Runs

Each row below is a real captured agent run, not a mockup — every model call, tool call, and reasoning step, held exactly as the agent produced it. Open one to see the full context window at the step that mattered, replay it from that exact point, and check the signed, hash-chained record against the runback.cassette/v1 spec yourself — no account required.

The error rows aren't failures we hid — some are a policy gate blocking a step before it reached a customer, the same mechanism walked in the incident guide; others are a genuine downstream failure, like a declined card. New to this? Start with the guide →