Why was Replai built?

It started as a prototype for one shop's own WhatsApp inbox, which was taking 80 to 150 messages a day, most of them the same questions about stock, sizes, shipping and discounts. Grounded in the shop's product sheet, it answered about 70 percent of them correctly, and other owners turned out to have exactly the same inbox.

What kinds of messages still need a person?

The ones worth your time: wholesale enquiries, damaged shipments, custom requests, complaints and anything the knowledge base does not cover. In the original shop that was roughly 30 percent of inbound. Replai hands those threads to a person rather than guessing, and flags complaints so a human sees them first.

Do you need an API key or a developer to use Replai?

No. The first version was code calling a model API, which is exactly what a shop owner will never do. Replai now pairs to the number you already have by QR as a linked device, takes a document, a PDF or a link as its knowledge base, and starts a 7-day trial with no card.

Does Replai answer chats that are already on the phone?

Not by default. New accounts answer new conversations only, so threads already on the phone stay with you and your staff. That default exists because the tool was first paired to a live business number where a good share of the inbox was mid-conversation with real customers who expected a human.

Why I built Replai for my own WhatsApp inbox

Quick answer

Replai began as a prototype built to survive one shop's own WhatsApp inbox, which was taking 80 to 150 messages a day. Grounded in the shop's product sheet, it was answering about 70 percent of inbound within a week, leaving wholesale enquiries, complaints and custom requests for a person. It became a product when other owners described the same inbox.

The shop behind Replai was answering 80 to 150 WhatsApp messages a day.
A bot grounded in the product sheet handled about 70 percent of inbound within a week.
The remaining 30 percent were wholesale enquiries, damaged shipments and custom requests.
Replai now pairs by QR and reads a PDF or document link, with no API key to supply.

This isn't an "I had an idea on a flight" story. This is an "I had a problem I could not get out from under" story.

What problem led to building Replai?

A small e-commerce business. Phone accessories, mostly — cases, charms, screen protectors. Sold on the website, through a couple of retail partners, and over WhatsApp.

WhatsApp was the channel that mattered. Customers found us through Instagram or word-of-mouth, and the second they had a question — do you have iPhone 17 Pro cases in stock? Is the magnetic charm compatible with X? Can I get free shipping over RM 100? — they pinged WhatsApp.

By the time the store was doing real volume, I was answering 80–150 WhatsApp messages a day. Same questions. Same screenshots. Same voice notes from older customers asking the same thing about MagSafe.

I was answering at 11pm. I was answering at 6am. I was answering while driving (badly). My partner was answering for me when I was on a call. The store was growing, and the support inbox was the bottleneck — not the product, not the marketing, the inbox.

How did the first version of Replai work?

I'd been building with the Anthropic API for unrelated things. One night I wired up a quick prototype: pull the recent WhatsApp messages, feed them to Claude with a system prompt about my products and prices, generate a reply.

It worked. Embarrassingly well. The questions are repetitive — do you have this colour, is this in stock, when does shipping arrive, can I get a discount — and Claude grounded in our product sheet answered correctly the first time, every time.

Within a week the bot was handling about 70% of inbound. The 30% that needed me were the actually interesting messages: a wholesale enquiry, a damaged-shipment complaint, a custom request. The repetitive 70% were gone.

I got my evenings back.

One day in the inbox, before and after. Before, every message from 6am to 11pm was answered by hand, 80 to 150 a day. After, about 70 percent were answered from the product sheet, leaving the wholesale enquiries, damaged shipments and custom requests. — The same day, before and after. The repetitive 70 percent were the ones worth giving away.

When I realised it was a product

The pattern is not rare. The WhatsApp Business app passed 200 million monthly active users in June 2023, up from 50 million in 2020 (Meta, reported by TechCrunch, 2023). Behind each of those numbers is an owner or a staff member typing the same answers, and every operator I spoke to had hit the same wall at roughly the same volume.

I told some other small-business owners about it. A tuition centre operator who was answering parent enquiries at midnight. A friend running a Raya promo who needed to capture contest entries by photo. A florist whose voice-note inbox was full of orders she couldn't transcribe fast enough.

Same problem. Same channel. Same shape of solution. And none of them were going to write Python and call the Anthropic API to fix it.

So I started turning the hack into a real product. Pairing flow that doesn't need code (a QR scan). Knowledge base that accepts a PDF or a Google Doc URL. Templates so a tuition centre owner doesn't have to write a system prompt from scratch. Outbound campaigns with ban-safe rate limiting because someone was going to want that. A safety-net classifier that flags complaints so you don't accidentally let one fester.

That's Replai now.

What I'm building toward

The honest pitch is: everything an inbox-only customer-support team would do — without the team. Pair your number, drop in your business knowledge, let it answer the predictable 70% so you can spend your time on the 30% that actually needs you.

If you're an owner-operator running a small business in SEA and your WhatsApp inbox is eating your evenings, Replai is built for you. It's the tool I wished existed when I was the one drowning in receipts and voice notes at 11pm.

7-day free trial. No credit card, no API key. Pair in 30 seconds. Let me know if it does for you what it did for me.

Frequently asked questions

Why was Replai built?: It started as a prototype for one shop's own WhatsApp inbox, which was taking 80 to 150 messages a day, most of them the same questions about stock, sizes, shipping and discounts. Grounded in the shop's product sheet, it answered about 70 percent of them correctly, and other owners turned out to have exactly the same inbox.
What kinds of messages still need a person?: The ones worth your time: wholesale enquiries, damaged shipments, custom requests, complaints and anything the knowledge base does not cover. In the original shop that was roughly 30 percent of inbound. Replai hands those threads to a person rather than guessing, and flags complaints so a human sees them first.
Do you need an API key or a developer to use Replai?: No. The first version was code calling a model API, which is exactly what a shop owner will never do. Replai now pairs to the number you already have by QR as a linked device, takes a document, a PDF or a link as its knowledge base, and starts a 7-day trial with no card.
Does Replai answer chats that are already on the phone?: Not by default. New accounts answer new conversations only, so threads already on the phone stay with you and your staff. That default exists because the tool was first paired to a live business number where a good share of the inbox was mid-conversation with real customers who expected a human.

Sources

WhatsApp Business crosses 200M MAUs, introduces personalized messages feature, TechCrunch, 2023-06-27
WhatsApp Business app, Meta, 2026-09-07

Keep reading

See it answer your own customers.

Pair your number, paste in what you know, and message it from another phone. Ten minutes, no card.

Start free trial