Telephony integration (SIP, Twilio, Exotel)

A voice agent is useless until it can answer a ringing phone. There are two main routes. Cloud communications platforms (CPaaS) such as Twilio internationally and Exotel or Knowlarity in India provide phone numbers, inbound and outbound call control through APIs, and a media stream that sends raw audio to your agent over a websocket. Alternatively, a SIP trunk connects the agent directly to an existing PBX, Asterisk or contact-centre platform such as Genesys, so calls flow through the infrastructure you already have and existing IVR, queues and recording stay in place.

The integration also carries the boring but essential parts: caller-ID (the dialled number tells the agent which business line was called), DTMF passthrough for keypad input, warm transfer to a human queue with the transcript attached, call recording, and concurrency limits. In India, Exotel-style providers handle TRAI DND scrubbing and regulatory number provisioning that a foreign CPaaS may not.

Telephony integration is distinct from CRM integration; the first moves audio and call state, the second moves customer data. A production agent needs both.

• CTO: the telephony choice determines audio quality, latency (region and codec), concurrency ceilings and how transfers work; changing it later is disruptive.
• Operations head: warm transfer with context is where most voice projects succeed or fail operationally; it depends on the telephony layer.
• CFO: telephony minutes are a separate line from AI minutes and vary by country and provider; Indian providers are typically cheaper for domestic calls.
• Compliance officer: number provisioning, DND compliance and recording storage location all sit with the telephony provider.

Phone networks predate the internet and speak their own protocols; a language model cannot pick up a call by itself. Telephony integration exists to bridge the PSTN and the agent's audio pipeline while preserving what contact centres already rely on: queues, recording, transfers and reporting. The problem it prevents is a voice agent that works in a browser demo and cannot be deployed on your actual number. The trade-off is between the simplicity of a CPaaS (fast to start, per-minute fees, a vendor in the path) and the control of SIP into your own PBX (more setup, more ownership). Eazyware has shipped both and recommends based on existing infrastructure.

• Connecting a hospital's existing Avaya or Asterisk PBX to a booking agent over SIP so the main number is unchanged.
• Provisioning Exotel virtual numbers for an NBFC's outbound reminder campaign with DND scrubbing.
• Using Twilio media streams for a US-facing SaaS support line with warm transfer into Zendesk Talk.
• Logistics customer-care line where the tracking agent transfers to a human queue with the transcript attached.
• University admissions helpline scaled to hundreds of concurrent calls during results week.

SkillA specialised integration skill covering SIP, WebRTC, media streaming, call control APIs and transfer logic. Eazyware provides it as part of Voice Agents, supporting Twilio, Exotel and direct SIP into enterprise PBX and contact-centre platforms.

Eazyware service that covers it: AI Voice Agents (multilingual). Starting prices are on the pricing page.