FaceTime audio, agent video, and caller vision: what to test

Start with the direction of each track

For an audio conversation, sound must travel from the caller to the agent and from the agent to the caller. A connected room does not prove either path. For a video experience, distinguish video the agent publishes from video the caller sends.

Chert’s Bring Your Own Room documentation describes two-way audio, agent video, and separately enabled caller-camera capture. Treat those as individual acceptance requirements. Your application may need only audio, an agent face, camera input, or all of them.

Test audio as a conversation

Ask a question whose answer cannot be a generic greeting. Follow it with a related question. Confirm that the caller hears a relevant reply, that the agent hears the actual words, and that interruption behaves as expected for your agent implementation.

If the call is silent, inspect the agent process as well as the bridge. Check model credentials and provider credits, caller mute state, track publication, and application errors. Repeatedly dialing does not isolate the cause.

Test agent video as changing output

Use a video-enabled line and a compatible agent. Confirm the person calling sees the expected video and that it changes during the session. A poster image, frozen frame, or earlier demo recording cannot establish live media flow.

The supplied video starter publishes an agent face, but that does not automatically add vision input. Follow the current video guide for the starter’s capabilities and dependencies.

Test caller vision with a new scene

First confirm that caller-camera capture is enabled for the assigned line. Then confirm your agent actually consumes that input. Show a new object or change the scene after the call begins, and ask a specific question about what changed.

Use a test whose answer is not already in the prompt or conversation. An agent that correctly repeats a description you just spoke has not demonstrated visual understanding. Record the media direction, agent configuration, test conditions, and result when qualifying a support workflow.

Check lifecycle and call direction

Verify the inbound and outbound paths you intend to use independently. Outbound audio or video requires an enabled, ready line and the documented media settings. Keep the agent silent until the bridge is ready so the greeting is not lost before the caller can hear it.

End the call from the caller’s device, wait for cleanup, and start a second call. Confirm that the previous conversation is not accidentally reused and that the line becomes available again. Check call duration and concurrency against your plan.

Write an honest acceptance record

Keep a small matrix: caller hears agent; agent hears caller; caller sees changing agent video; agent responds to changing caller view; resources released; second call succeeds. Mark irrelevant directions as not required and untested directions as untested.

Use that record to decide which experience to expose in your product. A voice-only assistant can be useful without video. A visual support flow should stay limited to the line and configuration whose camera path you have actually verified.