Exports
Four datasets, four formats. Exports are generated in the background; when the file is ready it appears in the list with its size and a download link.
Datasets
| Dataset | One row is | Use it for |
|---|---|---|
requests | One request | Full detail: timestamp, request id, model, task type and group, API key alias and last four characters, project, status, token counts, billable tasks, cost, latency |
daily | One day | Day, requests, cost — the series behind the daily chart |
models | One model family | Consumption and cost per model |
tasks | One task kind | What the work was, in the same vocabulary as the Usage screen |
The tasks dataset is offered when your organisation has the detailed task breakdown
enabled. Token columns likewise appear only where the contract includes them.
Formats
- CSV — the default for large volumes and for loading into your own tools.
- JSON — machine-readable, with the filters and period recorded in the metadata.
- XLSX — a workbook for manual analysis.
- PDF — a formatted report to attach to an invoice discussion.
The workbook and the report are unsuitable for hundreds of thousands of rows and are capped; CSV and JSON are streamed and have no such limit. When a heavy export is requested in a capped format, the cabinet says so rather than truncating silently.
Filters
Exports carry the same filters as the screen: period, model family, API key and project. The applied filters are written into the file metadata and into the file name, so a downloaded file is self-describing.
Retention
Generated files live for 24 hours and are then removed. Nothing is lost — regenerate the same export at any time. The list keeps the record of what was generated, by whom and how many times it was downloaded.