On this page
Create and Use a Dataset Bundle
A Dataset Bundle is an explicit group of datasets approved for one app. Use a bundle when an AI-built application needs several related datasets and should have clear, limited data context.
Create a Dataset Bundle
- Open Dashboard → Data Catalog → New bundle, or choose Add to bundle on a dataset.
- Set the bundle name, description, and target Site or new-app target.
- Select eligible datasets and assign their dataset roles from the choices shown. These describe each dataset’s purpose; they are not member authorization roles.
- Review compatibility warnings and create the bundle.
- Open the bundle and inspect Selected datasets and Safe manifest preview.
- Select Build app with this bundle and use the copied context or prompt in your AI workflow.
Review the bundle manifest
The manifest describes approved datasets, schemas, access context, and intended roles without supplying private credentials. A bundle is not a copy of all the dataset rows and does not invent relationships between them. Explain valid join fields and business rules in your prompt.
Edit and manage a bundle
Use Edit bundle to rename or describe it. Browse datasets adds eligible datasets with roles; Remove removes a selected item. Use Disable and Enable to control bundle availability, or Delete when it is no longer needed.
Disabling or deleting a bundle is not a substitute for changing the underlying dataset access or unpublishing an app that already uses the data.
Example: a parks and inspections app
Bundle a parks geometry dataset and an inspections Table. Describe park_id as the join field and ask the AI to verify matching values before building a map and inspection list. Require loading, empty, and error states and preservation of the dataset access settings.
Resolve ineligible datasets
Owner-only, private read-key-protected, unavailable, or target-incompatible datasets may be blocked. Review Access settings on the dataset and grant only the intended access. Do not put private keys into the bundle or make sensitive data public merely to clear a warning.