FalkorDB/GraphRAG-SDK

Build fast and accurate GenAI apps with GraphRAG SDK at scale 🌟

What it solves

GraphRAG SDK addresses the common failure points in standard Retrieval-Augmented Generation (RAG) systems, specifically LLM hallucinations caused by poor retrieval. While standard vector-based RAG often fails to capture multi-hop relationships or provide verifiable evidence, this framework uses a knowledge graph to store structured, typed facts that allow for more accurate retrieval and traceable, cited answers.

How it works

It functions as a modular framework built on top of FalkorDB. During ingestion, it extracts entities and relationships from documents to build a knowledge graph. When a user asks a question, the system traverses the graph to find connected evidence, rather than relying solely on vector similarity. It supports provenance tracking via MENTIONED_IN edges, allowing applications to verify the source of generated claims or abstain from answering if the graph lacks sufficient evidence.

Who it’s for

This SDK is designed for developers and engineers building production-grade RAG applications who need high factual reliability, multi-tenant support, and the ability to perform complex, multi-hop reasoning over their data.

Highlights

  • Benchmark-leading accuracy: Outperforms standard vector RAG and several other graph-based retrieval methods in multi-document and medical benchmarks.
  • Traceable provenance: Every answer can be linked back to specific source chunks, enabling validation and reducing hallucinations.
  • Incremental updates: Supports efficient document updates and deletions without requiring a full rebuild of the knowledge graph.
  • Flexible schema: Allows users to define custom ontologies (entity and relation types) to constrain and structure extracted information.
  • Production-ready features: Includes support for multi-tenancy, PDF ingestion, and configurable retrieval strategies.

Related

  • Project
    RManLuo/gfm-rag

    GFM‑RAG is a Python library that builds a graph index from documents (or an existing knowledge graph) and uses a pre‑trained Graph Foundation Model—a graph neural network—to retrieve relevant passages for Retrieval‑Augmented Generation. It supports out‑of‑the‑box use on new datasets, fine‑tuning for specific domains, and provides interpretable multi‑hop reasoning paths.

  • Project
    HKUDS/LightRAG

    LightRAG is an open‑source Python framework that blends vector embeddings with a knowledge‑graph index for fast, scalable Retrieval‑Augmented Generation. It supports multiple chunking strategies, multimodal document parsing, incremental updates, and a variety of storage back‑ends, and ships with a REST API and Web UI.

  • Project
    trustgraph-ai/trustgraph

    TrustGraph is an open‑source, ontology‑native platform that converts documents into a standards‑compliant RDF/OWL knowledge graph, adds full provenance, and exposes natural‑language and SPARQL query interfaces for LLM‑powered agents. It ships as Docker containers, includes ingestion pipelines, a UI, and a full LLM inference stack, aiming to eliminate hallucinations and improve explainability in complex AI use cases.

  • Project
    GiovanniPasq/agentic-rag-for-dummies

    A tutorial‑style, open‑source Python project that shows how to build a modular, agent‑driven Retrieval‑Augmented Generation system with LangGraph, hierarchical chunking, conversation memory, query clarification, multi‑agent map‑reduce, self‑correction, and evaluation via RAGAS. Works with Ollama, OpenAI, Anthropic, Google Gemini, Qdrant vector store, and optional Langfuse observability.

  • Project
    neo4j/neo4j-graphrag-python

    A Python library from Neo4j that lets you build Graph Retrieval‑Augmented Generation (GraphRAG) apps. It provides pipelines to turn text/PDFs into a Neo4j knowledge graph, utilities for vector indexing, hybrid retrievers that blend vector search with graph traversal, and a simple RAG API that plugs any supported LLM (OpenAI, Anthropic, Ollama, etc.). Install via `pip install neo4j-graphrag` (optional extras for your LLM), create a KG, index embeddings, then query with natural language.