Context Compression Engine

Long chats make AI slow and expensive. This sends the model only what it needs.

GitHub

1 A long conversation

    Use your own conversation

    Paste a chat, one message per line, starting with user: or assistant:. Nothing leaves your browser.

    2 Ask a question about it

    Pick one, or type your own.

    This page runs CCE Lite, a small version of the engine written in JavaScript that matches on shared words. The Python version understands paraphrases too. In our tests CCE finds facts said long ago as reliably as ordinary search while sending far fewer tokens than the whole chat; see the benchmarks and their limits.