The context window is the amount of text a language model can take into account at once. It includes instructions, documents, earlier messages in the conversation and the answer itself, and it is measured in tokens.
Anything outside the window is invisible to the model. In a long conversation, the first messages can slip out, and the model forgets what you agreed at the start. Newer models have very large context windows. But even when a whole document fits, the model does not pay equal attention to every part of it.
In practice
Start a new conversation when the task changes, and repeat the key parameters if the conversation gets long. Give the model the parts of a document that matter for the task rather than the whole archive. It is the same point as in the context ladder: the right context matters more than the amount.
The concepts in this glossary run through my AI courses and workshops, where we work with them on your own tasks.