From raw thoughts to connected graphs.
An intelligent note-taking and cognitive synthesis platform. It turns chaotic markdown notes into interactive, force-directed knowledge graphs and automatically schedules spaced repetition flashcards for lasting retention.
Try the deployed live version with pre-populated graphs, biology and CS subjects, notes, and flashcard queues:
Force-directed layout with degree-scaled nodes, interactive hover halos, collision avoidance, and progressive label reveal.
Centralized workspace showing active subjects, card review counts, total notes, and synchronization health.
Clean typography, sanitised markdown previews, associated knowledge graph tags, and version history.
Distraction-free flashcard study mode with flip-to-reveal answers and confidence ratings for optimal recall pacing.
Ingests raw markdown and leverages high-speed Groq LLaMA models to extract core concepts, link related ideas, and construct an interconnected conceptual topology.
Interactive force-directed graph with collision avoidance, degree-based node sizing, domain color clusters, N-hop neighborhood focus, and progressive label reveal.
Generates atomic Q&A and cloze flashcards directly from lecture notes with a daily-capped review scheduler to strengthen memory retention effortlessly.
A lightweight companion Node.js background process that monitors local markdown folders, detects changes via SHA hashing, and syncs with offline SQLite queueing.
Enterprise-grade privacy with note content encrypted at rest, Argon2id password hashing, and zero plaintext exposure even during in-memory full-text search.
Installable Progressive Web App with custom service worker caching, IndexedDB local persistence, and background sync reconciliation on reconnection.
Save markdown files locally in Obsidian, VS Code, or write directly in the cloud editor. The local watcher daemon computes SHA-256 checksums and debounces writes.
Notes are partitioned into contextual sections. Embeddings are generated via Pinecone inference (voyage-3-lite) into per-subject namespaces for semantic retrieval.
Groq LLaMA 3.3 processes chunks in parallel to extract atomic concepts, resolve cross-note synonyms, and prune duplicates using Jaccard similarity.
Relationships are inferred and mapped into Supabase Postgres as directed graph nodes and typed edges with automatic clustering above 50 nodes.
Synthesizes Q&A and cloze cards from key propositions, scheduled into a daily review queue capped to prevent learner cognitive fatigue.
By combining concept extraction with flashcard generation into unified Groq requests and batching Pinecone upserts, note processing dropped from 50s down to 14s.
You do not have to write notes inside a browser tab. The Graphite Watcher runs as a quiet background service on your workstation. Open Obsidian, Logseq, VS Code, or Notepad — the moment you save a file, it enters your knowledge graph.
watcher.db and retry on reconnection. # 1. Install & pair your local folder
cd watcher && npm install
node index.js --pair
# 2. Enter 6-digit dashboard pairing code
Pairing successful • Token saved to config.json
# 3. Start watching in the background
node index.js
Generates a secure pairing key linked directly to your authenticated Graphite user and target subject container.
| Security Layer | Implementation Detail |
|---|---|
| Content Encryption | AES-256-GCM authenticated encryption at rest |
| Credential Security | Argon2id password hashing (19 MiB memory, 2 passes) |
| Session Architecture | Tamper-proof signed JOSE JWT with revocation & auto-healing |
| Search Privacy | Decrypt-on-the-fly search without persisting plaintext indexes |
| Sync Protocol | Ephemeral 6-digit cryptographic device pairing & rate limiting |
Next.js 15
TypeScript
Groq API (LLaMA 3.3)
Pinecone Vector DB
Supabase PostgreSQL
Tailwind CSS
AES-256-GCM
Argon2id
Workbox PWA
Node.js Daemon
Experience the automated synthesis of thoughts, concepts, and memory cards. Explore the live demo or inspect the open-source codebase on GitHub.