If you're building AI agents that need to survive a compliance audit, this gives you cryptographically verifiable decision logs without shipping data to a third party. Every action gets wrapped in an Ed25519-signed receipt with SHA-256 hash chaining, so auditors can prove nothing was tampered with after the fact. The ReceiptCollector API captures input, safety checks, output, and final action (allow/reject/escalate), while built-in PII redaction strips personal data before hashing. It maps directly to EU AI Act Article 12, ISO 42001, and NIST AI RMF controls, and exports self-contained evidence packages that verify offline. Reach for this when you're in fintech, healthtech, or any regulated space where "we logged it" isn't enough and you need tamper-evident proof.
claude mcp add --transport stdio sundsoffice-tech-ai-audit-trail uvx ai-audit-trail