Experiment/graph UI improvements

[vrupak]

Memory Graph Changelog

Development Setup

To test changes, run these commands in separate terminals:

Terminal 1 - Build memory-graph in watch mode:

cd packages/memory-graph && bun run dev

Terminal 2 - Run the playground:

cd apps/memory-graph-playground && bun run dev

Then open http://localhost:3000 in your browser.

---

Features

Slideshow Mode

Auto-cycling through nodes with smooth animations and physics simulation

• Random node selection every 3.5s (avoids consecutive duplicates)
• Smooth pan-to-node animation with automatic popover
• Brief physics pulse (1s) on each selection
• Background dimming animation
• Single-click to stop

Node Popover with Background Dimming

Floating popover with smart positioning and focus dimming effect

• Smooth 1.5s cubic ease-out dimming animation
• Non-selected nodes: 20% opacity, unconnected edges: 10% opacity
• Smart edge detection with 20px gap from node
• Auto-flips to avoid viewport edges
• Close via backdrop click, X button, or Escape key
• Shows: title, summary, type, memory count, URL, date, ID

Document Type Icons

Canvas-rendered icons centered on document cards

• Supported: TXT, PDF, MD, DOC/DOCX, RTF, CSV, JSON
• Scales with card size (40% of height)
• Only renders when zoomed in

Physics-Driven Layout

Simplified initial positioning, letting physics create natural layouts

• Simple grid with random offsets (no concentric rings)
• 50 quick pre-ticks + smooth animation
• Eliminates teleportation on node interaction
• Faster, non-blocking initial render

Updated Color Scheme

Refined palette for better contrast and readability

Bug Fixes

Edge Viewport Culling

Fixed edges disappearing during zoom/pan

• Now checks both X and Y axis bounds
• Only culls when BOTH endpoints off-screen in same direction
• 100px margin on all sides

Memory Nodes Follow Parents

Memory nodes now move with parent documents when dragged

• Store relative offset instead of absolute position
• Automatically repositions based on parent location

Performance

k-NN Similarity Algorithm

Reduced from O(n²) to O(n·k)

• 3x faster: ~50ms → ~17ms for 100 docs
• 4,950 → 1,500 comparisons for 100 docs
• Separated into own memo (doesn't recalculate on UI interactions)

Memory Leak Fix

NodeCache now cleans up deleted nodes properly

Race Condition Fix

Atomic node/edge updates eliminate NaN positions

---