Prove what your AI agents actually did.
Cryptographically verifiable receipts for delegated AI work. The agent signs. The witness signs. The human signs. The verifier runs offline.
Reference implementation · local verifier · patent pending.
A delegated agent workflow, verified.
What a receipt can and cannot prove.
Sequesign shows what is verified, what is attested, and what remains only agent-asserted. The distinction is the product.
What the receipt can prove
- Evidence has not changed after receipt construction
- Action sequence and chain continuity are intact
- Agent and witness signatures verify
- Schema validation passed, when requested
- Workflow profile validation passed, when requested
- Human approval signature is present and valid
What a receipt does not prove
- The agent told the truth
- The invoice came from the real vendor
- The payment provider confirmed the instruction
- The LLM provider attested to the response
- A lying agent self-reported every unsupported claim
- Every natural-language claim is externally verified
How the protocol works.
The demo is invoice work. The implementation is general: any agent action, any schema, any workflow profile, any verifiability class.
One protocol. Multiple delegated workflows.
The same sequence-signing receipt pattern can be used across delegated agent workflows. The underlying receipt pattern applies wherever agents act on behalf of people or organizations.
Agent marketplaces
Require receipts before accepting completed delegated work or releasing payment.
Design targetRegulated trading
Record what an agent checked, decided, and executed for later audit review.
Future profileHealthcare workflows
Separate agent assertions from signed human approvals and external system evidence.
High-trust use caseContract and claim review
Verify which documents, schemas, approvals, and claims produced an agent output.
Profile candidatePatent pending.
Sequesign technology is patent pending. This demo explains protocol concepts and is not a production deployment.