Jargon:Context window
1 min read
Say this at the party
“How much an AI can keep in mind at once. Everything in the chat has to fit, and older bits fall out or get fuzzy.”
In plain English
An AI doesn't remember your chat the way you do. Every time you send a message, it re-reads the whole conversation from the top, plus any files you've added.
The context window is the size limit on that pile, measured in tokens. Today's big models fit hundreds of thousands, about the length of a couple of long novels.
When the pile gets too big, something has to give. Older parts get dropped or summarised, and the AI starts losing track of details.
Where it came from
The limit has been there since the first language models. Early ChatGPT could hold about 4,000 tokens, a few pages. Windows grew a hundredfold or more within two years, but researchers found models still pay most attention to the start and end, a problem nicknamed "lost in the middle".
See it in action
Three hours into planning a holiday, the AI suggests a hotel you ruled out in the first ten minutes. Your early messages have slipped out of view. A fresh chat with a short summary fixes it.
What people get wrong
- A bigger window isn't the same as memory. Close the chat and it's gone, unless the app keeps notes somewhere else.
- Stuffing the window doesn't help. The more you pile in, the more the AI misses.
Where Ugenie fits
That's why Ugenie keeps notes outside any one chat. The window forgets; notes don't.
Get Ugenie →Go deeper
- GuideAnthropic, Claude Platform DocsContext windowsA clear picture of what fills the window, turn by turn.
- PaperLiu et al., 2023Lost in the Middle: How Language Models Use Long ContextsWhy the middle of a long chat gets forgotten first.
- ArticleAnthropic, 2025Effective context engineering for AI agentsHow to choose what goes in the window, not just how much.
Where it comes up
Related jargon
Drafted with AI, checked by the AI Council (Claude, ChatGPT and Gemini) and signed off by Phin.