It’s not a log.
It’s evidence.
AuditAgent keeps a record of everything your AI agents do, asks a person before the risky parts, and turns it all into evidence your auditors accept.
Sent “Your refund has been processed”. The customer's email address was hidden before anything was stored.
Asked to refund $1,240 on order 48213. Your policy sends refunds over $500 to a person. Maya approved it in Slack in two minutes.
Read one order from the database to answer a support question. Nothing was changed.
Moved lead 8f21 from “contacted” to “qualified” after they replied to step 3 of the sequence.
One decorator. Every action, on the record.
Put @audit.track above any function your agent calls. From then on, every time it runs, AuditAgent writes down what happened: what went in, what came out, and when. Your agent doesn't slow down; the bookkeeping happens in the background.
AuditAgentAPP9:40 AMBilling agent wants to refund $1,240 on order 48213. Your policy asks a person for refunds over $500.
Risky actions wait for a human.
You decide which actions need a person: refunds over a certain amount, anything that deletes data, whatever matters to you. Before your agent runs one, it stops and asks your team in Slack. One tap to approve or deny, and the agent carries on.
- Rules are written in plain text, not code
- Falls back to email if Slack isn't connected
- Every decision is recorded too: who, when, how long it took
Nothing can be changed after the fact.
Each record is linked to the one before it and checked every day. Nothing can be edited or deleted later, not by your agent, not by your team, not by us. That's what makes it evidence rather than a log.
- Personal details and secrets are removed before anything is stored
- Auditors can verify the record themselves
Yes. Every action an agent takes is written to a record that can’t be edited or deleted. Sensitive actions need a person’s approval before they run.
Security questionnaires, answered from your logs.
Upload a security questionnaire. AuditAgent finds the records that answer each question and writes a first draft, with sources. You read and approve every answer before it goes anywhere, then export as a Word document or spreadsheet.
MeridianSee what your agents did today.
A timeline of every action, the approvals waiting on your team, your rules, and how it all maps to SOC 2, in one place and always up to date.
Ask Claude, ChatGPT, or Grok about your audit trail.
Connect AuditAgent to Claude, ChatGPT, or Grok and just ask: anything waiting on me? What did the billing agent do last night? It reads the same record as the dashboard, and it can only read.
It’s not a log.
It’s evidence.
Free while in beta · works with any Python agent · set up in an afternoon