A context window is the limit on how much information an AI model can work with when producing a response. It is usually measured in tokens, the units a model uses to process information. Instructions, conversation history, retrieved passages, and the generated response can all use this capacity, depending on the system.
The context window is not the model's training knowledge. Training changes the model's internal parameters; context supplies information for the current response without necessarily changing those parameters.
Practical relevance
An assistant comparing suppliers may receive only selected passages from their websites, not every page in full. Clear sections that state the product, audience, limitations, and supporting evidence together make those passages easier to interpret when retrieved.
A larger window can accommodate more material, but does not guarantee that the model will notice, accurately use, or cite every passage. Retrieval choices and the quality of the supplied information still matter.