An AI that knows your trade, not an AI that writes in your place
Why Kalepio's AI writes questions rather than answers, what it refuses to do, and how to count a quota when twenty messages are worth one questionnaire.
There are two ways to put AI into a tool for freelancers. The first is to have it write in your place: the emails, the quotes, the replies to clients. The second is to have it write the questions you never find the time to draft.
We chose the second, and not out of caution. It is that the first problem of commissioned work is not writing. It is knowing what to ask for.
The real bottleneck
A vehicle wrap does not start without the make, the model, the year, the panels to cover and the files in vector format. A brochure site does not start without the copy, the logo and access to the domain. A shoot does not start without the date, the duration and the place.
You know that. Your client does not: they do not know what they do not know. So the information arrives in pieces, a message here, a photo there, a phone call for the rest. The job does not begin.
An AI that drafts your chase-up email does not touch that problem. An AI that writes the sixteen questions of the questionnaire you never got round to does.
What it actually does
It writes a service's questions. You type "full wrap and lettering for commercial vehicles", it proposes the questions to ask for that kind of job, in the words the trade actually uses. You keep the ones that suit you.
It writes a follow-up round's questions, and this is where it gets interesting. A follow-up starts from work in progress: it can see what the client has already answered, so it does not ask the brief again. You are chasing the sponsor logos, it proposes the sponsor logos, not the make of the bike.
Without that, the help would be worse than nothing: you would read sixteen questions to discover the client answered them last week.
And it answers about your studio. A chat that reads your orders, your clients, your deliveries, your payments, and answers with your real figures. "What is waiting on me", on a Monday morning, is a question a tool ought to be able to answer.
What it refuses to do
This is the part we are proudest of, and it is a list.
It does not publish a round. It neither publishes nor unlocks a delivery. It does not send or settle a payment. It does not write to your client. It does not answer a round in the client's place. It deletes nothing.
What those six have in common: your client sees them within seconds, and they cannot be taken back. An AI that has the wrong order does not find out from an error message. There is no error: the action succeeds, and a client is told something untrue.
So it cannot. Not "it asks first": it cannot.
What it can do, it does properly: add a note, change a status, create a client, open an order. And every time, a card appears, says what is about to happen, and waits for you to confirm. Opening an order is seen by nobody but you: the link exists, and your client hears about it when you send it to them.
A quota you can explain in one sentence
The hard part about AI in a product is not technical. It is that the person paying should know what they are spending.
We have two uses that cost wildly different amounts. Writing a whole questionnaire costs something. Asking a question in the chat does not. Charging the same for both would have been absurd in either direction: either the chat is overpriced, or the questionnaire is free.
So: a questionnaire is worth one credit, a message is worth a twentieth. Twenty messages, one credit. And the count restarts on the 1st of the month, on the calendar, not on your subscription date, because "ten a month, it restarts on the 1st" is a sentence people remember.
What is left is on screen at all times: a green ring with the percentage, on the three screens that spend. Not in a billing tab nobody visits. Where you are actually clicking.
A refused generation costs nothing. An AI that does not answer costs nothing. A proposal that is shown is paid for, even if you close without keeping any of it, because the work was done: that is the one slightly hard rule, and it was better written down than left to be discovered.
What it does not replace
Your judgement about which questions to ask. The AI proposes sixteen, you keep nine, and the nine you keep are your trade. Nothing is saved until you press Save.
It is a shortcut on the thankless part. It is not somebody else talking to your clients.
Can Kalepio's AI write to my clients?
No, and that is not a technical limit. Writing in a client's conversation, publishing a round, sending a payment: those are seen within seconds and cannot be taken back. The AI has no access to them at all.
What spends an AI credit?
Writing the questions of a service or a round costs one credit. A message in the assistant costs a twentieth of a credit, so twenty messages are worth one questionnaire. A refused generation, or an AI that does not answer, costs nothing.
When does the AI quota reset?
On the 1st of the month, on the calendar, not on your subscription's anniversary. When you reach zero the button stays on screen but refuses, with the reset date beside it.
Are the AI's proposed questions saved automatically?
No. They arrive in a list, you add the ones that suit you, and your own Save button is what writes. A round your client has already answered is frozen and cannot be changed at all.
Does the AI see my clients' answers?
To write a follow-up it sees the labels of the questions already asked, so it does not ask them again. It does not receive the answers: knowing a question was asked is enough, and the answers are your client's own words.
Kalepio asks these questions for you
You send one link, your client answers once, and you start with everything you need.
Read next
September 4, 2026
Keeping a client in the loop during a project
The update rhythm that stops a client from asking where things stand, without turning your day into emails.
5 minute read
September 4, 2026
Onboarding a new client after they say yes
The first week is where jobs drift, not the last. The four things to send before work starts, the one to refuse, and how to set the tone of the whole job in the first round.
9 minute read
September 4, 2026
A client comes back six months later and wants their files again
The most avoidable forty minutes in this trade, the question that costs more than the missing file, and the three things to decide before it happens.
7 minute read