Skip to main content
AI Agents

Test sets

Manage the test sets that validate the agent's behaviour before you publish.

Test sets

The Test sets page is where you build the scenarios that check whether the agent behaves the way you expect. Each set defines inputs, expected outputs, and whether the run should succeed or fail.

Reach it through the Test sets icon in the editor's side navigation.

The list of sets

The table shows every test set on the agent:

ColumnWhat it is
NameThe scenario's name ("Happy path", "Invalid input with no configuration")
Expected to succeedYes (green) — the agent should finish without escalating · No (red) — the agent should escalate or reject the input
InputsHow many input arguments the set configures
Expected outputsHow many output arguments it expects
Last runHow long ago it last ran ("2 days ago"), or if it never has

Click any row to open that set's editing panel.

Running every set

To run them all at once:

  1. Click Run (the play icon) in the editor's top bar.
  2. Choose Test sets in the menu.
  3. The Running the test sets panel opens with a confirmation:

    Every test in this agent's test sets will run. This may consume tokens and take a few minutes.

  4. Click Run to confirm.

While they run, each set reports its status live:

StatusWhat it means
WaitingQueued, not started yet
RunningIn progress
PassedFinished, and the result matched what was expected
Did not passFinished, but the result differed from what was expected
FailedSomething went wrong during the run

Click a result card for that run's details, trace and logs.

Removing a set

Click the bin icon on the set's row. A confirmation appears:

Remove this test set? This cannot be undone.

Confirm with Remove.