What is important to know about the context window? Return JSON {"facts": [list of 2+ facts], "why": "briefly"}.
The model has a limited context window - how much text (in tokens) it sees at a time: the system prompt, history, documents, and response. Key facts: the window is finite - endless history does not fit in, the old is forced out; each token in the window costs money and time on each request; the model “remembers” only what is currently in the window - it itself does not store past chats; closer to the limit, quality and attention to detail decline. Therefore, in the application you decide for yourself what to put in the window. Rule: the context window is the model’s desktop, not its long-term memory; it is limited and paid, put only what you need there.