Skip to main content
A dataset is items, optionally with events, frozen as an immutable snapshot. Two ways one shows up in your workspace. Bring your own. Import items and events to fit a tokenizer on them or to evaluate a model against your own held-out behavior. Start with a small JSON sample; the shape is on Data format. Licensed from Jean. Datasets Jean sourced or collected appear in your workspace once the agreement is in place. What they cover, how they were collected, and what you may do with them is settled in that agreement, not here. Talk to us about data.

​
Access

Each dataset records whether discovery, export, and model training are permitted. Public discovery does not imply redistribution or training rights. Exports are available only where your workspace has access. A dataset ID pins a version; changing the contents creates a new one. The sample import contract accepts up to 1,000 items, 5,000 events, and 1 MiB per request. Larger imports are arranged during onboarding.