- Hash-chained ledger events — every row cryptographically links to the prior row
- Captures prompt, retrieved context, tool calls, and resulting journal entry per action
- Append-only: no UPDATE or DELETE permitted on the audit table at the DB level
- One-call proof export: produce a signed bundle for any transaction ID
- SOC 2 / ISO 27001 ready schema with role-based redaction for sensitive fields
- Tamper detection: nightly verifier surfaces any broken chain within minutes
- Replay support: reconstruct the exact agent reasoning that produced any entry
Agent Audit Trail: Immutable Provenance Log for Autonomous AI Transactions
Auditors do not accept "the model decided." They want the prompt, the retrieved context, the policy that fired, and the ledger entry that resulted — all linked and tamper-evident. An agent audit trail stores each agent action as a hash-chained record: tool call, prompt hash, retrieved doc IDs, the journal entry produced, and the cryptographic link to the prior block. Deleting a row breaks the chain and is detectable on the first verification.
Lock down an agent audit trail
AgentBooks is a double-entry ledger API built for autonomous systems. Join the waitlist for early access.
You're on the list.