516df98272ff7ca7110eac6)