AI chatbot for your website
A widget in the corner of the page that answers visitors’ questions from your materials — price list, policies, FAQ, support chats — and shows under every answer where it took it from. What is not in the documents, it does not make up: it hands the conversation to a person. The request is collected right in the chat, without a form with seven fields.
What it looks like in a chat
An example conversation. The wording and the flow are built for you — on your services, prices and rules.
- Answers frequent questions from your materials: prices, timing, terms, warranties
- Clarifies the details and collects the request in the chat — without a form with seven fields
- Moves the conversation to Telegram or WhatsApp if the person prefers to continue there
- Hands the dialogue to a manager with the whole history when the question is non-standard
- Works at night and at weekends — a request is not lost until the morning
- Shows what visitors actually ask: the best source of edits for the website
- Answers a question written in free text, without a tree of buttons
- Takes answers only from your materials: website, price lists, policies, chats
- Shows the source under each answer — you can see where it came from
- When it does not know, it says so and hands the dialogue to a manager with the history
- Creates the request in the CRM and books a time in the same chat
- Shows what people ask most often and where it stumbles
How the bot arrives at an answer
The difference from a scripted bot is in step four: it looks for the answer in your materials instead of picking a canned reply.
-
01
We collect your knowledge base
Website pages, price lists, policies, manuals, FAQ, the internal wiki, the history of support chats — everything you already have in writing.
from what is already written -
02
We cut it into fragments and index them
A whole document is useless to the bot: the materials are split into meaningful pieces, so that the question “are you open on Sunday” finds one line from the policy.
search works paragraph by paragraph -
03
The customer writes however they write
In free text, with typos and abbreviations. “hw much 4 an oil change on a kia rio” is a normal question, and the bot answers it.
no tree of buttons -
04
The bot searches your base — and only your base
The question becomes a search across your fragments, and the model answers strictly from what was found. That is the protection against made-up answers: it simply has nothing to answer “from memory” with.
answers only from your materials -
05
It answers with a link to the source; when it does not know, it calls a person
Under the answer you can see which document it came from. If the answer is not in the base, the bot says so and hands the dialogue to a manager together with the whole conversation.
the answer can be checked -
06
We test it on your questions before launch
We run a list of real questions and count the share of correct answers. Until that share satisfies us, the bot does not go to customers.
measure first, customers second
What we need from you
Do not prepare anything specially — send it as it is.
- Materials: website, price list, policies, FAQ, wiki — as links or files
- Twenty to thirty real customer questions to test on
- A decision on where requests land and who picks up the handed-over conversations
- A person who will confirm: this answer is right, and this one is not
- A list of topics the bot must not talk about
What it runs on
We are not tied to one vendor — and that is part of the savings.
We choose the specific model for the task and the budget: not everything needs the most expensive one. If the data must not leave your premises, we deploy an open model on your own infrastructure or in a cloud region you choose.
Tell us what needs to be done
We reply within one working day. Scoping and the estimate are free.