How to Test LiveKit Agents With TestMu AI

LiveKit is infrastructure rather than a hosted builder. The agent joins a WebRTC room as a participant, and phone calls arrive in that same room through a SIP trunk. The TestMu AI Agent Testing Platform joins the room or dials the number and runs your LiveKit agent through multi-turn conversations across personas and edge cases, scoring each run so you can test it without touching the infrastructure.

To connect one, you need a running LiveKit agent (self-hosted or on LiveKit Cloud), a SIP trunk with a provider such as Twilio or Telnyx for phone agents only, source docs for scenario generation (prompt, PRD, or knowledge base), and a TestMu AI workspace with agent-testing permissions.

Why do LiveKit failures often sit in the SIP trunk, not the agent?

LiveKit is infrastructure, so the unit under test is a room, not a hosted agent ID. That shapes how you connect and where failures sit:

Room, not an agent ID. The unit under test is a room, so setup differs from every hosted platform.
SIP REFER for transfers. SIP REFER must be enabled on the trunk for transfers, and Twilio additionally needs PSTN transfer switched on.
Connectors over WebSocket. Connectors bridge Twilio calls over WebSocket without a SIP trunk, which is a separate connection path.
Trunk-level settings. DTMF, cold transfer, and region pinning are trunk-level settings, so a failure may sit in the trunk rather than the agent.
No text channel. There is no text channel to test, so this page covers phone and voice only.

Do you reach a LiveKit agent by room or by number?

Phone covers inbound and outbound callers arriving over a SIP trunk: DTMF entry, SIP REFER transfers, voicemail handling, and carrier latency. Setup: phone agent testing.
Voice reaches the agent by joining a room directly rather than over a phone number, using WebRTC against the room. It isolates the speech pipeline, so STT and TTS accuracy, turn-taking, and interruption handling get tested without telephony noise. Setup: voice agent testing.