Voice Agent
Overview

How Voice Agent evaluation works

Follow a conversation from the agent you want to assess to evidence you can review.

Voice Agent helps teams assess conversations against the standards that matter to them. The product keeps the source conversation, the evaluation context, and the resulting evidence together so a score can be traced back to what happened in the call.

The journey

Most work follows this sequence:

  1. Choose a workspace.
  2. Register the Agent Under Test (AUT) — the voice agent you want to assess.
  3. Provide a recording or create a scenario for a new test.
  4. Voice Agent creates a test suite containing one or more test runs.
  5. Review scores, metric evidence, the transcript, and the audio as results become available.

An evaluation is the interpretation of a run through its metrics and evidence. It is not a separate item you need to create or manage.

The two sides of a conversation

The AUT is the agent being assessed. The Test Agent (TA) is the simulated counterparty used for a generated test.

  • When the AUT represents the business, the TA plays the customer.
  • When the AUT represents the customer, the TA plays the business.

This distinction keeps scenarios clear: define what the other side of the conversation should do, then evaluate how the AUT responds.

Suites and runs

A test suite is the evaluation group you see in Results. A test run is one physical execution or one uploaded-recording evaluation inside that group.

One requested test can create several runs. For example, a selection of driving metrics or optional robustness and bias variants may need separate calls. Compatible passive metrics can still be evaluated from the same call, so more metrics do not always mean more recordings.

Results arrive progressively

You can open a suite while it is still running. Its recordings, statuses, and score information appear as processing advances. Open a recording from the suite to inspect the individual run in detail.

Start with a real call

For a first successful evaluation, use an existing recording. It gives you a quick path from a conversation your team knows to an explainable result.

Follow the quickstart or learn the core concepts.

On this page