Live · Built in India
Apavin is an AI sales agent that answers on WhatsApp, Instagram and ten more channels, in the language and script they typed in, from the prices you uploaded. Then it follows up tomorrow, in three days, and once more after a week, and stops the second they reply.
Bookings, payments, refunds and anyone who sounds annoyed go to a person. Not because we prompted it nicely. The agent has no tool that could take your customer’s money, so it cannot, and every handover it makes says why it made it.
Priya S.
WhatsApp · first message
A worked example of the rules on this page, not a recording of a customer.
Why this channel
Three figures, each one a statement Meta made itself, with the date it made it. Every other number on this page is checkable the same way.
3 billion
people, every month
Mark Zuckerberg put WhatsApp past three billion monthly users on Meta's first-quarter 2025 earnings call. It is not a channel your customers have to be persuaded onto.
500 million
of them in India
Meta confirmed the Indian figure in December 2024, and India remains WhatsApp's largest market anywhere. Which is why the pricing on this page is quoted in rupees as well as dollars.
1 July 2025
per message, not per conversation
Meta stopped charging for a 24-hour conversation window and started charging per template message. Free-form replies inside an open service window still cost nothing, so an agent that answers quickly is an agent that costs less.
The shape of it
The same person messages you on WhatsApp on Tuesday and emails on Thursday. In most tools those are two strangers. Here they are one conversation, and whoever picks it up can see all of it.
Tue 11:40pm
asked on WhatsApp
Thu 9:05am
replied by email
Fri 6:20pm
sent a DM on Instagram
It assigns, drafts, creates tasks and writes notes by calling the same role-checked endpoints a person calls. There is no privileged path for the model, which is why an agent cannot use it to read a conversation they could not open themselves.
Delivery charges, returns policy, warranty terms, the catalogue. The model answers from those. When the answer is not in them it is instructed never to mark itself safe to send, and it tells the customer it does not know rather than inventing something plausible.
Rule, trigger, decision, result and timestamp, recorded for the AI and for people alike. Append-only, because a log an administrator can quietly edit is a log that proves nothing on the day it matters.
One message, traced
Six steps, in order, each naming the thing that performs it. This is the path the code takes, not a description of the kind of thing it might do.
Received
WhatsApp · 11:40pm
Threaded onto the customer, not onto the channel.
Language
Hinglish, Latin script
It answers in what they typed, not what your settings say.
Knowledge
catalogue.pdf · 2 passages
Prices come from your uploads. Nothing else is quotable.
Confidence
0.86
Above the 0.80 floor, so it may answer without a person.
Tools offered
read and safe writes only
No payment tool exists for it to reach for.
Sent
1.9s after it arrived
Logged with every step above, for anyone who asks why.
“bhai ye 500ml wala kitne ka, aur kal tak aa jayega kya?” Two questions, one sentence, two scripts. Nobody on your team is awake.
The webhook is matched to your workspace by the phone number id it arrived on, then threaded onto the contact who emailed you in March. Same person, one conversation, not two strangers.
The price comes from the catalogue row you uploaded. The delivery window comes from your shipping policy document. Nothing else on the page is allowed to supply either number, and a figure that is in neither is not rounded, guessed or inferred.
Low risk, both facts grounded, confidence above the 0.80 floor. Nobody is online, so the unattended branch applies: waiting until morning is worse than answering.
Because that is what they wrote in. Your price, your delivery window, and a next step. The reply goes out inside the free service window, so it costs you nothing in Meta fees.
One list, filtered to ai.auto_applied_unattended, with the two documents each answer drew on. If the shipping policy was out of date, you fix the document rather than argue with a prompt.
Held. That is a price negotiation, which is money, which is the risk class that never sends unattended no matter how confident the model is. It sits in the approvals queue with a drafted reply and a reason, and it is the first thing you see at 09:02.
The rule
Every product in this category describes its AI with adjectives. None of them will tell you the condition. Here is ours, in the same words the code uses.
Low risk, 80 percent confidence or higher, and the answer is grounded in a document you loaded.
Opening hours. Delivery timelines. Is this in stock.
Money, refunds, complaints, or anything it is not sure about, for as long as somebody is on the inbox.
Where is my refund. This arrived broken. Can you do better on price.
Nobody on the team is online. Waiting until morning is worse than answering, so it answers.
Logged under its own event so you read that night's replies as one list.
Anything it cannot ground in your knowledge base. It is instructed never to rate an ungrounded answer safe to send.
Made-up policy. Invented stock. A price nobody gave it.
Both switches are off when you sign up. With both off, Apavin drafts and a person sends, every single time. Turn on triage and it proposes an assignee, a status and a reply into an approvals queue. Turn on autonomous and the table above applies.
Out-of-hours replies are audited under their own event, ai.auto_applied_unattended, so the first thing you can do in the morning is read exactly what your business said overnight, in one list, and correct the knowledge base where it was wrong.
