Audit and recovery
Follow changes and investigate incomplete operations.
Read the audit trail
The audit explorer records the actor, action, project, environment, time, and target. Record activity gives you the same context close to the record. Field changes depend on the environment's redaction settings.
Use environment, actor, and record filters to investigate an operation. A successful action result describes what the connector reported; external systems may have their own separate processing lifecycle.
Start from a record
Open a record and select Activity. After the connector delivers the event, a saved edit appears as record.update with the actor and changed field.

Investigate across the project
Select Audit log in the project sidebar. Use the filters to narrow the events, then expand a row with Show details. Check the event, target record, source, status, and request identifier together.

Choose a redaction mode
Full mode includes field values except always-redacted fields. Field-names-only mode records which fields changed without their values. None omits field changes. Secret-like field names are redacted, and you can configure additional fields.
Review data flow before enabling full values or streaming audit events to another destination.
Keep the outbox durable
The connector queues events locally and retries delivery when ingestion is unavailable. Persist its data directory and monitor backlog growth. The health response exposes queued events, pending mutations, and dead-letter events.
A write intent is saved before a governed mutation. If the connector crashes between the customer write and audit completion, recovery records an unknown outcome with identifiers for investigation. The customer database and audit outbox are separate stores, so this is not a single atomic transaction.
Reconcile before retrying
For an unknown mutation or an uncertain approval, inspect the target record and external side effects before repeating it. Retain the operation identifiers when escalating to your development team.
Inspect and replay rejected events
Rejected events remain in the connector's durable outbox. Heartbeats report them separately from the delivery backlog; the dashboard marks the connector's audit status as needing attention until the count returns to zero. Older connector versions do not report this count, so upgrade connectors before relying on central monitoring.
On the connector host, inspect up to 100 events at a time:
aviato audit --data-dir /var/lib/aviatoUse --payload only when you need the stored event body; it can contain values allowed by your redaction policy. Fix the rejection cause first. Replay selected event IDs with their original scope:
aviato audit --data-dir /var/lib/aviato --replay EVENT_ID --org-id WORKSPACE_ID --project-id PROJECT_ID --environment productionReplay moves the event back to the delivery queue without changing its ID or body. It does not rerun the original customer operation. Events from a reused data directory must be replayed by a connector for their original scope; the command refuses a mismatched scope. Back up the outbox before operational recovery.
Webhook delivery guarantees
Webhook and S3 streams leave out preview environments and members' development environments, whose events are kept for 7 days at most.
Webhook delivery is at least once. A pending batch is stored before sending; if the receiver accepted it but saving progress fails, retries retain the same payload and webhook-id. Receivers should deduplicate that ID. A renewable lease and fenced cursor updates prevent a stale worker from advancing progress; a network request already accepted by a receiver cannot be recalled.