Skip to main content

Group Voice Lab

Group Voice Lab is an unlisted internal prototype for testing the raw LiveKit + Gemini multi-participant audio path. It is not the supported maturity-linked group-session workflow.

Reach it only by its direct route:

/app/voice-lab

For normal collaboration, use Group sessions and Group voice.

Who should use it?​

Only authorised testers who understand that the Lab:

  • is unlisted and English-only;
  • creates an AI live-audio session;
  • consumes monthly AI budget;
  • depends on deployed LiveKit and voice-agent infrastructure;
  • does not attach its conversation to a normal company maturity session;
  • may change or be removed without notice.

The current lab-token path permits a signed-in user with budget capacity to open their own isolated lab room. The hidden URL is not a substitute for an administrator-only access policy, so do not distribute it broadly.

Open a lab room​

  1. Navigate to /app/voice-lab.
  2. Enter a room label.
  3. Select Open.
  4. In the Voice Room panel, join the room.
  5. Grant microphone permission.
  6. Wait for the AI agent to become available.

Room labels are normalised to the permitted lab namespace. The server also namespaces a lab room to the signed-in user, preventing unrelated accounts from colliding on the same simple label.

To test two browser clients, use authorised sessions belonging to the same test identity and room context. Do not use real customer conversations.

Change rooms​

Select Change room, enter another label and open it. Leave the previous voice room before switching.

What this Lab validates​

  • LiveKit token issuance;
  • browser microphone/media connection;
  • multi-client audio transport for the same isolated test context;
  • voice agent joining;
  • Gemini Live audio response;
  • basic transcript/room panel behaviour.

What it does not validate​

  • multiplayer session invitations or presence;
  • participant authorisation for a real session;
  • shared maturity updates;
  • task generation;
  • group-session completion and summaries;
  • company knowledge integration;
  • normal feature-flag discoverability.

Run those checks in an actual gated group session.

Privacy and safety​

  • Use synthetic test content.
  • Do not speak credentials, personal data or confidential company strategy.
  • Ensure everyone in the room knows audio is being processed by the AI service.
  • Stop and leave after the test.
  • Record only the minimum diagnostic metadata required.

Troubleshooting​

“LiveKit not configured”​

The required voice infrastructure is absent from that environment. Stop testing and ask the deployment owner to verify LiveKit and agent configuration.

Microphone is unavailable​

Check browser/site permission, operating-system input selection and whether another app has exclusive access.

The agent does not join​

The browser room can connect even when the separate voice agent is not deployed or healthy. Capture environment, time and room label, then check the agent service.

Budget error​

Lab rooms are subject to the signed-in user’s enforced monthly budget. A System Admin must review limits; do not disable budgets merely to make a prototype test pass.

Another account cannot join the same simple label​

That is intentional. Lab rooms are namespaced per user for tenant isolation.

Last updated on