Managing Datasets

Datasets group ground-truth items by modality and challenge type. A dataset can contain thousands of items. Multiple API keys can draw from the same dataset.

Create a dataset

hiveguard datasets create "My Dataset" --modality image

Options:

Flag	Type	Description
--modality	string	image, text, or audio (default: image)
--challenge-type	string	Default challenge type for items in this dataset
--prompt	string	Default prompt shown to solvers

List datasets

hiveguard datasets list

Output (table format):

ID                                    Name             Modality  Items
------------------------------------  ---------------  --------  -----
00000000-0000-0000-0000-000000000001  ImageNet Cats    image     2500
00000000-0000-0000-0000-000000000002  News Headlines   text      800

Inspect a dataset

hiveguard datasets show DATASET_ID

Shows item count, creation date, challenge type, and label statistics.

Export a dataset’s labels

hiveguard datasets export DATASET_ID --fmt csv --output labels.csv

Streams the export. Safe for large datasets. See Exporting Labels for format details.

Delete a dataset

hiveguard datasets delete DATASET_ID

You’ll be prompted to confirm:

Delete dataset "ImageNet Cats" (2500 items)? [y/N]: y
Deleted.

Deleting a dataset removes all its items and labels. This cannot be undone.

Attach a dataset to an API key

When creating an API key, you can restrict it to specific datasets:

hiveguard keys create "production-key" --dataset DATASET_ID

Requests with that key can only draw challenges from the allowed datasets.

Dashboard

All dataset operations are also available in the dashboard under the Datasets tab.