Aitetsu

TODO

• Implemented high performance canvas using pure DOM from scratch. Can handle ~450 cards while providing smooth panning and zooming.
• Decide where I wanna take this project

Ideas

• For seamless edit/view mode for cards:
  • When input box, buttons are hovered, they work immediately and you can't dragging while pressing those
  • Everywhere else (e.g. messages, empty areas) are for dragging
• Collabsible responses for easier navigation
• Fix titlebar on macOS: https://v2.tauri.app/learn/window-customization/#macos-transparent-titlebar-with-custom-window-background-color
• Have 2 modes for each card: AI chat and Note. Can switch modes for existing cards as well: AI chats -> become fully formatted markdown Note, Note -> populate input box for AI chat
• Pin a message like a bookmark (e.g. wanna bookmark an explanation response in a long conversation about smth)
• Command palette to switch models, ...

Potential Improvements

• Blurry text after zoom: Currently using React key-based remount on zoom-end to force repaint. Consider CSS-only approach for better performance:

  // Toggle will-change to force layer re-rasterization without React reconciliation
  planeRef.current.style.willChange = "auto";
  requestAnimationFrame(() => {
    planeRef.current.style.willChange = "transform";
  });