Skip to main content
AI Chatbot

It doesn't answer from a script. It answers from your material.

Instagram, WhatsApp, Telegram, Facebook, TikTok, your website and your phone line all call the same engine. Only the delivery differs; how it thinks is identical everywhere.

  1. A question arrives
  2. Context is retrieved
  3. The answer is built
  4. The sources are shown
The problem

There are three good reasons to distrust a chatbot.

All three come from experience, and all three are what should be asked immediately after the sentence “AI answers your customers”.

  1. 01

    It invents what it does not know

    This is the one everybody fears: the bot states a wrong price with complete confidence and the customer believes it. Nobody notices until somebody complains.

  2. 02

    It gives the same answer to everything

    A keyword bot sees which words a customer used, not what they asked. “And the red one?” means nothing to it.

  3. 03

    It doesn't know when to stop

    When the conversation gets difficult, or the customer gets angry, the bot keeps talking. And when a person tries to step in, both of them are typing at once.

This page is the answer to those three: which knowledge the reply is built from, how the question is understood, and who stops the bot — and when.

How an answer is built

One question, five steps, one answer and its sources underneath.

The bot does not start writing the moment a message lands. This is what happens in between.

The customer writes

«And in red, size 42?»

  1. The language is identified

    For every message separately. Whichever language the customer wrote in, the answer is built in that one — and you can restrict the list of languages if you want to.

  2. The type of question is sorted

    A greeting, a request for a human, a signal to end the conversation, or a real question — each takes a different path. A greeting never triggers a knowledge-base search.

  3. The question is rebuilt as a standalone query

    “And the red one?” means nothing on its own. The history is read and the question becomes “the red version of that model, size 42” — which is what the search can actually find.

  4. Does it need live data?

    If the question wants this minute's answer — stock, order status, availability — the bot calls your API and folds the result into the reply.

  5. Context is retrieved from your knowledge base

    The query goes to vector search and the closest passages from your own material are pulled out. If an image was sent, it is also matched against your product photos.

The bot answers

Yes — red in size 42 is in stock at 89 AZN. Delivery within Baku takes a day.

This answer was built from

  • Product catalogue — colour and size variantsCatalogue
  • Delivery termsPolicy
  • Price list — spring collectionPricing

That last block is the most important thing on this page. Every bot message stores which passages of knowledge went into it — so “why did the bot say that?” has an answer. And when no relevant context is found, the bot does not invent one: it keeps the conversation going and hands it to your team when it should.

Control

You can stop the bot at any moment.

There are three ways to do it, and all three work mid-conversation.

  1. 01

    “.” or a stop word

    An agent types one character

    A full stop, or any of the words you choose, is enough — the bot goes quiet for 30 minutes and you take over.

  2. 02

    A toggle in the panel

    The bot is switched off for that conversation

    You turn it off entirely for one customer. Messages keep being recorded; only a person replies.

  3. 03

    Automatic

    The bot steps aside by itself

    It goes quiet when a customer becomes abusive. When someone asks for a human, your team is notified with a link straight to the conversation.

Set in advance, so nobody has to intervene

Answer length
Short or long
Conversation memory
Last 15 messages
Working hours
Your own out-of-hours message
Channel rules
Separate for public comments
AI model
Selectable
Prompt versions
Only one active

Having separate rules for public comments is not incidental: a price or a personal detail must not appear under a post, and must appear in a private message. The bot knows which channel it is speaking on and behaves accordingly. Every new version of a prompt gets a number, and you can return to the previous one at any time.

What it can do

Four jobs the engine does.

The question is understood, the answer is built, the details are collected, and everything is recorded. Open any of them to see what is inside.

Voice notes
When a customer would rather record than type, the note is transcribed and answered like anything else.
Images, and the text on them
An image is analysed, any writing on it is read, and both go into the conversation.
Product recognition from a photo
The image is matched against your catalogue photos; on a strong match, the answer comes straight from that product's details.
Many languages
Whichever language the customer writes in, the answer is built in that one.
Consecutive messages are merged
Three separate messages get one complete answer — the bot does not write three times.

Scenarios

Where the difference actually shows.

All four are the moments an ordinary bot and a reasoning one come apart.

01

The half-finished question

«And the red one?»

A keyword bot stops here. Vexvon reads back through the conversation, completes the question and finds the right product.

02

A photo arrives

«Do you have this model?»

No text, just an image. It is matched against the catalogue and the answer comes back with that product's price.

03

A question it cannot answer

«What are your financing terms?»

If that subject is not in the knowledge base, the bot does not invent a number — it holds the conversation and passes it to your team.

04

An agent steps in

«I'll take this customer myself»

The manager types a full stop and the bot goes quiet for 30 minutes. The customer is never answered in two different voices.

The Vexvon platform

One engine, seven channels.

Instagram, WhatsApp, Telegram, Facebook, TikTok, your website and your phone line all connect to the same account.

A channel is only a delivery method. How the question is understood, which knowledge the answer is built from and where the details are written are identical everywhere — which is why a conversation that starts on one channel can carry on in another without the customer introducing themselves again.

Demo

Let's test it on your own material.

The best test is your own difficult question. Tell us briefly about your company — our AI Call Center agent will call you shortly and answer your questions.

  • Answers from your own material
  • Every answer shows its sources
  • You can stop the bot at any moment
Book a demo

Your details are used only for the demo and to get in touch.

Book a Meeting with Vexvon

Pick a time that suits you in our calendar.