Knowledge Graph
Your knowledge connects itself. No manual linking.
The Auto-Linking Engine discovers and builds connections between your notes automatically, using three parallel strategies. Navigate your knowledge visually — desktop ships a production vis-network graph today, with experimental WebGPU/3D renderers on web.
The Problem
Notes exist in silos. You wrote something about 'machine learning' in January and something about 'neural networks' in March — but there's no automatic connection between them, or between every note mentioning a specific person across 3 years.
The Solution
The Auto-Linking Engine (ALE) runs 3 parallel strategies: entity overlap (cosine-similarity KnowledgeNode matching), semantic similarity (pgvector ≥0.65 cosine distance), and temporal proximity (±7 days with shared content). 11 entity types extracted via LLM (audit: entity-extraction.service.ts:73). Desktop graph renders with vis-network (sidebar + minimap + semantic zoom); experimental WebGPU/Three.js/Sigma/Cytoscape renderers live on web only.
Key benefits
Connections build automatically — no manual linking ever required
11 entity types extracted: people, companies, places, events, concepts, technologies, and more
Desktop graph: vis-network with minimap, semantic zoom, and drift highlights — production-ready
Experimental renderers (WebGPU, Three.js 3D, Sigma, Cytoscape) available on web
Entity graph: find every note mentioning a specific person or concept in one click
Semantic zoom: from domain clusters to individual notes
Technical Depth
ALE runs asynchronously after capture. Entity extraction uses LLM-based NER (11 types — PERSON through TECHNICAL_TERM). Semantic similarity computed with paraphrase-multilingual-MiniLM-L12-v2 (384-dim), stored in pgvector. Desktop graph uses GraphVisCanvas (vis-network); clustering via Louvain at multiple granularity levels.