Context Window

Every interaction with a language model must fit inside a limited working space called the context window. It contains the tokens the model can consider at one time, including instructions, conversation history, retrieved documents, and the response being generated. When the total exceeds the model’s limit, older or less important material must be removed, summarized, or processed separately. A larger context window can support long reports, extended conversations, and analysis across many files, but capacity alone does not guarantee that the model will use every detail correctly. Information placement, prompt structure, retrieval quality, and computational cost still matter. Context windows differ from permanent memory because their contents apply to the current request unless another system explicitly stores them.