Storage ownership
Keep files in your own storage
Raw files stay in S3, GCS, Azure, or on-prem storage. Compute reads them directly and writes typed results to Dataset DB—without staging, ingestion pipelines, or duplicate copies to govern.
- Dataset DB stores file pointers, schemas, and derived columns. Each row references its source URI.
- Access only what you need: stream a clip, decode a slice, or read a column.
- Pay the egress costs of reading your files, without duplication.


