In short: what you now have
You now have a digital employee. Not a chatbot that answers questions,
but a worker with a memory, tools and a workplace.
The difference is simple. A chatbot gives advice. Your assistant gets things done: writes and sends emails,
handles correspondence, gathers documents, does the maths, publishes posts, remembers your clients
and your agreements. And remembers them tomorrow, next week and three months from now.
It is born a newborn. It knows how to work, but doesn’t
know you yet: your business, your clients, your rules, your tone. The first week is training
for your employee. Put an hour into it — you get a worker. Don’t — you get a polite
conversation partner.
First 15 minutes
1
Write to it
Say hello and tell it in two or three sentences what you do. It will ask about the rest itself.
2
Give it a real task
Not a test one — something from today. For example: “write a reply to this client”, “put together a list of suppliers”, “make a poster for Saturday”.
3
Correct the first result
It will almost certainly not get it quite right. Say what exactly is off: not “bad”, but “too formal” or “drop the last paragraph”. A correction is remembered for good.
How to talk to it
- Like a person, not a program. No commands needed — write in plain words. It understands voice messages.
- One task — one message. Five jobs in one paragraph get done worse than the same five sent one by one.
- Say what the result should be, not how to get there. “I need an email that brings the client back to the conversation” works better than “write three paragraphs and add a question”.
- Changing the subject? Type
/new. It’s like closing one folder and opening another. It won’t forget you or your work — only the current conversation. It’s free and noticeably cheaper (why — see below).
- Skills and balance — via chat commands.
/skills shows what the assistant can do and lets you add or remove skills. For the balance, just ask in plain words — it will answer right there.
- Correct it straight away. Don’t let frustration pile up: say it in the moment, briefly and directly. It won’t take offence — it will remember.
What you pay for and what gets used up
There are two different bills here, and it’s important not to mix them up.
| What it is | Who pays |
|---|
| Subscription | the assistant itself, its development, our support, training, new skills | paid for your term |
| Fuel (tokens) | the model’s work: every time the assistant calls on the artificial intelligence | you, based on actual usage |
Fuel is not our markup. It goes straight to the model provider. We don’t earn on it.
What it costs. A short question costs pennies. A big job (building a spreadsheet,
sending out emails, preparing a document) costs noticeably more: the assistant takes many steps.
The thing that surprises everyone. The cost of a question grows not with how hard it is, but with the length
of the conversation: each time, the assistant rereads everything you’ve discussed in that chat.
The same question in the evening can cost eight times more than in the morning. There’s only one reason:
the conversation got long.
Hence the rule: new topic — new conversation (/new).
It’s the simplest way to pay less.
Ask about the cost up front. Before a big job, ask: “roughly how much will this
cost?” — it will work it out and tell you before it starts.
When the fuel runs out, the assistant goes quiet. It doesn’t break — it simply goes quiet,
because it has nothing to think with. Topping up takes a couple of minutes, and then it carries on by itself:
memory, work in progress and training are all still there. We’ll warn you in advance when the balance is running low.
How to check the work
Ask for evidence. Not “I sent the email”, but “to whom, when, to which address”. A good answer
always comes with facts: a number, a time, a link, a file name. If an answer sounds confident but has not a single
detail you can check — ask again. It’s a work habit, not distrust.
Check closely at first, then less often. Just like with a new employee.
What it won’t do
These aren’t limits born of weakness — they protect your money and your reputation.
- An email goes out only after you’ve seen it. It drafts the text and shows a card with a button — until you press it, the email isn’t sent. You’ll hear the word “sent” only once it actually has been.
- Publications are shown in advance too — a post, a story, a reply in DMs all go through the same card.
- It won’t touch money. No payments, transfers or trades.
- It won’t retell the contents of your documents in the chat. A passport, a certificate, a contract — it names the type of document and quietly uses what it needs.
- It won’t translate an official document. Its translation has no legal force. Embassies and government bodies need a sworn translator — it will tell you so directly.
- It won’t make up a contact or a fact. What it doesn’t know, it will say it doesn’t know. That’s its most valuable trait — protect it: don’t make it guess.
If the assistant is quiet
Three reasons, from most to least common:
- The fuel ran out. The most common one. It will say so in the chat itself, once a day, so as not to spam. Top up — and it carries on.
- You’re writing in a group, and it’s set to answer only when called by name. Address it with
@. Tell us — we’ll set it up to always answer in your work chat.
- Something broke on our side. Send us one line — we’ll sort it out and tell you honestly what happened.
Don’t wait in silence. Quiet for more than an hour is a reason to write, not a reason to decide the service doesn’t work.
Who’s responsible for what
| Your side | Our side |
|---|
| top up the fuel · teach the assistant and give it tasks · check the results · organise how your team works with it · tell us if something’s wrong |
keep the assistant working and not falling over · make sure it can do what was promised · new skills and development · supporting and training you · being honest and the first to tell you about outages · not charging you for our breakdowns |
Where to reach the people. We’re around for the whole term of your subscription. Write in the shared work chat —
both the assistant and we see you there. Specific questions get solved faster: not “everything’s bad”,
but “this email came out dry” or “yesterday it didn’t answer at 15:40”.
