Run history is where you find out what your Agent actually did. Chats, test chats, scheduled runs, webhooks, and replies it gave in Slack and elsewhere all land here.
Evaluations are the one thing that doesn't. Those replies live in the report on the evaluation run that produced them.
Go to Agent → open an Agent → Monitor → Runs.
How do I find a run?
Each run is one row: its status, its title, and when it started. The newest run is at the top, and the list updates on its own as runs start and finish.
- Search: type in Search runs… to find runs by title, or by the message that started them: what you typed in a chat, or what a trigger sent, such as a ticket's subject line. Webhook and integration events are also found by what was sent to them, even when no run started. A run found by its message shows the matching text under its title.
- Filter: click the filter icon beside the search box, then choose what to keep.
| Filter | Keeps |
|---|---|
| Trigger | Runs started a certain way: Manual, Scheduled, Webhook, Integration, Channel, or API. Pick more than one. |
| Review | Runs your team marked Approved or Rejected, or runs that are Not reviewed |
The number beside Runs counts the runs that match. A number on the filter icon shows how many filters are on. Choose Clear filters in the same menu to see every run again.
Hover a run to see more without opening it: its full title and status, when it started, what triggered it, how long it took, what it cost, and how it was reviewed. The same facts are in its Details once it's open.
What can I see in a run?
A run opens with its title and status at the top, next to Create test case and a ... menu. Under them are two tabs, and the Details panel sits beside the run.
| Part | What it shows |
|---|---|
| Run | The messages in and out |
| Trace | Every step the Agent took, including which model it used and which tools it called |
| Details | Its review, when it started, what triggered it, how long it took, its spend and tokens, and which version answered |
| Scores | Grades from any scorer that graded this run, and why |
| Tools | Which tools it used, and how often |
Scores and Tools are in the Details panel, under its list. To give the run more room, click the Hide details icon at the top of the panel.
Runs started by a webhook also show what was sent to them.
The version in Details opens the exact Agent that answered, and tells you if it's changed since.
How do I mark a run as good or bad?
Open the run's Details → Review and choose Approved or Rejected. Then click Add note in the same menu to say why. Choose Not reviewed to clear it.
There's one review per run, shared with your team, and the most recent edit wins. Auto-improve reads these reviews, so saying what went wrong is worth the ten seconds.
How do I turn a run into a test case?
Open the run and click Create test case at the top, then choose which dataset it goes in. Real runs make the best test cases, because they're the questions your Agent actually gets asked.
Runs you've already saved are marked in the list.
How do I grade a run after the fact?
Open the run and click ... → Run scorer, then pick a scorer. A run with no grades yet also offers Run a scorer under Scores. Only scorers that don't need an expected answer can grade real runs.
Can I reply to a run?
Yes, when it's a chat you started. Open it in Runs and type in the message box under the conversation, the same as on the Agent's Chat tab.
Runs started by a trigger, and chats your teammates started, are read-only.
Additional FAQ
How do I start a run myself?
Start a chat from the Agent's Chat tab. To try changes you haven't published, use the test chat, which opens from the bar above any Build page. Run history doesn't start new runs, but you can keep replying to a chat you started.
A run stopped early. Why?
The status beside its title tells you: Completed, Failed, Stopped, Out of credits, Stopped at its spend limit, or Stopped at its time limit. The last two mean it hit a run limit.
Can I see what a run cost?
Yes, under Spend in the run's Details. For totals across every run, see Analytics.
How do I delete a run?
Open the run and click ... → Delete run. Or, in the list, hover the run and click ... at the end of its row, or right-click it, then choose Delete. Its messages and trace are removed, and this cannot be undone.
How far back does history go?
The list loads more as you scroll, so you can keep going back through older runs.