Backed by: @pithy-sh/audit, and the control plane’s connection listing. Scopes: audit:events:read · audit:events:read_detail · connections:read. Plan: Team.
What it answers
Who did what, when and to whom: your audit trail searched by every filter its listing takes, combined, over any range, with each event’s detail one click away and an agent’s acts told apart from a person’s.
The audit trail is the pane that shows the newest events on every plan, querying the trail is the listing this searches, and retention is how long your Worker keeps what it searches.
The explorer sits in the rail’s Compliance group. Every search is your Worker’s own GET /audit/events, sent from your browser with the listing’s query, and the answer is dropped with the screen, like every other pane. Nothing from it is written to our database, cache or logs.
The search
| Filter | How you set it |
|---|---|
| Actor type, client kind, outcome, severity | Chosen from the values the trail records |
| Actor id, action, resource type, resource id, tenant, project, environment, worker | Typed |
| Range | All they keep, the last 24 hours, 7, 30, 90 or 365 days, or between two times |
Every filter goes to your Worker as the listing’s own query, checked against it before it is sent, and nothing is filtered on our side. A value the trail cannot be searched by is refused before the call, naming the field. Results are newest first, and Older and Newer follow the listing’s keyset cursor.
Tenant has a choice of its own, No tenant, which asks for the events that belong to no tenant, rather than leaving the filter off.
The window is yours
The explorer states how long your Worker keeps events, read from the effective retention on its manifest: so many days, or no end date. A range is sent as you asked for it, however far back, and no plan shortens it. The search reaches whatever your Worker still holds.
A Worker on a kit too old to report its retention is named as one that does not say, and the search still reaches whatever it holds.
Each event’s detail, on its own scope
A row opens GET /audit/events/:eventId: the IP, the user agent and the metadata. It needs audit:events:read_detail and nothing else.
When the connection does not hold that scope, every row is still drawn in full, with no way to open its detail, and the explorer names the scope. Scopes is how to add it.
Agents told apart
A row an agent wrote reads as an agent acting for somebody, never as the person alone. There are two kinds:
- A control-plane connection of kind
agent: a management client acting on your Worker. Its row names the principal it acts for, and the connection it acted through once its detail is open. - A tool call inside your app: a user row of kind
agent. The OAuth client is named from the event’s detail. The name is the client’s own and is not verified; the client id after it is.
An agent’s row leads to the member who answers for it, on their member record, where the connection can be revoked. The explorer itself acts on nothing.
On Free and Solo
The explorer is in the rail, locked, saying what Team adds, and reads nothing until the plan holds it. The audit trail keeps showing the newest events on every plan. Plans and pricing has what each plan holds.