A browser call proves that your agent can join an audio session. It does not prove that a customer's phone number reaches that session, or that audio works in both directions over the telephone network.
When adding telephony, treat the call route as its own integration. Keep a record of which component should receive the call at each step.
What does SIP do in a voice-agent system?
SIP is a signaling protocol used to establish and control communication sessions. In a telephone integration, it helps connect your carrier's call to the voice application. The audio path has its own behavior and can fail even when signaling appears successful.
LiveKit supports inbound calling through configured telephony connections. Its inbound calling overview describes the trunk and dispatch components, alongside its hosted phone-number option. Choose the setup that fits your carrier and account rather than combining steps from different connection methods.
For a carrier-backed example, the route looks like this:
Caller -> Twilio phone number -> SIP connection
|
v
LiveKit room
|
v
LiveKit agentThis is a conceptual call route, not a complete account configuration. Authentication, allowed destinations and carrier settings still need to be applied for the chosen setup.
A phone number, a trunk and a dispatch rule do different jobs
The number identifies where the caller dials. The trunk connects the carrier and LiveKit. Dispatch determines where the caller goes inside the application.
A room can exist without your intended agent joining it. That is why a ringing number followed by silence should not immediately become a speech-model investigation.
Keep caller sessions separate unless a shared room is intentional. An accidental shared room creates a much more serious problem than an awkward greeting: callers may be connected to each other.
LiveKit's dispatch-rule reference explains room routing and agent configuration. Reuse the appropriate routing configuration instead of inventing a fresh routing object for every call.
Test the call route one boundary at a time
Use a dedicated test number and fictional details. Record the call reference and approximate time so you can connect carrier evidence with application evidence.
| Observation | First boundary to inspect |
|---|---|
| The carrier has no call record | Dialed number and carrier routing |
| Carrier sees the call, LiveKit does not | Trunk configuration and authorization |
| A room appears without an agent | Agent dispatch and worker availability |
| Agent joins but hears nothing | Audio routing and media negotiation |
| Caller hears speech, then loses the call | Disconnect events and telephony state |
The table suggests where to start, not a diagnosis for every failure. A network restriction or provider outage can produce similar symptoms at more than one boundary.
Keep the first test deliberately boring: the agent says a short greeting and repeats a fictional phrase. Add the customer-system lookup after the basic audio route works. This makes the next failure easier to locate.
What my telephony work does and does not establish
My work at Voho included SIP calling alongside LiveKit agents and backend integration. The portfolio case describes that contribution within a team platform.
A new agency deployment still needs its own account checks, number configuration and call tests. An existing browser demo cannot certify your carrier setup, permitted destinations or production routing.
Before handover, leave a routing map, the test number, relevant account ownership, and a rollback or fallback procedure. Avoid putting credentials in the map itself.
Questions about LiveKit phone connections
Do I always need an external carrier?
Not always. Check the current hosted phone-number offering and its supported regions and call directions. An existing business number may require a different path.
Is an inbound test enough for outbound calling?
No. Outbound dialing has separate destinations, permissions, caller identity and failure conditions to verify. Scope it independently.
What should I send someone who is diagnosing the setup?
Send a redacted routing diagram, the affected direction of calling, a test-call time and the last component that recorded the call. Keep account secrets out of the initial message.
For an existing blocked connection, those details are a useful starting point for a telephony diagnosis.