SAG

Why we included this project

Teams that keep tripping over plain vector search when questions span several documents will find SAG worth a look. It is a knowledge base application built on a retrieval architecture that handles semantic search and relational reasoning in one pass, replacing the usual pairing of dense RAG and GraphRAG. During indexing the system extracts events and entities, and at query time it builds dynamic hyperedges, which is how multi-hop questions get answered. The app turns uploaded files and web sources into a searchable, source-traceable knowledge base, and ships with a desktop client, a self-hosted REST API, an OpenAI-compatible chat endpoint, and an MCP server you can connect to coding agents. Since it runs local-first on SQLite and LanceDB out of the box, a single machine is enough to evaluate the whole workflow before moving to production backends.