Audit log
Every change to your project, who made it, and whether it worked — including the attempts that did not.
Every state-changing action is recorded. Not a sample, not the interesting ones: a mutation that writes no audit row is treated as a bug in the platform, and there is a check that fails the build over it.
Your project’s audit view is usually the fastest answer to “what happened to this”, and it is faster than asking.
What a row tells you
- Who — the account, and the IP the request came from.
- What — the action, named after the operation rather than the endpoint.
- On what — the project, workload, namespace or key it touched.
- When.
- Whether it worked. Rows carry a status, and failures are recorded too.
That last one is the part people do not expect. A refused action leaves a trail: a deploy rejected for quota, a login with the wrong password, an API key reaching for something outside its scopes. If you are trying to work out whether somebody tried something, absence of a success is not the same as absence of an attempt, and this is where the difference shows.
Finding things
Filter by actor, action, status and date, or search. The most useful habit is filtering to failures over a window — that is where misconfigured automation and expired credentials surface, usually before anyone reports them.
Credentials are named
An action taken with an API key records the key, not just its owner. So when something unexpected appears under your name, the log distinguishes “you, in the console” from “the CI key on the build server” — which is the difference between changing your password and revoking one credential.
What it is not
It records actions against the platform: deploys, stops, member changes, key mints, bindings. It is not an application log. What your container printed is in its logs, and what a request did inside your own service is yours to record.
Audit rows are retained on a schedule your operator sets, and they are evidence rather than telemetry — nothing edits a row after it is written.