Runs and Reports
Every time a job runs, it leaves a run: how it went, how long it took, what it cost, what it saved, and the agent's report.
The jobs list
AI Studio → Jobs shows every job in a row:
| Column | Shows |
|---|---|
| Enabled | The on/off switch. |
| Job | The name, what it runs (Prompt or the workflow) and its agent. Under them: › Last report, which opens the last run's report right in the list — shown only when that run left one — and Artifacts: N, which goes straight to the job's artifacts. |
| Trigger | When it runs, in words, with a chip saying how it was set: Manual, Schedule or Cron. |
| Last run | When, and how it ended: Completed, Failed, Cancelled — or Queued and Running while it goes. |
| Next run | When the next scheduled run is due. |
| Owner | Who owns the job. |
| … | The job's menu: Run now, Edit, Subscribers, Reset artifacts, Delete. |
Opening the last report or the artifacts from the list does not open the job; clicking anywhere else on the row does.
A job's page
Three tabs:
- Overview — the Brief · Prompt switch, and the facts: what it runs, the trigger, the agent, the owner, the subscribers, the last and next run.
- Runs — every run, newest first.
- Artifacts — what the job keeps; see Artifacts.
A run, in one row
On the Runs tab each run reads on one line:
| Part | Shows |
|---|---|
| Status | Completed, Failed, Cancelled, Queued or Running |
| Started by | The schedule, or the person who pressed Run now |
| Duration · when | "Duration 13m 28s · 2h ago" |
| › Report | Opens the agent's report in place |
| Artifacts | "Artifacts: 1 created · 2 updated" — only when the run changed any |
| Tokens | Output, input at full price, and the share read from the provider's cache |
A run that failed says why on one line, with Show more for the whole message. Opening a run shows its report in full, the artifacts it touched, and — while it is queued or running — Cancel run.
Reading a report
The report is the agent's account of the run. A good job asks for a fixed shape: what is new first, then what changed and what was archived, then what could not be read. Links in a report point at the job's artifacts and open them. How to ask for that shape: Writing a Job That Works.
The Runs page
AI Studio → Runs lists everything agents are doing for jobs across the workspace:
- Active — running now, and waiting. A waiting run says its place in the queue and why it waits: behind other work, or no runner available.
- Finished — newest first.
Filter by agent or by source. Each run reads exactly as on its job's Runs tab, with › Report in place.
Notifications
When a run finishes or fails, the job's owner and its subscribers hear about it in the Inbox and by email. The notification shows the agent that ran it and links to the run.
What a run costs
The tokens a run uses are billed by the agent's provider on the agent's key — or, for a trial agent, on XPT's. A first run, which builds everything from nothing, costs the most. Later runs read what is already recorded and only check for changes, so they are much cheaper: in our own monitoring job, the second run used about a third of the first run's input.
A run of a harness agent has the limits of a scheduled job on its runner. On XPT's runner that is 2 CPUs, 4 GB of memory, 16 GB of scratch disk, and it is stopped after 30 minutes of silence. Long work is fine as long as the agent keeps working.