magic note

Inspiration

Neural Zettelkasten is an AI-powered knowledge management system that transforms how you capture, connect, and discover ideas. Unlike traditional note-taking apps, it actively helps you think:

🔮 Smart Resurfacing: Automatically surfaces relevant old notes (7+ days) while you write, using semantic similarity and temporal decay—helping you rediscover forgotten insights at the perfect moment ⚖️ Argument Mapper: Analyzes your notes by topic to reveal supporting evidence, contradictions, and evolution in your thinking over time 📚 Reading Integration: Import content from text, URLs, or PDFs with AI-powered extraction that automatically links to your existing notes 🗺️ Interactive Knowledge Map: Visualize your thoughts as a beautiful force-directed graph with color-coded clusters and right-click synthesis 🔗 Wiki Linking: Connect ideas naturally with [[double bracket]] syntax 🏷️ Auto-Tagging: AI suggests relevant tags based on semantic understanding Everything runs 100% locally in your browser—your notes never leave your machine. No servers, no subscriptions, instant performance.

How we built it

Tech Stack:

Frontend: React 18 + TypeScript + Vite for blazing-fast development Styling: Tailwind CSS with custom Harry Potter-themed parchment aesthetic State Management: Zustand with localStorage persistence Database: IndexedDB (via Dexie) for local-first architecture AI Engine: Google Gemini 2.5 Flash API with temperature-tuned prompts Visualization: react-force-graph-2d with D3.js physics simulation PDF Processing: PDF.js with custom formatting preservation Architecture: We built a local-first architecture where all data lives in IndexedDB. AI features call Gemini API with carefully crafted prompts—for example, semantic search uses temperature 0.4 for focused matching, while auto-tagging uses 0.6 for creative consistency.

The knowledge graph uses force-directed layout with tuned physics:

Charge repulsion: -300 for spread Link distance scales with shared tags: 180 - (shared_tags × 30) Center gravity: 0.03 for gentle clustering PDF extraction analyzes position and font-size data to detect headings and preserve paragraph structure—not just dumping raw text.

Challenges we ran into

We ran into a wall while trying to get the various AI agents to create some of the assets for the project. Most notably, the books, bookshelves, and map.

Accomplishments that we're proud of

✨ Built something genuinely unique - Not another Notion clone. Smart resurfacing and argument mapping are features we haven't seen elsewhere.

🎨 Beautiful, cohesive design - The Harry Potter parchment theme makes knowledge work feel magical, not mechanical.

🚀 100% local-first - No backend, no database servers, no authentication. Everything just works, instantly, privately.

🧠 AI that enhances thinking - Not AI that writes for you, but AI that surfaces connections you'd miss, reveals contradictions you'd ignore, and links ideas you'd forget.

📄 PDF extraction with formatting - We didn't just extract text—we preserved structure by analyzing position and font data.

⚡ Performance optimization - Force-directed graphs are expensive to render, but we tuned D3 physics for smooth 60fps interaction even with 100+ notes.

What's next for magic note

Export Formats: Obsidian markdown, Roam JSON, compiled PDF "knowledge books" Mobile App: React Native version with optional P2P sync via CRDT (Yjs or Automerge)

Built With

• chatgpt
• claude
• cursor
• gemini