Twelve channels
You connect your own business accounts. Your numbers and pages stay yours, and the platforms bill you directly at their own rates rather than through us.
Getting started
Step two is the whole job. Connecting a channel is a form; deciding what your business is willing to say without a human present is the part nobody can do for you, and it is worth an hour whether or not you keep the product.
WhatsApp is the usual first one, on your own Cloud API number. Your number and your Meta account stay yours, which is why Meta bills you its message fees directly instead of us marking them up.
Price list, delivery charges, returns policy, warranty terms, the catalogue. PDF, Word or a spreadsheet. This step is the product: everything the agent may state to a customer comes from these files and from nowhere else.
Both autonomy switches are off when you sign up. Apavin drafts, you read, you send. This is the hour where you find out whether your own documents actually answer the questions your customers ask, which is worth knowing whether or not you keep the product.
Triage first, so it proposes an assignee, a status and a reply into a queue. Autonomous second, when the drafts have stopped surprising you. The table above is what changes when you do.
One thing to plan around, because it is outside everybody’s control: Meta verifies your business before a WhatsApp Cloud API number can send at volume, and that queue is Meta’s, not ours. Start it on day one and use the other eleven channels while it clears.
After the conversation
Somebody asked, somebody answered, and then what. Three things sit behind the inbox and answer that, and none of them ask you to keep a second spreadsheet.
A closed conversation might be a sale, a refusal, or a question that got answered. Those are three different things and no dashboard can add them together. Drag a deal to Won or Lost - losing one asks you why, and that answer becomes the most useful chart you have.
Conversion, win rate, average deal, how long customers wait for an answer, and who on the team is carrying it. Medians rather than averages, because one customer who waited three days should not make a good week look bad - and every rate prints the number it is a rate of.
See who is reading which page right now, where they came from, and chat with them in the same inbox as everything else. Tell it who someone is when they log in and their whole anonymous history attaches to their contact. No IP addresses stored, and no tracking across anybody else’s site.
For agencies and resellers
If you run inboxes for twenty clients from one Meta app, every message from all twenty arrives at the same URL, and something has to work out whose it is. Chatwoot's own issue #13426 is a queue of people asking for exactly this, and answering it today means a proxy or an n8n workflow you now maintain.
Apavin routes every inbound Meta webhook to the right tenant by accountRef, the phone number id, page id or Instagram id it arrived on, before anything else touches it. It was designed in rather than added later, which is the only way this particular thing is ever correct.
// one URL, twenty tenants
POST /api/webhooks/meta
entry[].id → page_id
metadata → phone_number_id
resolve(accountRef) → org
// then, and only then, ingest
Pricing
Every thirty days, bought again to extend. Indian workspaces pay about ₹4,702 and ₹1,440, with the exact amount shown before you confirm.
Thirty days at a time. No card is stored to charge you later, and there is no cancellation flow to find because there is nothing to cancel.
Rupees by UPI, card or netbanking in India; dollars everywhere else. You are never shown two payment buttons, because the wrong one fails at your bank rather than on the page.
WhatsApp message charges are billed to you by Meta, on your own account, at cost. We cannot mark them up because they never pass through us.
See who is on your website and talk to them before they leave. It sells on its own at $15 a month with unlimited seats, and an Apavin Pro workspace gets it at no extra charge, threaded into this same inbox. See IntellQ.
No per-message AI charge, no per-resolution fee, no separate agent licence. A month where the agent answers a thousand questions costs the same as a month where it answers ten.
All twelve channels, no card, no invite code. We would rather give you everything for a week than a crippled version forever, because a limited free tier tells you nothing about whether this works for your team.
CSV export during the trial, after it, and after you stop paying. Nothing is deleted when a period lapses and nothing is held to make leaving expensive.
Before you sign up
You would find all four out in the first week. It costs us nothing to say them now and it costs you a week to discover them later.
Who builds it
No account manager, no support tier, no ticket number. The address below is the one that is read.

Cofounder & CEO
Product and growth, from user behaviour to shipped roadmap
Questions
It answers your customers directly on WhatsApp, Instagram, Messenger and nine other channels, in the same language and script they wrote in, using only the prices, products and policies you loaded into the knowledge base. It gives them a next step rather than a brochure, follows up on enquiries that go quiet, and hands the conversation to a person the moment a human is genuinely needed. It is off by default and takes two separate switches to turn on.
It is built not to, in two places. It may only state prices, stock, timings and discounts that appear in the material you loaded, and it is instructed never to guess a number or round one it was given. Separately, before any reply is sent, a claim that something has already been done - a booking made, an order placed, a payment taken - is caught and the conversation is handed to a person instead, because the customer-facing agent has no tool that could have done it. That check runs in any language, not just English.
No, deliberately. Bookings, payments, refunds and price changes sit behind an approval engine with a risk level on every tool and a person on the other side of it. When a customer is ready to buy, the agent gives them your own booking or purchase link and hands the chat over. The only things it can write on its own are a non-sensitive customer preference and a reminder the customer asked for.
