Context Window
A context window is the amount of text an AI model can take in for a single request, which is not the same thing as memory.
Everything the model considers, the instructions, the retrieved documents, the conversation so far and its own reply, has to fit inside the window. When it does not, something is dropped or summarised, and the model reasons from what is left. Larger windows have made this much less painful than it was, but the window is still a per-request budget, not a store.
The distinction matters because the two are easily conflated. A model with a very large context window still starts the next conversation empty unless something outside the model puts the relevant history back. Persistence is a property of the system around the model, which is why context windows and persistent company memory solve different problems.