Audit log

#How audit logs work

Each entry records an event, the time it happened and its outcome. Entries are appended in order and kept for later review. A useful log is complete for the events it covers, easy to read by the person who needs it, and kept apart from the system it describes, so a failure in that system does not erase the record.

#Why it matters for AI agents

An agent acts on your behalf, often in places you do not watch as it happens. A log lets you see what it did after a run. It helps you find a mistake, check a suspicious action or show what changed on a site. Without a log, you rely on the agent's own summary of what it did.

#Wield and audit logs

Wield logs every request on the user's computer. Each entry holds the method, host, path, status and timing. Request and response bodies are never logged, and neither are query values. The logs are not sent to wield.run. Telemetry is a separate setting, off by default. If it is turned on, it sends only recipe_id, version, ok, duration_ms and error_code. Paid plans add audit export for teams, so an organization can review its members' activity.

#Related terms

• Human-in-the-loop: the approval step that decides whether an action runs at all.
• Least privilege: the limit that keeps each logged action within a narrow scope.
• Security model: what Wield keeps on the user's computer and what it sends.

#FAQ

#What does Wield log?

For each request, Wield logs the method, the host, the path, the response status and the timing. It never logs request or response bodies, header values or query values. The log stays on the user's computer.

#Are logs sent to wield.run?

No. The log is written on the user's computer only. wield.run sees downloads of public recipes and anything the user chooses to send, such as a site request. Telemetry is off by default, and when it is on, it sends only the five fields listed above.