You have probably noticed: after a long conversation, the AI starts forgetting what you told it at the beginning. It is not a bug, it is a matter of "working memory". Here is why it happens and how to avoid it, no jargon required.
What is a token? A small piece of a word
An AI does not read words the way we do. It chops them into tokens, that is, small pieces. A token is roughly three quarters of a word.
For example, the word "freelancer" might split into several tokens: "free", "lan", "cer". A short sentence is a few tokens; a long email is hundreds of them.
Why is this useful to know? Because everything the AI reads and writes is counted in tokens. The more text there is, the more tokens there are to process. And that is where the limit shows up.
The context window: a desk of fixed size
Picture the desk of a very efficient person. As long as the documents fit on the desk, they see everything at a glance and answer fast. But the desk has a fixed size. When it is full, to put down a new sheet, an old one has to go.
That is exactly the context window: the amount of text the AI can keep "in front of its eyes" at the same time. Once that window is full:
- the earliest pieces of the conversation slide off the desk;
- the AI no longer "sees" them, so it forgets them;
- it may then contradict itself or ask again for something you already said.
The AI does not forget out of laziness. It forgets because its desk is full.
That is why a marathon conversation sometimes goes off the rails: your original brief has vanished under a pile of more recent messages.
Why re-explaining every time is exhausting
With a plain conversational assistant, everything lives inside the conversation. Your business, your prices, the tone of your emails, the names of your clients: if it is not in the window, it does not exist for the AI.
The result: you start from scratch with every new chat. You paste the same information in, again and again. Beyond the wasted time, it is frustrating: you feel like you are training someone who retains nothing.
How chyll keeps the thread
chyll separates two things a chatbot mixes up: the conversation of the moment and the durable context of your business.
That context (your activity, your offers, the way you speak) is learned once, then stored apart in the business memory. It does not clutter the desk: it is pulled out at the right moment, when chyll needs it.
Concretely, this changes everything:
- you no longer have to re-explain your business every time;
- all your tasks (a quote, a payment reminder, a draft reply to a client) rely on the same business memory;
- a long task no longer pushes your important information out of the window.
The AI is still bounded by its window, like every AI. The difference is the organization around it: you do not ask the desk to carry everything. You keep the essentials in a reliable binder, and you only put on the desk what is needed right now.
One last useful habit, even outside chyll: for a specific task, open a new conversation rather than stretching an endless thread. An empty desk always works better than a saturated one.