Get an AI voice agent answering your phone

The phone still rings, and half the time nobody can pick it up. Callers hit voicemail, wait on hold, or hang up and try a competitor. Some questions are easier to say than to type, and a caller who gives up has left no record of what they wanted.

An AI voice agent answers every call in natural speech, 24/7. It runs on the same knowledge as your chat widget, so it answers questions, books appointments, and captures details by voice, then hands off to a human when the call needs one.

What you'll need

• A Evoriqa workspace (create one free; the voice agent is on every paid plan)
• A trained chatbot, the same one that answers your other channels
• A Twilio account with a voice-capable phone number you own

Step 1: Have your agent trained

Your voice agent is only as good as what it knows. In Evoriqa, crawl your website, upload your docs, or add FAQs so the agent can answer from your own content and say ‘I don’t know’ instead of guessing. Our training guide walks through it. Do this once and the same brain answers on the phone, chat, WhatsApp, and everywhere else.

Step 2: Get your webhook URL from Evoriqa

In the dashboard, go to your chatbot → Channels tab and pick the phone or voice channel. Choose the chatbot that should answer calls. Evoriqa gives you a webhook URL. Copy it, you’ll paste it into Twilio in the next step.

Step 3: Point your Twilio number at the webhook

Head to twilio.com and buy or select a voice-capable phone number.

1. 1In the Twilio Console, go to Phone Numbers → Manage → Active numbers.
2. 2Click the number you want the agent to answer.
3. 3Scroll to the Voice configuration section.
4. 4Under "A call comes in", choose Webhook.
5. 5Paste the webhook URL from Evoriqa.
6. 6Set the method to HTTP POST.
7. 7Click Save.

That’s the whole connection. Twilio now routes every inbound call on that number to your agent.

Step 4: Keep your published number

You don’t have to hand out a new Twilio number. Set up carrier call forwarding on your existing business line so calls forward to the Twilio number. Customers keep dialing the number they already know, and the AI agent answers behind it.

Step 5: Place a real test call

Call the number yourself. Ask a question your knowledge base covers and listen to the agent answer out loud. Try booking a time or leaving your details to see it capture a lead by voice.

Then open the unified inbox in Evoriqa. The call appears tagged as a voice conversation, with the full transcript so you can read exactly what was said.

What to listen for on your test call

The first thing a caller registers is the gap between finishing a sentence and hearing the agent start to answer. That gap is the most useful thing to judge on your own test call. Half a second sounds natural; a two- or three-second silence after you stop talking usually points at a slow webhook round-trip — the request Twilio makes out to Evoriqa and back on every turn — and a real caller hears that silence as a dropped line and hangs up. If every reply lags like that, treat it as a setup problem to chase down, not a quirk to accept.

While you have the line open, run three checks on purpose:

• Barge-in. Start talking while the agent is still speaking. It should stop and follow what you said, not read the rest of its answer over you.
• Read-back. Give it a phone number or a booking time and confirm it repeats the digits out loud before it saves them. That read-back is what you are verifying here. Without it, a misheard digit rides straight into a saved booking.
• Recovery. Say one word too quietly to catch, or trail off mid-number. The agent should ask you to repeat just that piece rather than save whatever it guessed.

Why a call behaves differently from chat, and how the agent is built for it, is the subject of the voice channel page. For setup, the shorter version is that this test call is where a bad connection or a misrouted number shows itself, while you are the one on the line instead of a customer.

What a warm transfer hands the human

Some calls should reach a person, and the handoff is where a lot of phone setups fall down: the customer explains the whole thing to a machine, gets transferred, and repeats it all to whoever picks up.

A warm transfer skips the second telling. When the agent passes a call to a human, it passes the context with it: who is calling, what they asked, what the agent already tried, and any details it captured along the way. The person taking the call starts with the thread in front of them instead of a cold "how can I help you?"

You decide when a transfer fires: a topic the agent should not handle, an explicit request for a person, or a question outside what it knows. Set those rules once and the agent follows them on every call. When no one is free to take it, the agent captures the caller's details instead, so the call still becomes a follow-up rather than a lost voicemail. The same human handoff runs across your other channels, so a caller and a chat visitor reach a person the same way.

Where the transcript lands

Every call is transcribed and written to the same unified inbox as your chat, WhatsApp, and email threads. A phone call is not a separate system to check; it is one more conversation in the list.

Open a call and you get the full transcript, the details the agent captured, and any action it took: a booking made, a lead saved, a ticket opened. You can read exactly what was said instead of guessing from a missed-call log. Because everything lands in one place, a customer who called yesterday and messages on WhatsApp today stays one thread of history, not two separate events.

This is worth the most where the phone is still the front door — dealers and service centres book most of their workshop diary by voice, and an engaged line is a booking that goes elsewhere.

An AI voice agent vs. an old phone tree

A traditional phone tree — "press 1 for sales, press 2 for support" — routes callers; it does not answer them. Every question has to be squeezed into a menu option, and anything the menu did not anticipate ends in a voicemail or a long hold.

A voice agent works the other way around. The caller says what they want in their own words, and the agent answers from your knowledge, books the appointment, or captures the details, then routes to a person only when the call needs one.

A phone tree is not wrong for every business. If all you need is to send callers to the right extension, a menu is simple and cheap. The agent earns its place when callers have real questions you want answered, not just sorted into queues.

Take it further

• Let the agent capture and qualify leads on calls that come in after hours
• Have it book and reschedule appointments straight from the conversation
• Set up a warm transfer so complex calls hand off to a human with full context
• Turn on the same agent for WhatsApp, Messenger, SMS, and email, no extra training
• White-label the whole thing so it’s your brand answering, not ours

Your phone line stops being a bottleneck and starts working around the clock. See what it does on the voice channel page, or create your workspace and hear the voice agent handle a call for your business.