Skip to main content
The Evaluation tab lets you measure how well your agent’s prompt performs before you rely on it with real callers. It lives on the agent page, alongside Conversation, Capabilities, Behaviour, and After the call.

​
When to use it

Use Evaluation whenever you change an agent’s prompt or want to confirm it still behaves the way you intend. Instead of judging a single test call, an evaluation runs your agent across a set of representative conversations and scores the results, so you can see patterns rather than one-off impressions.

​
Run an evaluation

  1. Open the agent and select the Evaluation tab.
  2. Select Run Eval.
  3. The evaluation runs in the background and updates automatically as it progresses — you can stay on the tab and watch it complete.

​
Review your results

Once the evaluation finishes, the tab shows you:
  • Summary cards with the agent’s overall scores. These cover both quality (how well the agent does its job) and conduct (whether it stays within the boundaries you expect).
  • A per-conversation breakdown. Expand any conversation to read the full turn-by-turn transcript and see the metric scores for that exchange.
Use the summary cards to spot whether a prompt change helped or hurt overall, then open individual conversations to understand why a score moved.

​
If you cannot run an evaluation yet

Evaluation needs a prepared set of conversations to score your agent against. If that set is not ready for your agent, the tab tells you so and the Run Eval button stays unavailable until it is in place. If you would like evaluations enabled for an agent, contact support with your workspace ID and the agent name.