Skip to content
WeZend
One assistant on SMS, WhatsApp, e-mail and your website — answering from your own knowledge.

AI Chatbot

Answers customers on SMS, WhatsApp, e-mail and a website chat from the text, files and pages you give it, hands over to a person when it cannot answer, and drafts for your approval where you want that.

Most chatbots are built to sound sure. This one is built to know when to stop. You give it your own material — text you paste, files you upload, pages you import — and it answers only from that. When the answer is not there, it hands the conversation to a person in your Inbox instead of guessing, and every answer shows which sources it rested on.

It works on SMS, WhatsApp, e-mail and a chat widget on your site, all through the same Inbox your team already uses. On WhatsApp and e-mail it starts in shadow mode: it drafts a reply, a person approves, edits or rejects it with one click, and nothing reaches the customer until then. You switch a channel to "send by itself" yourself, after you have seen how the drafts turned out.

It also goes beyond FAQ text:

  • A webshop assistant that talks about your products and orders from your shop's own data.
  • Lookups and actions in your own systems — read-only on its own, with a person approving anything that changes something.
  • A weekly summary of what it could not answer, so the gaps in your knowledge become a to-do list.

Benefits

Why it matters

Answers you can check

It answers only from your own knowledge, and each answer shows the sources it used. If your knowledge does not hold the answer, it hands over instead of inventing one.

A person stays in charge

Shadow mode on WhatsApp and e-mail means nothing is sent until someone approves it. Once a person replies in a conversation, the assistant goes quiet.

Every channel, one Inbox

SMS, WhatsApp, e-mail and web chat all end in the same Inbox, so your team sees what the assistant answered and can take over at any point.

Capabilities

What you can do

Grounded in your knowledge

The assistant is told to answer only from the excerpts it is given and to hand over when they do not contain the answer. Malformed or unsure output becomes a hand-off, not a guess.

Knowledge from text, files and pages

Paste text, upload .txt, .md, .html or text PDFs, or import a web page — or a whole site section — and re-read it later. Starter templates for webshop, clinic, restaurant and software turn plain answers into knowledge.

Hand-off to a person

When it cannot answer, or the customer asks for a person, the conversation lands in your Inbox and your people are alerted. Outside opening hours the customer gets your away message instead.

Shadow mode

On WhatsApp and e-mail the assistant drafts and a person approves, edits or rejects. Sending by itself is something you turn on per channel, after you have reviewed enough drafts.

Web chat widget

One script tag puts the chat on your site. Quick-reply buttons, a typing indicator and an optional 1–5 rating when a conversation is resolved. An agent can take over from the Inbox.

Webshop assistant

Finds products with price, stock and sizes as stored in your shop, and gives order status — only after the shopper proves the order with the order number plus the phone or e-mail it was placed with. Return requests become a note and a queue entry for your team; nothing is refunded by the assistant.

Discounts from your own code pools

It can offer a code from a pool you own, only in situations you allow, with a daily cap and one code per person per period. The assistant asks; the server checks the rules. Back-in-stock requests on SMS and e-mail send one message when the product or size returns.

Lookups and approved actions

Define read-only lookups in your own system, such as "where is order 12?". For actions that change something, the assistant can only file a request; a signed-in admin or owner approves it, never an API key.

Weekly summary of what it could not answer

Once a week your own people get the questions it could not answer, knowledge that was never used, and answers your team gave that are waiting to be saved to the knowledge base — offered, never added automatically.

API example

Build it in minutes

Web chat widget
<!-- Paste before </body>. Use your PUBLIC key. -->
<script src="https://api.wezend.com/chat.js" data-key="YOUR_PUBLIC_KEY" defer></script>

Frequently asked questions

Will it make things up?

It is built not to. It answers only from the knowledge you gave it, every answer shows its sources so a person can check it, and when the knowledge does not hold the answer it hands over to a person instead of guessing. No tool can promise a model never gets something wrong, which is why shadow mode exists on WhatsApp and e-mail.

Can it refund an order?

No. It can request, and a person approves. A return request becomes a note and a queue entry for your team. Any action that changes something in your systems is filed as a request that a signed-in admin or owner has to approve; one approval makes exactly one call.

Does STOP still work?

Yes, and it always wins. Opt-out words are handled before the assistant ever sees the message, on every channel, and a contact who has opted out is never answered by it.

Does it message people first?

Never. It only replies to a message a customer has sent. It does not start conversations, and it does not send follow-ups on its own.

Can I see a reply before the customer does?

Yes. On WhatsApp and e-mail that is the default: the assistant writes a draft, your people are alerted, and nothing is sent until a person approves, edits or rejects it. A draft older than 24 hours expires. You can also try questions in a test console that sends nothing.

What happens when a person steps in?

The assistant goes silent in that conversation. The thread, the sources of earlier answers and an internal note with any details it collected are all in the Inbox, and the person writes as themselves.

One platform. Every customer interaction.

Replace your patchwork of messaging APIs, CDP and automation tools with a single engagement platform built for scale.

No credit card required · EU data residency · 99.99% uptime SLA