Context window
The total tokens a model can consider at once: your prompt, the conversation so far, and the answer it is writing.
Also written: context length, context limit, context windows.
Everything the model is thinking about has to fit in one budget: the system prompt (Standing instructions sent ahead of every conversation, setting the role, rules and format.), any files you attached, the whole conversation so far, and the answer it has not finished writing yet. That budget is the context window.
It is not memory. Nothing persists between calls. Each time you send a message, the entire conversation is sent again from the beginning — which is why a long chat gets slower and more expensive with every turn, and why prompt caching (Paying a reduced rate for a prefix the vendor has already processed, instead of full price for sending it again.) exists at all.
When you hit the limit, something has to go. Most tools (Letting a model call functions you define — search, read a file, send a request — and read back what they return.) silently drop the oldest turns, which is the single most common reason an assistant “forgets” a decision you made twenty messages ago. A big context window is not the same as a model that uses it well.