A context window is the working memory of a language model: the total span of tokens — prompt, retrieved documents, conversation history, and the response being generated — that the model can attend to in a single pass. Anything outside it is invisible to the model.
Window size shapes what is possible. A larger window lets an assistant hold more of a codebase, a longer specification, or a fuller conversation in view at once, enabling reasoning across files that a smaller window would force you to summarize or omit. But a big window is not free: relevant information can still get lost among irrelevant context, and cost and latency scale with how much you include.
Effective AI-assisted engineering is partly context management — supplying the model the right subset of a large system, not simply the most.