Skip to main contentVaultGraph is designed to be a trust and verification layer for AI agents. The goal is not only to measure outcomes, but to make outcomes auditable.
High-Level Architecture
Participants
- Vendor: the AI agent provider (the entity that runs and signs receipts)
- Consumer: the organization using an agent fleet (the entity that audits receipts)
- VaultGraph Platform: infrastructure that ingests receipts, stores them, computes scores, and supports audits
JobReceipt
A JobReceipt is the core unit of proof. It represents a completed AI job (chat, task, action) and is intended to be:
- Signed by the vendor
- Verifiable by the consumer
- Privacy-preserving, so sensitive context does not need to be public
The exact schema and signing format are planned and will be documented once the SDK/API are implemented.
Attestation (Optional)
VaultGraph may optionally publish verifiable proofs of work completion (e.g., attestations) so third parties can validate that some work occurred, without exposing private job context.
Trust Scoring
VaultGraph computes vendor and agent-level trust summaries derived from receipt aggregates.
Audit & Dispute
Consumers are able to:
- Access receipts for their own jobs only
- Verify authenticity and integrity
- Export proofs for internal review
