What actually is an ‘AI model’? An explanation for people with real jobs
Everyone talks about AI models, but what *are* they? An explanation without a single difficult word. Well, almost.

You hear it everywhere: “this AI model”, “that new model”. But what actually is it?
Think of a very well-read intern
Imagine an intern who, in a few months, has read an enormous library. Books, websites, manuals, in all sorts of languages. He hasn’t memorised everything word for word, but he has learned how language works and how things relate to each other.
That is an AI model: the result of all that reading, stored in one big file.
Why does it matter where that file lives?
With most AI services, that “intern” sits in a data centre belonging to a big tech company. Every question you ask travels there. Along with every document you attach.
With Kameleon, the intern sits in your office. In a cabinet. With eyes.
And “open” models?
Some companies and universities give their models away for free, so that anyone can run them themselves. We call those open models. France’s Mistral does this, for example, as do various other parties around the world.
The great thing is: those open models have improved enormously in recent years. Good enough to do real work, and small enough to run on a single cabinet.
Does it learn from our documents?
It reads them, when you ask it a question. Just like a colleague opening a case file. But it doesn’t send them anywhere, and it doesn’t tell anyone outside your organisation. There is no “outside”, after all. The cabinet is offline.


