Every AI conversation lives inside a window. See tokens pile up, watch old messages fall out, and learn the trick that keeps long chats sharp.
Every word you send, and every word it replies, is tokens. Every turn re-sends the whole window, so the meter climbs faster as the chat grows.
When the window fills, the oldest messages fall out. The model never announces it. Run the demo to the end and ask it about the price.
Compacting folds old messages into a short summary. Space comes back, detail quietly drops. Claude and ChatGPT do this behind the scenes.