layman/terms
AI words

Context window

kon · tekst win · doh

In shortThe amount of text an AI can look at in one go: your messages, its replies and any documents you've added.

An AI doesn't remember your chat the way a friend does. It re-reads it. The context window is how much it can re-read at once.

Imagine an assistant with a desk. Everything they’re working on has to fit on it: your instructions, their notes and any papers you’ve handed over. Once the desk is full, the oldest pages slide off the back.

An AI’s context window is that desk. It’s the most text the AI can look at in one go, counted in tokens. Everything in the current chat has to fit.

What fills it up

Why it matters

Bigger windows mean the AI can read a whole report or a long chat history at once. Smaller ones mean it starts forgetting earlier parts sooner. Each time you send a message, the AI reads the whole desk again from the top.

What it isn’t

It isn’t long-term memory. When you start a new chat, the desk is cleared. Some apps add a separate memory feature, but that’s an extra on top.

Where the comparison breaks down

A person with a full desk at least knows pages fell off. An AI usually doesn’t say so. It just quietly works with what’s left.

How the words fit together

Your prompt is split into tokens. They all have to fit in the context window. RAG is a way of putting only the most useful pages on the desk.

Pro tip

In a very long chat, the AI may lose track of early details. When that happens, start a fresh chat with a short summary of what matters.

When you'll hear it

“It has a million-token context window.” It can take in a huge amount of text at once, roughly several long books.

“That's outside its context.” It can't see that part any more, because the chat got too long.

“Paste it into the context.” Add the document to the chat so the AI can read it.

Say it in a meeting

“Does this AI need to read whole documents at once, or just short questions?”