Replay

Replay lets you re-execute a run from any step. This is useful for debugging, testing fixes, and reproducing customer-reported issues.

How replay works

Since every step is persisted with its full context (input, output, tool calls), Papayya can reconstruct the conversation history up to any step and resume execution from there.

Original run: step 0 → step 1 → step 2 → step 3 (error)
Replay from step 2: step 0 → step 1 → [resume] → step 2' → step 3' (success)

Steps before the replay point are reused as-is. Steps after the replay point are re-executed (and may produce different results if the agent's behavior or tools have changed).

When to use replay

• Debugging — a run failed at step 15. Replay from step 14 to see what happens with a fix applied.
• Reproducing issues — a customer reports unexpected output. Replay the exact run to observe the execution.
• Testing tool changes — you updated a tool's implementation. Replay a previous run to see if the new version produces better results.

Triggering a replay

Dashboard

On the run detail page, click Replay and select the step to replay from.

API

POST /v1/runs/{runId}/replay
{
  "from_step": 14
}

This creates a new run that inherits the original run's steps up to the specified point.

Replay creates a new run

Replay does not modify the original run. It creates a new run with a new ID. The original run remains unchanged for audit purposes.