Run your tool with test values to see it work before agents use it. The debugger executes the flow step by step and shows what every step produced.
Starting a run
Press the play button at the bottom of the page and fill in the tool's parameters. Autofill parameters fills them with AI-generated values based on their descriptions. Run & Debug starts the run; if it is disabled, a required parameter is missing or its value has the wrong format.
To start partway through the flow, open a step's ⋯ menu and choose Run from here.
During the run
Steps execute one by one and the focus follows along. When a step fails, the run stops there.
Select any executed step and inspect it:
-
Step output: what the step produced, or the error if it failed.
-
Collected data: the value of every parameter and variable at that point in the run. You can edit it (keep it valid JSON) and restart to test different data.
-
Context: the conversation context values the run is using. Use Restart with different context to test other ones.
-
Media: files involved in the run, when there are any.
Use the arrow buttons to move through the executed steps and watch the collected data change.
When the run reaches an Ask user step, you answer it yourself: pick a quick reply or type a response.
Restart runs the tool again with the same values. Stop and play starts fresh with new ones.
Editing while testing
You can change a step between runs. Close the step before restarting so the debugger picks up the change. The debugger follows steps you add or remove; if the view looks out of sync, restart the run.