The context budget is the number of tokens available in the model's window, to be split between instructions, history, tool results, and the answer. As it nears the limit, you summarize, truncate, or remove items to avoid losing the essentials.
Example
Summarize the first 20 exchanges in 200 tokens to free up space.