Skip to main content

Run a private browser test

A private browser test exercises the current saved voice-agent draft without requiring a published version or phone number.

Before starting​

Confirm:

  • the editor shows a successful save;
  • the browser can use the microphone;
  • the active workspace has usable credits and is not payment-locked;
  • attached integrations point to safe test destinations; and
  • the selected model, speech, voice, and knowledge configuration are ready.

The editor attempts to flush pending autosave changes before opening Test. If that save fails, do not treat the next result as a test of the visible editor state.

Start and run the test​

  1. Open the voice agent.
  2. Select Test.
  3. Allow microphone access when the browser asks.
  4. Start the conversation.
  5. Follow one planned scenario at a time.
  6. End the conversation and record the result.

The test uses the voice runtime without the telephone carrier path. It can display a live transcript and tool events.

What the test creates​

A private browser test creates stored operational records, including a test lead/contact representation and call record. Its transcript is saved when the session closes. The session is billed from its duration using the tested configuration's cost.

Browser tests do not currently run the live post-call AI analyzer, so the enriched summary or classified outcome expected from a live call can be absent.

Tool behavior​

Private authenticated tests allow attached and connected tools. Calendar events, leads, messages, webhooks, and other side effects can be real.

For every consequential action:

  1. Use safe test data.
  2. Verify the agent confirmed the inputs.
  3. Inspect the tool event or transcript.
  4. Verify the destination system directly.
  5. Clean up test records where appropriate.

Knowledge search remains a retrieval capability and should be evaluated for both relevant and missing results.

What a browser test cannot prove​

It does not prove:

  • phone-number assignment or inbound routing;
  • carrier audio transport;
  • caller ID behavior;
  • live concurrency admission;
  • published-version selection; or
  • live-only post-call enrichment.

After publication, place a new real call through the assigned number.