AgentAudit: Immutable AI Audit Trail

AgentAudit solves the critical gap in AI agent governance: enterprises run agents in production today — updating CRMs, generating reports, sending emails — but when audits or incidents occur, there's zero proof of what happened, when, or why. Our solution: every agent action passes through Lobster Trap (deep prompt inspection + policy enforcement), generates a SHA-256 hash, and is permanently recorded on Base L2 blockchain. The result is an immutable, verifiable audit trail that compliance officers can present to regulators immediately. The architecture: 1. Agent action → Lobster Trap :8080 (policy evaluation: ALLOW/DENY/LOG) 2. If ALLOW → LLM processes → audit.log (JSON) 3. AgentAudit Bridge → SHA-256 hash → Base L2 Sepolia blockchain 4. Supabase stores metadata + blockchain transaction hash 5. Next.js dashboard shows real-time events with VerifyBadge (click hash → see on-chain proof) Why it matters: 80% of enterprises already use AI agents. 0% have immutable audit trails. This is the missing infrastructure. Technical stack: Lobster Trap (Go), Base L2 Sepolia, Supabase, Next.js, Solidity smart contract, n8n orchestration, Node.js bridge. We're Diego Rivera (Vision4AI), implementing agents for LATAM enterprises. Built AgentAudit because we needed it first. Every AI action. Immutably verified.