SQLite as a Graph Database: Recursive CTEs, Semantic Search, and Why We Ditched Neo4j

The post describes ctxgraph, a project that replaces the typical Neo4j + LLM + vector DB stack for knowledge graphs with a single SQLite-based binary. By leveraging recursive CTEs for graph traversal and FTS5 for search, it achieves 0.800 F1 on extraction benchmarks with zero API calls and no Docker dependency. The schema uses three core tables (episodes, entities, edges) with proper indexing to handle tens of thousands of nodes efficiently.