AgentAudit is being built to record every tool call and human approval your AI agent makes — append-only and tamper-evident (hash-chained) — and produce a report your auditor accepts. Your debug log doesn’t do that; an enterprise security review asks for exactly this.
# one line — every tool call + approval → tamper-evident record from agentaudit import audit @audit def agent_action(...): ...
↑ Illustrative — the API isn’t final yet. This shows the intended developer experience.
Occasional updates about early access and pricing. We only use your email for launch news — ask us to delete it anytime.
Designed as an append-only, hash-chained record — who, what, when, which approval — in the format an auditor asks for. A debug log generally won’t pass a security review.
The goal: a ready answer to “does the agent have an audit trail?” — the security-review question that stalls deals.
EU AI Act Article 12 logging becomes required on 2 Dec 2027 (deferred via the Digital Omnibus). Built so you’re ready when it does — a bonus, not a scare tactic.
Three steps, framework-agnostic.
It’s designed to be tamper-evident, at the agent tool-call and approval level, with a retention policy and an exportable report. A plain debug log generally won’t satisfy an auditor.
Planned Python first (LangGraph and common agent SDKs), then JS/TS, with a framework-agnostic webhook option.
The plan is to let you choose self-hosted or hosted. Final architecture and data-handling details will be published before launch.
Article 12’s logging obligation applies from 2 Dec 2027 (deferred by the Digital Omnibus). We’re not manufacturing urgency — most teams want this today for SOC 2 and enterprise trust, and it also prepares you for 2027. This is informational, not legal advice.