This web app is a proof-of-concept for token-efficient context engineering. The premise is that upon every turn of a chat, the backend generates a running summary of the conversation based on the previous running summary and the new messages.