AI calling that picks up the phone for you
Build a voice agent from a written brief, give it a number, and let it place and answer ordinary phone calls. It runs beside your WhatsApp conversations, on the same contacts, in the same platform.
The problem
Some conversations still have to happen out loud
Messaging handles most of what a business needs to say. But a share of your list never opens a message, and a share of your work is calls nobody enjoys making: confirming tomorrow's appointments, checking a delivery address, ringing forty leads to find the six worth a real conversation.
Those calls are repetitive, they are time-sensitive, and they are the first thing to be dropped on a busy day. So they get dropped, and you find out later through a no-show or a failed delivery. Hiring for them is expensive, and the work is dull enough that nobody stays in the seat long.
That is the gap AI calling fills. Not replacing your best salesperson, but clearing the predictable calls off the queue so the people on your team spend their time on the conversations where a human voice genuinely changes the outcome.
What you get
An AI phone agent you can brief, dispatch and read back
The whole loop lives in one place: define the agent, give it a number, call one contact or a thousand, then read what happened.
Agents built from a behaviour prompt
You describe how the agent should behave — who it is, what it is calling about, what it must ask, what it should never say — and that prompt becomes the agent. Create, edit and version agents through the dashboard or the API, and keep a different agent for each job.
Click-to-call from a contact
Open a contact record and start a call with one click. The agent dials out on your number, runs the conversation you defined, and the result lands back against that same contact instead of in a separate silo.
Bulk calling
Point an agent at a list and let it work through it. Bulk calling is for the repetitive end of the queue — reminders, confirmations, first-pass qualification — where every call starts from the same script and only the answers differ.
Call history and transcripts
Every call is recorded in a history you can filter and read back, including the transcript of what was said. That is what makes voice agents improvable: you read the calls that went badly and tighten the behaviour prompt.
Numbers from your wallet
Buy a phone number from inside the platform, paid from your organisation wallet, and attach it to an agent. No separate telco account to provision before you can make the first test call.
Metered billing you can see
Voice usage is metered per call against a rate card, so the cost of a campaign is visible rather than inferred from a monthly invoice. Rate cards are set per organisation, alongside the choice of voice model.
Stop a call in flight
A running call can be stopped from the platform. During a bulk run that is the difference between a bad script reaching six people and reaching six hundred.
Managed or self-hosted
Run agents through the managed telephony integration, or run self-hosted voice agents on your own infrastructure while keeping prompts, click-to-call and history in the dashboard.
Model and rate card per organisation
The voice model and the rate card applied to your account are configurable per organisation, so voice quality and cost can be tuned to what the workload actually needs.
Call events arrive back from the telephony provider over signature-verified webhooks, so the call records and billing events in your account can be trusted to have come from the provider and not from anyone else.
Typical setup
Getting your first AI voice agent onto a live call
Voice is less forgiving than chat: a person hangs up in seconds if the opening is wrong. Build it in this order and you will find the problems while the list is still small.
- 1
Buy a number
Pick a phone number in the dashboard and pay for it from your wallet. This is the number that will show on the recipient's handset, so choose the country and format your audience will recognise.
- 2
Write the behaviour prompt
Describe the agent's job in plain language: the opening line, the two or three things it must establish, how to handle an objection, and when to stop and hand off. Short and specific beats long and clever.
- 3
Test it on yourself
Use click-to-call against your own number first. Listen to how the agent opens, where it talks over you, and where it fails to recognise an answer. Read the transcript afterwards and fix the prompt.
- 4
Run a small batch
Send a bulk call to a few dozen contacts rather than the whole list. Check the call history for hang-up points. A call that is abandoned in the first eight seconds is usually an opening-line problem, not a model problem.
- 5
Wire the follow-up to WhatsApp
A voice call is a poor place to send a link, a price list or a booking form. Have the call end with a promise to send the details, then deliver them as a WhatsApp message on the same contact record.
- 6
Watch the meter
Check metered usage against the rate card after the first proper campaign, so you know what a qualified conversation actually costs you before you scale the list.
Where it fits
Be honest about what a voice agent can do
AI calling is strong on defined tasks with a clear finish line. Confirm an appointment. Verify a delivery slot. Ask three qualifying questions and route the answer. In those jobs the agent knows what a successful call looks like, and so do you.
It is weaker the moment the conversation opens up. Long consultative sales calls, complaint handling, anything where being misheard costs money — keep those with people. Voice models still stumble over accents, background noise and interruptions in ways a chat interface never has to deal with.
The practical pattern is a split. Let the agent take the volume and the routine, let it hand off cleanly when a call goes off script, and send anything that needs a link, a document or a price as a follow-up message through your shared team inbox.
Works with the rest of the platform
A call rarely stands alone. The contact you dial usually lives in your contact database and its custom fields, the reminder that triggered the call was probably set up as part of a no-code automation flow, and the detail you promised on the call goes out as a template message from your broadcast campaigns. For the conversations that are better handled in writing, the AI chatbot answering from your own documents covers the same ground with far less risk of being misheard.
New to the messaging side of this? Start with our explainer on what WhatsApp automation actually involves and the guide to automating customer support without annoying people.
Browse all featuresFAQ
Frequently asked questions
Is this calling over WhatsApp?
No. ChatSetu AI's voice agents work over ordinary telephony — a real phone number that rings a real handset, whether or not the person uses WhatsApp. It sits alongside your WhatsApp conversations rather than inside them, which is why the follow-up to a call is usually a WhatsApp message on the same contact.
What is a behaviour prompt?
It is the written brief that defines the voice agent: its identity, the reason for the call, the questions it must ask, and the boundaries it must not cross. Changing the prompt changes the agent's behaviour on the next call, so tuning a voice agent is mostly an editing job rather than an engineering one.
What is AI calling actually good at?
Defined, repetitive tasks with a clear end state: appointment reminders, delivery and order confirmations, first-pass lead qualification, and re-engaging a contact who has gone quiet. It is much weaker at open-ended conversation, negotiation, or anything where being misunderstood is expensive. Use it to clear the predictable calls so your team can take the difficult ones.
Can I stop a call that is going wrong?
Yes. An in-progress call can be stopped from the platform, which matters most during a bulk run when you spot a problem in the first few conversations. You can end the run, fix the behaviour prompt, and start again rather than letting a bad script work through the whole list.
How is voice usage billed?
Voice calling is metered per call and charged against a rate card set for your organisation, with numbers bought from the same wallet. Because the cost is per call rather than per seat, it is worth checking usage after your first campaign before scaling to a larger list.
Can I run the voice agent on my own infrastructure?
There are two paths. Managed voice agents run through a signature-verified telephony provider integration, and self-hosted voice agents let you run the agent yourself while still using the platform for behaviour prompts, click-to-call and call history. Which one suits you usually comes down to your data-residency requirements.
Want AI calling in your first month?
Create your account, or message us on WhatsApp and we'll set up your first voice agent with you.