GraphRAG-SDK
GraphRAG-SDK is a GraphRAG framework built on the FalkorDB graph database that grounds LLM answers in a knowledge graph rather than vector similarity alone, tracing each answer to its source chunks so an app can inspect the evidence or decline to answer.
GraphRAG-SDK is a GraphRAG framework built on FalkorDB, aimed at reducing the hallucinations that come from relying on vector similarity alone. It builds a knowledge graph from source documents so that multi-hop facts spread across several chunks can still be retrieved through relationship traversal, and it can turn raw documents into cited answers in a few minutes of setup.
Provenance edges trace each extracted entity mention back to its source chunk, and an option to return the retrieval trail lets an application validate a generated claim or gate generation and return an explicit evidence-insufficient response when the retrieved context does not support an answer. An ontology constrains what can be extracted into the graph so it holds typed, checkable facts rather than free-form model output, and the project reports benchmark accuracy against several other retrieval systems.