AI-Generated Note Graphs

Joplin users accumulate hundreds of notes over time, but existing graph plugins only visualize explicit links (:/noteId) and shared tags, leaving semantically related yet unlinked notes disconnected. This project introduces an AI-powered Joplin plugin that uncovers these hidden relationships using text embeddings as the primary mechanism, with optional LLM-based enrichment for labeling and categorization. The default embedding provider is Transformers.js, enabling fully local, zero-setup inference with no external dependencies, while advanced users can opt into Ollama, Google Gemini, OpenAI, or compatible APIs. The system combines explicit links, tag relationships, and embedding-based semantic similarity into a unified graph, ensuring comprehensive note connectivity. The plugin renders this enriched structure as an interactive graph using Cytoscape.js with the fCoSE layout, where node size reflects centrality, node color represents Louvain-detected communities, and edge styles distinguish relationship types (solid for links, dotted for tags, dashed for semantic connections). It includes features such as click-to-navigate, focus mode, filtering, and export to PNG/SVG/JSON. A SQLite-based cache stores embeddings for incremental updates, ensuring scalability and performance, while optional Pass B LLM enrichment adds human-readable relationship labels and categories. The final deliverable is a production-ready .jpl plugin with secure API key handling, provider abstraction, and a comprehensive test suite including deterministic validation of AI outputs.