Debug with AI
Use a run's failure explanation to identify the step that needs attention. Debug with AI can add a diagnosis and suggest a draft correction. Free, Pro Lite, and Pro can request this assistance with sufficient GNG credits.
Before you begin
- Open the failed run from Runs.
- Read its failure explanation and inspect the first failed step's inputs.
- Keep GNG credits available. Diagnosis is metered separately from the original run.
Steps
Review the explanation
Read what went wrong and the suggested next action. Use Fix this node or Open the node if the correction is already clear.
Request a diagnosis
Select Debug with AI. Wait for AI diagnosis and the likely cause to appear.
Review the proposal
Read the diagnosis before choosing an action. If a correction is available, select Apply suggested fix. Otherwise, use Open the node to edit the identified step.
Inspect the draft
Review the affected step and its connections in the editor. An applied suggestion changes the draft; it does not publish automatically.
Test the correction
Select Test and check the result. Confirm that required data reaches the step and that any intended output is produced.
Save the reviewed version
Return to the editor and select Save. Automatic runs then use the corrected published version when the workflow is active.
Outputs
The diagnosis explains the likely cause and may supply a draft edit. Common corrections include replacing a missing source, selecting an available model, choosing the right field, or completing an output's required settings.
A successful but unhelpful AI result can need a data correction rather than an error fix. If the step ran on condition results alone, select the earlier research through Choose specific fields.
An older run can refer to a step you removed later. The editor shows a notice when that step no longer exists. Review the current workflow before applying an old diagnosis.