Replay re-executes a response ClaudIA already gave, using the current knowledge and configuration, and shows the result side by side with the original response. It's a way to check the effect of a change before trusting it.
Replay is only available to users with the Administrator or CX Engineer role, and only on responses generated by ClaudIA's AI. If you don't see the replay button on a ClaudIA response, your user most likely doesn't have that role — talk to an administrator of your account.
When to use
- After adjusting the knowledge base, to see whether it would change a response ClaudIA already gave.
- After changing an instruction or configuration in ClaudIA, to validate the effect in a real conversation.
- To understand why ClaudIA answered a certain way — which knowledge base articles it used.
Step-by-step
Step 1 — Open the conversation
Open the conversation with the ClaudIA response you want to re-execute.
Step 2 — Trigger the replay
Hover over the ClaudIA response and click the replay button (circular arrow icon, "Replay ClaudIA response"). The Replay ClaudIA response panel opens over the conversation.
Step 3 — Run the replay
In the panel, click Run replay. ClaudIA re-executes the response live, from scratch, with the current knowledge. It takes a few moments.
Step 4 — Compare the results
When it finishes, the panel shows three things:
- The verdict at the top: Identical responses or Different responses.
- The Context for the responses: which knowledge base articles were used in the original response and in the replay — and whether they changed.
- The Original response and the Replay response (now), side by side.
If the response changed, compare the knowledge base articles of the two versions to identify what changed.
Replay is only a simulation for comparison: it does not change the conversation or the original response the customer received.
Notes
- Replay is not available for responses generated by multiple agents (a supervisor that triggers specialist agents). The panel warns you when that's the case.
- Responses that didn't come from the AI agent — such as Controlled Flows or manual responses — don't have replay, because there's no agent execution to reproduce.