Events, audit and export
Guard keeps two records:
- Events: what the agents tried and what Guard decided. Executors send them.
- The control audit log: what people and tokens did to the project, such as publishing a policy, approving a ticket or revoking a connection.
Events
An action usually leaves several events, one per phase: proposed (with the decision), decided (allowed, denied, degraded, pending approval), started and completed (with the outcome). They share the action's run, trace and action IDs, so you can follow one call, or one whole task, from start to end.
Each event is redacted before it leaves the executor: names, decisions and digests, never argument values or results. See What leaves your environment.
Investigating
"Events" in the console opens the investigation view. Filter by agent, trusted subject, resource, run, policy version, decision and time range (with a time zone). Results come 100 to a page. Open an event to see the decision, the matching rule, any approval and up to 100 other events of the same run.
Retention
A project keeps events for 30 days, up to 100,000 events, by default. When the store is full, new events are refused (EVENT_CAPACITY_FULL) and executors keep them in their local queue. On the hosted control plane, tell us if you need a different limit or an export.
The control audit log
"Audit log" lists the latest 50 control operations: when, what (the method and path, and for a rollback the two versions and the reason), the result (allowed, refused or failed) and the role of whoever did it. Refused attempts are listed too, for example a viewer trying to retire an executor.
Operations our staff carry out on your project are in the same log, each with a reason reference.
Exporting to your log system
Two read-only feeds, each scoped to one project and paged with a cursor:
| Endpoint | Content |
|---|---|
GET /v1/events/feed | Events |
GET /v1/audit/feed | The control audit log |
Both take after (the cursor) and limit. A viewer token can read them. preworld-guard siem-export pulls both into an append-only NDJSON file and remembers where it stopped:
PREWORLD_GUARD_TOKEN="$GUARD_VIEWER_TOKEN" preworld-guard \
--endpoint https://guard.example.com --project-id "$PREWORLD_GUARD_PROJECT_ID" \
siem-export --output /var/lib/preworld-guard-siem/events.ndjson \
--state /var/lib/preworld-guard-siem/cursor.json- It writes and syncs the new lines first, then saves the cursor. A crash between the two can repeat a few lines: deduplicate on
kind+project_id+record_id. - Run one export at a time.
- It doesn't push to any particular SIEM. Point your existing log collector at the NDJSON file.
--limit(default 500) and--max-pages(default 20) bound one run.
GET /v1/assets summarizes the agents, tools, destinations and enforcement points seen in events. It can only show actions that went through Guard. Check your tool, MCP and network configuration for anything that never did.
Handling exports
Exports hold no argument values or results, but resource names, destinations, agent IDs and subject IDs can still be sensitive. Protect the files and their backups as your log policy requires.