Verify
Drop in any file. Dara extracts the embedded manifest, checks its canonical integrity, and compares the bytes against the trusted published record.
PublicNO PROVIDER CALL
Live on Backblaze B2Make the work.
Make the work.
Keep the record.
The control plane for AI-generated media: governed pipelines, verifiable provenance, and a spend ledger that counts what never shipped.
Published SHA-256 · whole fileTrusted match
efaf24d3c4cbeeb2497acd5fcba1e485be529a0ece944190c4caef8720244c25
Genblaze's source hash and the delivered file's hash are deliberately different — embedding changes the bytes. Dara records both and never pretends they should match.
How it works
Govern.Generate.Account.
Generation tooling is excellent. The operational layer under it did not exist — so Dara makes provenance and cost part of the media supply chain.
1
Policy clears the brief
Cost is estimated from the model registry and checked against budget before a provider is ever called. A run that would overspend is rejected at zero cost.
2
The pipeline runs under watch
Every step streams. Fallback routes, QA revisions, and failed attempts are all preserved and linked by parent run, so the version tree shows the real history.
3
The record outlives the file
The manifest is embedded, both hashes are indexed in B2, and immutable Parquet accounts for the work — including the attempts that never shipped.
One bucket, distinct rolesBackblaze B2
LEDGER
Immutable Parquet, queried in place by DuckDB
See the numbersWhat Dara does
Answers you can hand to a client.
Govern
Typed policies enforced at four points — pre-flight, before every provider step, after QA, and after embedding but before publication.
Pre-spendBLOCKED AT ZERO COST
Account
Immutable per-run Parquet in B2, queried in place by DuckDB. Failed, rejected, and policy-blocked work stays visible.
HonestCOST PER APPROVED ASSET
Keep the
receipts.
No account required. The Studio demo replays a committed corpus at zero cost, and Verify never contacts a provider.