Context window
The amount of text a language model can consider at once, counted in tokens.
Everything the model knows about your scene has to fit in the context window: your instruction, the surrounding prose, and whatever reference material you send. A novel does not fit. That is the whole problem.
So the real question is never "how big is the window" but "what goes in it". Sending more is not better — a model given forty thousand words of lore and one paragraph of instruction will write lore-flavoured mush. Selection beats volume.
Context window in NovelKnow
Context is assembled per scene: the beat, the neighbouring prose, and the codex entries this scene actually mentions — not a summary you had to maintain by hand. You can preview exactly what would be sent before spending anything on the generation.