AgentAudit OS
Compliance observability platform for AI agents. 3-line SDK integration captures every AI decision with tamper-evident logs (cryptographic hash chain). Multi-framework compliance: EU AI Act, NIST AI RMF, ISO 42001, SR 11-7, HIPAA. Enterprise SIEM (Splunk, Sentinel, QRadar) and SSO (SAML/OIDC/SCIM via WorkOS).
Datadog for AI agents — but the auditor cares.
The problem
The EU AI Act, NIST AI RMF, and ISO 42001 all require auditable AI decision logs, but the agents themselves emit nothing useful. Existing observability tools (Datadog, New Relic) capture spans and metrics — they don't capture *the decision* in a form a regulator can read. Companies deploying AI agents in regulated sectors are flying blind on compliance.
The approach
Built a 3-line SDK that wraps any LLM or agent call and emits a structured decision record: inputs, prompt version, model, output, tool calls, scoring, latency. Records are written to an append-only Postgres table protected by a cryptographic hash chain — any tamper to a historical record breaks the chain and is detected immediately. Records are then mapped to controls across EU AI Act / NIST AI RMF / ISO 42001 / SR 11-7 / HIPAA so a compliance officer can answer 'show me every decision your AI made about this user last quarter' with one query.
Tech decisions
Outcomes
- 3-line SDK integration — instrumenting an existing agent takes minutes
- Tamper-evident audit log with cryptographic hash chain
- Out-of-the-box mapping to EU AI Act, NIST AI RMF, ISO 42001, SR 11-7, HIPAA controls
- SIEM export to Splunk, Sentinel, QRadar
- Enterprise auth (SAML / OIDC / SCIM) via WorkOS
What I learned
Compliance is a UX problem disguised as a regulation problem — the auditor doesn't need new data, they need the existing data in a shape their tooling can read. The hash chain was the cheapest 'we won't tamper with our own logs' guarantee we could offer; full blockchain was overkill.