Managing your clients with an AI, without handing over the keys

You already have an assistant open in a tab. You use it to reread an email, summarise a spec, find a better sentence. And in another tab sit your orders, your clients, your deposits, and the assistant knows nothing about any of it.

A connector is the bridge between the two. You ask "where is Camille's order?" and it answers, because it is actually looking.

What it is good for, concretely

These are not invented examples. They are the four questions everybody asks themselves on a Monday morning.

"What is waiting for me this week?" It reads your orders, finds the ones where the client has answered and you have not moved yet, and lists them. It is the one piece of work nobody enjoys and that decides the whole day.

"Chase the clients who have not answered." It finds the briefs still unanswered after ten days and tells you who, on what, and since when. And it can write the chaser, on one condition: it shows you the text before sending and waits for you to approve it.

"Open an order for this new client." You dictate three lines, it creates the sheet and the order, and hands you back the brief link to send.

"Prepare a 30% deposit on Mathieu's quote." It works out the figure, prepares the line, and stops there. The line exists, your client knows nothing about it, and sending it is your decision.

What it will never be able to do

This is the part that matters, and it matters more than the list above.

An assistant never does more than you. On every call, your own rights in the studio are checked. If your colleague has not given you access to the money, your assistant will not see it either. A right taken away from you it loses the same day. This is not a checkbox: it is the mechanism.

It never sends anything without your say-so. Writing to a client is possible, and it is exactly the act where it has to show you the precise text first: the tone you take with a client who has been silent for three weeks is decided, not delegated.

It attaches a file by fetching it from a public address, not by taking it off your computer: it runs elsewhere, and nothing can hand it a file from your machine.

It can write down a brief given over the telephone, and the record then says who typed it: a brief stays a record of what the client said, not of what somebody understood.

It cannot touch your services, your public page, your team or your subscription. What you sell and who you work with is settled by looking at it, not by dictating it.

That leaves six acts your client sees within seconds: writing to them, publishing questions, publishing a delivery, unlocking one, sending a payment, cancelling one. On those, the assistant is instructed to repeat back what it is about to do, naming the order and the client, and to wait. Because an assistant can pick the wrong order, and on those six the mistake reaches somebody.

Which assistants

The protocol is called MCP, for Model Context Protocol. It is an open standard rather than a bespoke integration, which means one connection works everywhere it is understood.

For everyday use: Claude, on the web and in the app, and ChatGPT. For development tools: Claude Code, Cursor, VS Code, Windsurf, Zed, Gemini CLI. And anything else that speaks MCP, today or tomorrow.

The list moves quickly and so does the place the setting lives. The safest route is to search for "connectors" or "MCP" in your assistant's own settings: if it has them, that is where.

If you would rather write code

Kalepio also has an ordinary REST API, with keys you create yourself. That is the right tool for plugging in your own site, a form, a dashboard, or anything you write. A connector is for talking to an assistant; a key is for programming. Both respect exactly the same rights.

What it costs

Nothing extra. The API and the AI connectors are included in every plan, the free one included.

Connecting yours

It starts from your assistant, never from Kalepio: you add the server address in its settings, it sends you to Kalepio to sign in and choose the studio, and no key has to be copied anywhere.

The step by step, with the exact address and how to disconnect, is in the guide: AI connectors.

Frequently asked

Can an AI really manage my clients?

It can read your orders, keep your client sheets up to date, open an order, prepare a payment and write to a client. Everything that leaves the studio stays a decision you make: for those acts the assistant repeats back what it is about to do, and for a message it shows you the exact wording, then waits.

Does my assistant see all of my data?

It sees what you see, in a single studio, the one you pick when you authorise it. Your rights apply first: whatever your studio does not let you do, it will not do either.

Which assistants can be connected to Kalepio?

Any that speak the MCP protocol. Today that covers Claude, ChatGPT, Claude Code, Cursor, VS Code, Windsurf, Zed and Gemini CLI, and the list keeps growing. Look for "connectors" or "MCP" in your assistant's settings.

Do I need to know how to code?

No. You paste an address into your assistant's settings and sign in to Kalepio as usual. There is no key to create and none to copy. Writing code is for plugging in your own site, and that is what the API is for.

How do I cut off access?

From your settings, under AI connectors: one click and the assistant loses access immediately. Remember to remove Kalepio from your assistant's settings too, or it will ask you to authorise it again.