Activity is the audit trail. Every turn you run, every workflow execution, and every automation firing is logged here with its status, how long it took, and what it cost in credits. When you need to know what actually happened — not what you remember asking for — this is the page.
What gets logged
Three kinds of work land here, in one timeline:
- Turns — requests you made to the operator directly.
- Workflow runs — every execution of a workflow, whether you triggered it or a schedule did.
- Automations — standing automations, each time they fire.
Every entry carries the same three facts: status (finished, failed, or still running), duration, and credit cost. That consistency is what makes the page scannable — one glance gives you the health and the price of a stretch of work.
A stretch of the log reads something like this:
completed workflow · competitor-summary 2m 41s $0.18
failed automation · inbox-rules 0m 12s $0.01
running turn · "tidy the pricing page" 4m 03s —Filtering
The list filters by workspace, type, and status, and the filters combine. Some useful shapes:
- Workspace + failed — "is anything broken in the client project?"
- Type: workflow, this week — "did the scheduled runs all go out?"
- Status: running — "what is the agent doing right now?"
Click through to the trail
Each row is a summary; the substance is one click deeper. Open any entry and you get the full trail of that run — for a workflow, step-by-step results for every node; for a turn, the plan, the branches, and what each produced. It is the same trail described in The Operator, reached from the log instead of the thread.
A worked example
Monday morning, credits look lower than expected. Open Activity, filter to last week, and scan the cost column — one Thursday workflow run cost several times its usual amount. Click through: the trail shows the second node pulled a far larger dataset than normal, and every following step paid for it. You now know the cause (an upstream data change), the fix (constrain that node's input), and that nothing else was responsible. About two minutes, no guesswork.
Edge notes
- Failed runs stay in the log with their trail intact — failure is recorded, not swept away. Diagnose from the trail, fix, and re-run.
- Every run is logged and stays clickable, so last month's "what happened here?" is still answerable next month.
- Activity is the record, not the control panel. To change a workflow or automation, follow the entry back to Workflows or the automation that owns it.
Good to know
- Who can see it? Collaborators working in a workspace see its activity, and viewers can look without spending credits — see Team.
- Where do the costs come from? Each entry's credit cost is the metered compute and model time of that run — see Billing and Usage.
- Why keep checking it? You mostly should not need to. Activity earns its place on the days something looks off — a spend spike, a missing result, a failed schedule — and on those days it answers in minutes.