The maximum amount of text, measured in tokens, a model can see and take into account in a single request or conversation.
The context window is the model's short-term memory: everything it can see at once, including your instructions, the conversation so far, attached documents and the reply it is writing. It is measured in tokens and varies from model to model. Anything outside the window is simply invisible to the model.
In an app, this explains why an assistant forgets details mentioned at the start of a very long chat, and why you cannot send every company file with every request. Even when the window is large, filling it with irrelevant text raises cost and can distract the model. Practical tip: send only what the model needs. For large knowledge bases, use RAG to fetch just the passages related to the question, and in long conversations summarize earlier turns instead of resending them in full.