Datasets

Datasets on this API are primarily custom CSV-backed resources: rows (for example prompt, output, golden answer), an optional description, and an optional default_selected_metrics set. Metrics on the dataset are not required. A run from the dataset may send its own metric set or reuse the dataset default, so the same dataset can be evaluated with different metrics. Custom datasets also support optional extra columns — user-defined free-text fields you can add to the dataset and fill per record. You can create, list, update, delete, and download custom datasets. Proprietary catalog datasets are readable: pass their dataset_type_id on the paginated list, or fetch / download by id. Create, update, delete, and extra-column management remain custom-only; the partial list excludes proprietary rows.

Base path: /datasets. All routes require a Bearer API key.

To resolve dataset_type_id for the paginated list endpoint, use Dataset types.

Shared HTTP semantics: Introduction — status codes and errors.

Endpoints

• Create dataset — POST /datasets
• Get datasets — GET /datasets/all-partial, GET /datasets (paginated), GET /datasets/{dataset_id}
• Update dataset — PUT /datasets/{dataset_id}
• Delete dataset — DELETE /datasets/{dataset_id}
• Download dataset — GET /datasets/{dataset_id}/download
• Extra columns (custom datasets only)
  • Create extra column — POST /datasets/{dataset_id}/extra-columns
  • Update extra column — PUT /datasets/{dataset_id}/extra-columns/{column_id}
  • Reorder extra columns — PUT /datasets/{dataset_id}/extra-columns
  • Delete extra column — DELETE /datasets/{dataset_id}/extra-columns/{column_id}
  • Set record extra values — PUT /datasets/{dataset_id}/records/{record_id}/extra-values