Beyond Text: Image Compression

CRUX captures meaning, not bytes. That principle extends to images — extract the semantic visual description, discard the pixels.

How It Works

  1. 1
    Run /crux-compress @path/to/image.png in Cursor
  2. 2
    The agent uses vision to extract semantic content into a .crux.md file
  3. 3
    To “decompress”, feed the .crux.md to any LLM with image generation