Voice Agent documentation
A field guide to evaluating voice-agent conversations and running controlled tests.
Voice Agent is a workspace for evaluating recorded and generated voice-agent conversations against the standards your team cares about. It keeps each interaction, its evaluation context, and the resulting evidence close together so reviewers can move from what happened to a defensible conclusion.
The recording-evaluation journey
The core journey begins with a recording. Add or select the conversation you want to understand, choose the evaluation context that applies, and run the review. Voice Agent then presents the resulting observations and metrics beside the source material, giving reviewers a practical way to trace a result back to the conversation that produced it.
Use that loop to make review work more consistent:
- Start with a representative recording.
- Apply the relevant evaluation context and criteria.
- Review the evidence and results in the context of the conversation.
- Keep the evaluation current as metrics evolve, or create a fresh comparable run when the conversation itself needs to change.
Find the right guide
Start with the path that matches your goal:
Evaluate an existing call
Take the shortest path from an audio recording to evidence you can review.
Run a generated test
Choose a testing mode, create a scenario, and submit a controlled interaction.
Understand results
Move from suite-level patterns to the evidence behind one run.
Refresh an existing evaluation
Apply newly available or changed metrics to a saved call without repeating it.
Use the reference
Check current limits, statuses, troubleshooting guidance, and terminology.
Use core concepts and terminology when your team needs a shared language for discussing evaluations.
The guide stays close to the product: it explains what is visible in Voice Agent and how to use it without asking you to leave the evaluation workflow.