Audit trail

How do you create an audit trail for AI agent actions?

Send one event for every action your agent takes: the tool it called, the input, the decision and result, tagged with a run ID and timestamp. Store them in order, outside the agent, so you can review any run later.

AgentTracer is not tamper-evident and it is not compliance-grade. It stores the JSON you POST, in order. It does not score output quality, and it does not replay a run. Retention tops out at the Team window.

What to log, and the three signals to watch, are on what are the best practices for AI agent observability? A single weird step is on how do I debug my AI agent when it does weird things?.

Steps

  1. Keep the record outside the agent. POST each action to /api/v1/events. AgentTracer is not a proxy, so a model call does not pass through it. The copy you send is the copy you can read later. The agent should not keep the only record.
  2. Send one event per action. A tool call, a decision, a result, and an error are separate events. Put the input in the payload. Recommended types are tool_call, decision, message, error, and span. Other type strings are stored. Up to 100 events can ride in one request.
  3. Tag the run and the time. Set run.id and an ISO timestamp on each event. When the run ends, set run.status to completed, failed, or cancelled. The project timeline shows the run id under the name. It does not show status as its own column. The status is stored on the run.
  4. Leave secrets and personal data out. AgentTracer stores the JSON you send and does not redact it. Do not put secrets, raw credentials, or personal data in the payload. The request body must stay under 1,000,000 characters.
  5. Read one run from the timeline. The project timeline lists the newest 100 events, newest first, with the time, the type, the name, the run id, and the payload. A payload longer than 180 characters is cut off on that screen. There is no separate page for one run, and there is no filter. The run id on the row is how you keep one conversation together.

This record is not tamper-evident and it is not compliance-grade. Free keeps events for 14 days, Pro for 30 days, and Team for 90 days. Retention tops out at 90 days.

curl -sS -X POST https://agenttracer.net/api/v1/events \
  -H "Authorization: Bearer $AGENTTRACER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"agent":{"id":"support-bot"},"run":{"id":"run_01h","status":"completed"},"events":[{"type":"tool_call","name":"search_docs","timestamp":"2026-10-29T00:12:00.120Z","payload":{"query":"refund policy"}},{"type":"decision","name":"cite_policy","timestamp":"2026-10-29T00:12:00.880Z","payload":{"choice":"cite_policy"}}]}'

Checklist

Audit questions and the event fields that answer them
Audit questionEvent fieldWhere you read it
Which tool ran?events name, with type tool_call.The name column on the project timeline.
What was the input?events payload.The payload column. A value longer than 180 characters is cut off on that screen.
What did it decide?type decision, plus the name and the payload.A decision row on the project timeline.
How did the run end?run.status: running, completed, failed, or cancelled.Stored on the run. The timeline shows the run id under the name, not a status column.
When did it happen?events timestamp, as an ISO time with a timezone.The time column, shown in UTC.

Questions

Is an AgentTracer audit trail tamper-evident or compliance-grade?

No. AgentTracer is not tamper-evident and it is not compliance-grade. It stores the JSON you send, in order, outside the agent. It does not score output quality, it does not replay a run, and it has no loop detector.

How long does the audit trail last?

Free keeps events for 14 days, Pro for 30 days, and Team for 90 days. Retention tops out at 90 days. The windows are on the pricing page.

What should the audit events leave out?

Leave out secrets, raw credentials, and personal data. AgentTracer stores the payload you send and does not redact it. The request body must stay under 1,000,000 characters, and one request can carry up to 100 events.

Do I need a special SDK to record the trail?

No. One JSON POST to /api/v1/events is enough. OpenTelemetry is not required. The same habits are on What are the best practices for AI agent observability?