It depends on what is making your bill move. Wati's entry plan is USD 29 a month and includes three users, but no further users can be added to it; the USD 99 Pro plan includes five and charges USD 24 for each one after that. Apavin is USD 49 for the company and USD 15 for each additional employee, with no tier to jump to add a person, and the AI is included rather than sold as an add-on. Wati is the more established product with a larger template and campaign toolset. Apavin threads twelve channels rather than WhatsApp alone, publishes the exact condition under which its AI will send a message without asking, and routes one Meta app to many client workspaces for agencies. Prices checked against each vendor's own pricing page in September 2026.
The difference is where the AI sits in the bill. Interakt's Growth plan is USD 55 a month and its Advanced plan USD 69, both with unlimited seats, and its WhatsApp AI Agents add-on is USD 74.99 on top, which includes 100 AI messages a month and charges USD 0.011 for each one beyond that. Apavin is USD 49 for the company plus USD 15 per additional employee with the agent included and no per-message AI charge, so a busy month costs the same as a quiet one. Interakt's unlimited seats are better value for a large support team; Apavin is built for a small one that wants twelve channels in one thread and a written rule about what the AI is allowed to send on its own. Prices checked against each vendor's own pricing page in September 2026.
There are two bills and they are separate. Meta charges you directly for messages: since 1 July 2025 it prices per template message rather than per 24-hour conversation, in marketing, utility and authentication categories, with India rates varying by category and volume, and free-form replies inside an open customer service window still costing nothing. Then there is the software you use to send and receive them, which is what Apavin, Wati, Interakt and the rest charge for. Apavin is USD 49 per company every 30 days plus USD 15 per additional employee, payable in rupees by UPI, card or netbanking, and it never touches Meta's fees because your Cloud API account is your own and Meta bills you on it at cost.
Yes, and it is the specific thing Apavin was architected for. Meta allows one webhook URL per app, so every message for every client arrives at the same endpoint and something has to resolve whose it is. Apavin routes each inbound Meta webhook to the right tenant by accountRef - the phone number id, page id or Instagram id it arrived on - before anything else touches the payload, so there is no proxy layer or workflow tool to maintain in front of it. Talk to us before you start and we will set the client workspaces up with you.
There is a free trial rather than a free plan: seven days, the entire product, all twelve channels, no card and no invite code. After that Apavin Pro is USD 49 per company every 30 days plus USD 15 per additional employee. We would rather give you everything for a week than a crippled version forever, because a limited free tier does not tell you whether the product works for your team.
Nothing. Conversations, contacts, channels and your knowledge base stay exactly as they are, and everything is available again the moment a payment lands. Nothing is deleted, and you can export contacts to CSV at any point, including after the trial has finished.
Apavin is a shared omnichannel inbox for small teams. It connects twelve channels, WhatsApp Business, Instagram, Messenger, Telegram, LINE, Viber, WeChat, SMS, Dialpad calls and texts, Zoho SalesIQ, any web chat widget and business email, and threads them per customer rather than per channel, so one person is one conversation. An AI teammate triages and drafts from your own knowledge base and product catalogs, and waits for human approval until you turn autonomous mode on. Every workspace starts with a full seven day trial, no card, and Apavin Pro is then USD 49 per company every 30 days plus USD 15 per additional employee.
Yes. You connect your own business accounts, which means Meta bills message fees directly to you at cost, and you keep full ownership of your numbers and pages.
Only if you turn autonomous mode on, and even then only for cases it rates low risk with 80 percent or higher confidence and that are covered by your knowledge base. Everything else waits for a human. Every autonomous action is logged.
Apavin Pro is USD 49 for the first user plus USD 15 per additional employee, every 30 days. The price is set in US dollars. If your business is in India you pay the rupee equivalent by UPI, card, netbanking or wallet, and outside India you pay in US dollars. The exact amount is always shown before you confirm. Nothing renews automatically, so you extend whenever you choose.
WhatsApp Business, Instagram DM, Facebook Messenger, Telegram, LINE, Viber, WeChat, SMS via Twilio, Dialpad for texts and calls, Zoho SalesIQ website chat, any web chat widget through our webhook bridge, and business email over IMAP and SMTP. Zoho CRM syncs contacts alongside.
Yes, and in the script the customer wrote in rather than a translation of it. A message typed in romanised Hindi is answered in romanised Hindi, one in Devanagari in Devanagari. The safety check that catches an agent claiming to have taken a booking or a payment runs in every language too, not only in English, because a guardrail that only reads English is not a guardrail in India.
In India. The product is designed, built and supported by our team here, and it is priced with Indian businesses in mind, which is why every price is shown in rupees as well as dollars.
The honest answer: Apple and carriers only allow those through partner programs, for every vendor and not just us. We cover every channel that offers a self serve business API.
If it does not hold up, export your contacts to CSV and go. That export works during the trial, after it, and long after you have stopped paying us.