What is a context window?
A context window is the maximum amount of text, measured in tokens, that a language model can take into account in a single request, including your instructions, any documents you provide and the answer it writes back.
Think of it as the model's desk. Everything it needs for this task must fit on the desk at the same time: the prompt, the reference material and the conversation so far. Anything that does not fit is simply not seen. Modern models have large desks, enough for a long contract or a full day of call notes, but not unlimited ones.
A bigger window is not the same as better reading. Models can still miss details buried in the middle of a very long input, and stuffing everything in raises cost and slows responses. Often the better design is to retrieve only the relevant pages first, then hand those to the model, which is the idea behind retrieval-augmented generation.
When a vendor quotes a context window, translate it into your documents. How long is a typical engagement letter, lease or patient intake packet? Will the tool need to compare several at once? Test with a real, anonymized example and check whether the output picks up details from the end of the file as reliably as the beginning.
Example
For example, a family law paralegal might want AI to summarize a 120-page discovery production. Rather than pasting the whole thing into one chat, the firm splits it by document, summarizes each part, and then combines the summaries. An attorney reviews the combined summary against flagged pages before relying on it. The page count is illustrative.