How it fits together
Voice AI platforms take calls over SIP like an ordinary trunk. Your PBX sends the call to the platform’s address, with the number the agent is attached to in the Request-URI. Audio flows over RTP directly between the PBX and the platform, so the firewall must allow the media port range as well as signalling. When the agent transfers a call to a person, it calls back into your PBX, configured on the platform as the termination / outbound address.
Common problems
- Call connects with no audio: NAT. Set the public address in the transport (external_media_address) and open the RTP ports.
- 404 Not Found: the Request-URI number does not match the one imported on the platform (often the “+”).
- 488: no common codec. Keep G.711 (PCMA/PCMU) or G.722.
- The agent does not hear key presses: use RFC 4733 DTMF (telephone-event).
For costs, see the AI voice agent cost calculator. Try ElevenLabs Agents: elevenlabs.io.
Addresses, transports and IP ranges come from the platforms’ documentation as of October 2026. Some links on this page are affiliate links.