Context Window

The context window is the model’s working memory for a single response, measured in tokens. Whatever you want the model to use — the instruction, the chat history, retrieved documents, the output of a tool it just called — all has to fit inside that window. Anything that doesn’t fit gets truncated or never makes it in.

Bigger windows let you pack in more, but more isn’t automatically better: a window stuffed with marginally relevant text can bury the part that matters and degrade the answer. Managing what goes into the window — and what to leave out — is its own discipline, and it’s where a lot of agent reliability is won or lost.