Audit log
The Audit log is the per-event record of activity in your tenant. Every action that matters — a schema saved, a connection added, an API key minted, a member invited, a sign-in attempt — lands here with the actor, the request path, the outcome, and a timestamp.
It’s the page you reach for when something unexpected happens and you need to know what changed. It’s also the page security and compliance teams export from during an audit cycle.
What’s audited
Events are grouped into categories, which is also how you filter them:
Governance. Schemas uploaded, edited, deleted or bound to a connection. API keys created and revoked. Members invited and invites accepted.
Security. Sign-ins and sign-outs, cube access denied by an ACL, rate limits exceeded, and connection tests blocked as SSRF attempts.
Connections. Warehouses created, deleted, and credentials rotated. Connection tests, both passing and failing. File datasources created and reaped.
Files. Uploads, downloads and deletions, plus the profiling runs that read a warehouse or file to draft a cube.
Agent activity. Every query against a cube — the path, the key or user that issued it, the duration, the status — split by whether it arrived over MCP, direct API, or schema discovery. On a busy tenant this is most of the log.
System. Tenant purges.
What’s not logged either way: query result contents (we never log the rows you fetched), per-cell access patterns, or your clicks in the Saiku UI. For the full data-handling picture see Tenant isolation.
Filtering
The log opens on the last 24 hours. Widen or narrow it with the filter bar:
- From / To — the date range.
- Category — Governance, Security, Connections, Files, Agent activity, System, or Other. The fastest way to cut a noisy log down to the class of event you’re chasing.
- Actor — the user or API key responsible.
- API key id — narrow to one specific integration.
- Outcome — success or failure.
Set what you want, then click Apply filters. Nothing re-queries as you type, which is what keeps a busy log usable. Filters land in the URL, so a particular view is something you can bookmark or paste to a colleague.
Each row shows the time, the actor, the event and its category, the target, and the outcome. Click one and a details panel opens underneath the table, carrying the event kind, the actor’s display name and email, the request ID, and then the raw event summary — HTTP method, path, status code and latency.
That request ID is the useful part: it’s the same one threaded through the server logs, so it’s what turns “something failed at 14:32” into a specific trace. Close dismisses the panel.
CSV export
Download CSV exports the currently-filtered set. Useful for offline analysis, for attaching to a SOC 2 readiness package, or for feeding your own SIEM. There’s no export-specific limit beyond your account’s overall rate limit — pull as often as you need.
Retention
Your window is a ceiling, not a floor: events older than it are deleted by a daily sweep, automatically, on every tier. We don’t hold onto audit history indefinitely and we don’t ask you to file a request to have it removed.
| Tier | Window | Changeable? |
|---|---|---|
| Starter | 30 days | Fixed |
| Team | 90 days | Fixed |
| Business | 7 days – 7 years (1 year default) | Yes, self-serve |
| Enterprise | 7 days – 7 years (1 year default) | Yes, self-serve |
On Business and Enterprise you set it yourself — pick a number between 7 days and 7 years and it applies from the next sweep. Make it shorter and the events past the new window go at the next sweep; make it longer and nothing that has already been swept comes back. Changing the window is itself an audited event, so your compliance trail records who chose it and when.
Shortening a window deletes data. If a control or regulator requires you to keep something for a fixed period, set the window above that period and let your archive take the rest — export on a schedule into S3, GCS or your SIEM. Deleting your tenant purges everything, audit history included; see Account.
Two practical notes:
- The sweep is daily, not instant. An event that crosses your window mid-afternoon is deleted the following morning, so treat the window as accurate to within a day.
- Sweeping the index doesn’t rewrite history elsewhere. Events you exported earlier still live in your own archive, under your own retention policy.
Common investigations
A few patterns we see often:
“Why is this cube broken?” Filter to the Governance category. Schema edits and uploads live there, so the most recent one tells you when the definition last moved.
“What has this contractor been doing?” Put their name or email in the Actor box and skim. Worth doing before you revoke anything, so you know what you’re about to interrupt.
“Is this API key compromised?” Put the key’s UUID in the API key id box and narrow the range to the last day. You’re looking for paths it has no business touching, error spikes, or volume out of its usual pattern.
“Did somebody get refused?” Filter to Security. Denied cube access and rate-limit rejections both land there — the two events that mean somebody asked for something they weren’t allowed.
Related
- Members — actors you’ll see on this page.
- API keys — actors you’ll see on this page.
- Tenant isolation — what we log, what we deliberately don’t.