What gets sealed
The platform or customer system supplies the record as hashes and metadata. Lumra never receives the audio or the transcript.
- Transcript SHA-256 and audio SHA-256, sealed into the signed record
- Call duration, platform, and call identifiers
- The moment the record was sealed, and by which key
- AI participation and disclosure flags, as asserted by the source
- Signing status: cloud-signed by Lumra at intake, or source-signed by the platform and verified when provided
- Exported as a ProofPack anyone can verify offline, no PriviNet servers involved
What this proves, and what it does not
Proves
The sealed conversation record and its artifact hashes are unaltered since intake. When the record was sealed, and by which key. Whether the record is cloud-signed or source-signed, and if source-signed, that the signature was verified and preserved.
Does not prove
What was actually said. Transcript fidelity to the audio. Speaker identity, or that a speaker was human. Consent or disclosure legality. Admissibility, which is always a decision for the court.
Sign. Chain. Verify.
Break it yourself
Below is a real sample pack generated by the live engine from a synthetic conversation. Open it, or download the machine JSON, change a single character of the transcript hash, and run the standalone verifier. Verification fails. Change it back, it passes. That is the whole product.
For voice platforms and carriers
The strongest version of this record is the one your platform signs at the source. Lumra verifies and preserves that signature, and every ProofPack then carries your attestation, checkable by anyone with open tools. If you run voice or AI-agent infrastructure and want your signature to carry independent weight in a dispute, that is the version worth building together.
brad@privinet.net