The call button in a WhatsApp Business chat is a standing invitation: tap here, talk to us. Most businesses can keep that promise for about nine hours a day. On Sautikit, the same managed AI voice agents that answer ordinary inbound calls answer WhatsApp calls — prompt-defined, interruptible mid-sentence, able to call your API in the middle of a conversation, and handing the call to a human with a written summary when a human is what the moment needs. Setup is three moves in the console: connect WhatsApp, write the agent's brief, bind it to your number. From then on, every tap gets answered.
Enable WhatsApp Business Calling on your Sautikit number and a call button appears in your chat. Customers tap it and the call arrives on your number like any other inbound call — same routing, same rules, same everything.
That button is a promise, and it is a more public promise than a phone number ever was. A ring-out on the phone network is invisible; the caller shrugs and moves on. A ring-out inside a chat happens in front of your business name, your catalogue, your last cheerful reply — and the customer is still sitting in that chat when they decide to try someone else. The chat remembers.
Here is the part that changes the maths: a WhatsApp call costs the caller nothing. It travels over data, not the phone network — no airtime, no international rates, just a tap in a conversation they already had open. A customer in London, Dallas or Dubai reaches a Nairobi business as cheaply as a neighbour down the road, which for any business with diaspora customers is not a nice-to-have, it is the whole game.
Remove the cost of calling and people call more. Volume goes up — and it goes up at exactly the hours you are least staffed to take it: evenings, weekends, other people's time zones. Which is precisely when an agent that answers every single call earns its keep.
Picture a homeware shop in Nairobi, lights off since six. A customer has been scrolling the catalogue in WhatsApp and has one question standing between her and paying: does the six-seater dining set come in mahogany, and can it reach Kileleshwa on Saturday?
She taps the call button. Second ring, a warm voice: good evening, thanks for calling, how can I help? She asks. The agent checks — a live, signed lookup against the shop's own stock API, mid-conversation — confirms the finish, books the Saturday delivery window, reads the order back to her. Ninety seconds. She goes to bed a customer.
The owner wakes up to a transcript, not a missed call. Who called, what they wanted, what was promised, word for word. Multiply that by every evening, every weekend, every public holiday, and the shadow ledger of calls you never knew you missed starts posting entries in your favour.
Run the arithmetic on your own week and the scene stops being hypothetical. Suppose your chat produces just four calls outside working hours each day — perfectly plausible once calling is free, and modest for a business with customers abroad. That is roughly 120 conversations a month that used to end in a ring-out. If one in ten becomes an order, an agent has paid for its keep before you have looked at a single transcript. The numbers are illustrative; the pattern is not.
These are the managed AI agents we shipped for ordinary inbound calls, answering a new kind of call. Nothing about them is diminished by the channel:
They hold a real conversation. The caller can interrupt mid-sentence — "actually, Thursday works better" — and the agent stops and adjusts, the way a person would. No menus, no "press one".
They act, not just talk. Attach HTTP tools — your endpoints, described with JSON Schema, every request signed so your API can verify the caller. "Where is my order?" becomes a lookup against your system and a spoken answer in the same breath. Balances, bookings, reschedules: if it lives behind an endpoint, the agent reaches it on the call.
They know when to stop. When the conversation needs a human — the caller asks, or the brief says so — the agent hands over with a written summary of everything established so far. Your teammate joins a call that arrives pre-briefed, and the customer repeats nothing.
Everything is on the record. Sautikit's platform records the call and produces a per-turn transcript: who said what, where the caller interrupted, every tool call and its arguments. Read the first week of transcripts and you will find the question you forgot to put in the brief. Fix the draft, publish, and the next call is better.
Connect WhatsApp. Open your number in the console — Numbers, then the WhatsApp tab — and connect your WhatsApp Business account through Meta's official Embedded Signup. It is a popup and a few clicks, not an afternoon in a developer dashboard. Once your account qualifies for calling, the button is live in your chat.
Write the brief. An agent is a prompt with a voice. Give it a name, a description of the job — what it books, what it answers, what it must never promise, when it must fetch a human — a model from the Sautikit catalog (ids like sautikit-flash), a voice, and a hard cap on call length. Publishing freezes a revision, so you can keep sharpening the draft while yesterday's version works the phones.
Bind it. Choose always-on, or after-hours only — humans by day, agent by night. The console does it in a toggle; the API does it in one call:
A call at 11 am reaches your team exactly as before. A call at 9 pm — or from a time zone where 9 pm is lunchtime — reaches the agent.
The surrounding controls live in the same console tab and sync to WhatsApp: show or hide the call icon, restrict it by country, set per-day business-hours windows in your timezone, and turn on voicemail so even the calls you choose not to answer leave something behind. And the failure path is deliberately boring — if the agent is unavailable, the call falls through to your normal routing. Callers never hear an error because of us.
It is worth naming the machinery this replaces, because most businesses are running some version of it right now. The pinned "call us on 07xx" message that asks the customer to leave the chat, find their dialler, and pay airtime for the privilege. The SMS auto-reply that gets read, if it gets read, hours after the moment has passed. The staff WhatsApp number on somebody's personal handset, answered when that somebody is free, silent when they are asleep, gone when they resign. Clunky USSD menus for anything self-service, where one mistyped digit starts the whole dance again.
Against all of that, the call button is almost unfairly direct: the customer is already in your chat, the tap costs them nothing, and what answers is an agent that knows your business, speaks in full sentences, checks real systems, and writes everything down. The infrastructure — media, models, barge-in, transfer logic, recording — runs inside Sautikit. The part left to you is the part only you can do: describe the job.
Not every customer wants to talk. The same agents hold two-way message conversations on the same number — reading each inbound WhatsApp message, replying in the customer's language with your business context, escalating to your team with the full thread when needed. We wrote that story up separately: AI that texts back. One number, one brief, both channels — and when a text conversation outgrows the keyboard, the call button is sitting right there in the same chat.
It runs the other way too. A business can request call permission from a customer over WhatsApp messaging and, once granted, place the outbound WhatsApp call itself — permission-first outbound, on the customer's terms.
The next tap on your call button does not have to ring out.
A dedicated number activates instantly from KES 116/month — no sales calls, no setup queue. Rates and AI pricing are on /pricing; everything draws on a prepaid wallet you top up via M-Pesa.
Developers: the routing, webhook and agent APIs are documented at /developers.
Or skip the docs entirely: sign up at sautikit.com, claim a number, open the WhatsApp tab, and let Embedded Signup do its popup. Your call button can have an answer before you close the laptop.