Most of the messages are the same few questions. You answer them at night, because by morning the person has written to someone else.
Your terms, your process and your best replies all exist — but they live in three different files and in your head. A new hire spends months learning them.
An off-the-shelf bot answers in someone else’s voice and invents things you never offered. One reply like that, and the trust is gone.
An assistant solves this not by knowing everything, but by knowing exactly your material — and stopping honestly at the edge.
An assistant is worth having for where it stops, not for what it says
Someone else’s assistant cannot be put on display: it is trained on a client’s private material. So instead of a widget — three ordinary questions and what happens to each.
Answer
What does the work include and how long does it take?
This is already written in your materials — the assistant answers from there, in your wording and your tone. The questions you have explained a hundred times stop reaching you.
Rule: if it is in your text, it answers on its own
Source
Will it be the same in my case?
The assistant answers and shows which of your documents the answer came from. You see a line you can open and check, not a decision made by a model.
Rule: every answer has an address in your files
Silence
Can you take a rush job for tomorrow?
This is not in the materials — and the assistant does not invent it. It says it is handing the question over and sends you the whole conversation. The client gets no invented promise, and you do not lose the thread.
Rule: not in your text — it does not answer, it calls you
That is the whole service: not a smart bot, but a boundary you can see and move yourself.
03
What is included
An honest list, with no “and more”
Going through your materials
I take what already exists: your correspondence, documents, site pages, voice notes. Nothing has to be written from scratch — I show you what is missing.
Training on your own text
The assistant answers in your wording, not in the model’s generic phrases. You hear the difference in the first line.
The boundary rules
A list of topics it answers on its own, and a list where it stays silent and hands over to you. The list is open — you see it and edit it.
Where it lives
Telegram, a window on your site, or a chat inside your admin panel — wherever people already write to you. The client gets no new channel to learn.
Handover to you
When the assistant stops, you get the whole conversation and reply yourself — the full context, not a bare “a client is waiting”.
Full chat log
Every conversation is kept. After a week you can see what people ask most often, and what your materials are missing.
What is not here: the assistant does not sell for you, does not promise deadlines, and does not stand in for you where you are the one who is needed.
04
How I work
Five steps — and each one says what is needed from you
1
The conversation
A short talk in Telegram: which questions reach you, and which of them you are ready to hand over.
From you — examples of real questions
2
Materials
You send whatever you have. I go through it, return a list of the gaps and say which of them are critical and which can wait.
From you — the files and access to your correspondence
3
Assembly and boundary
I build the assistant on your materials, write the answering rules and the silence boundary, and show you the first version.
From you — nothing
4
Your own test drive
You ask it your own questions and hear where it does not sound like you. Every such spot is a signal for me to change the tone or the boundary.
From you — your questions and your notes
5
Launch and handover
I put it where people write to you and show you how to change its answers without me.
From you — nothing
05
Questions
What people ask about assistants specifically
Won’t it start making things up?
It can — any model can. That is why the assistant answers from your materials and shows where it took the answer from. Where there is no source, it is blocked and hands over to you, and that is a separate list you can see, not a default setting.
Where does my data live and what does the model see?
Your materials stay in your own storage. Only the fragment needed for a specific answer goes to the model, not the whole archive. What went out is visible in the log.
Can I ban it from a particular topic?
Yes. The topics where it stays silent are a separate list, and you edit it yourself, without me. The change applies from the next message.
What if someone writes in another language?
It answers in the language of the question, provided your materials exist in that language. If they do not, it says so plainly and calls you, instead of translating your terms at random.
How much does it know, and can I add more?
Materials can be added at any time: a new condition means a new file. The assistant does not have to be rebuilt for that.
How is this different from a Telegram bot?
A bot follows a script: buttons, a form, a booking. An assistant reads your text and puts an answer into words. They are often installed together — the bot drives the action, the assistant does the talking.
06
What clients say
«We were losing buyers in the evenings: people asked about sizes and delivery, and we answered the next morning. Gordey set up an AI assistant that replies at once, taking answers from our own product descriptions instead of inventing them. Anything tricky goes to a human. Evening orders are up 30%.»
Lesia ShulhaMalecha children’s store · Dnipro
«We wanted the store itself to suggest what would genuinely suit a customer. Gordey built that and taught our team to run it — it will not stop working without him. Those recommendations now bring in 18% of our turnover. Smart suggestions, it turns out, sell better than any mailing.»
Isabelle DuboisCEO, Maison Élégance luxury retail · Paris, France
Send me the five questions you get asked most
I will tell you how many of them the assistant can close from your own materials, and where you are still the one who is needed. The answer will be specific: for each question — an answer, a source, or silence.