Right now, when test calls fail or error out, there’s no clean way to pull the data needed to debug — transcripts, agent responses, tool calls, latency, and error details are scattered or hard to extract in bulk.
Two asks that would meaningfully improve troubleshooting:
-
A filter on the test runs view for Failed and Error statuses
-
An Export action on that filtered view (CSV or JSON) that includes transcript, tool call payloads/responses, error messages, and timestamps
This becomes critical at scale. Once you’re running multiple agents across clients, manually clicking into each failed run to copy details isn’t sustainable — and the failed calls are exactly the ones we need to learn from fastest.
Happy to share specific examples from our agents if useful.