Self-host Personal External Brain Knowledge Graph

This local-first application constructs a personal knowledge graph from varied inputs such as notes, web pages, Markdown files, PDFs, audio recordings, and video. It pulls out claims with exact source references, filters ingestion by topic relevance, and delivers topic-isolated recall where every retrieved fact gets a confidence score. The system deliberately declines to answer when evidence is lacking, making refusal rate a quality measure. It includes decision tracking with outcome logging, learning loops with spaced repetition and misconception tracking, and a three-panel workbench interface. The stack uses Kuzu for graph storage, LanceDB for vector search, a FastAPI backend, and a Next.js/React frontend. You can deploy via Docker Compose or locally with Node.js 18+ and Python 3.11/3.12. The project is in early alpha with a strict design constitution that prioritizes evidence-backed answers over fluent but ungrounded generation.