nano-graphrag and kg-rag
One is a simple, easy-to-hack implementation of GraphRAG, while the other is a comprehensive framework for Knowledge Graph Retrieval Augmented Generation (KG-RAG) focusing on financial data; thus, they are ecosystem siblings, where the former provides a foundational GraphRAG concept and the latter offers a specialized, domain-specific application of the KG-RAG paradigm.
About nano-graphrag
gusye1234/nano-graphrag
A simple, easy-to-hack GraphRAG implementation
Builds knowledge graphs from text by extracting entities and relationships, then performs retrieval-augmented generation through both global and local graph traversal modes. Supports pluggable components including multiple LLM providers (OpenAI, Bedrock, Ollama), vector databases (FAISS, Milvus, HNSWlib), and graph backends (Neo4j, NetworkX), with full async/await support and MD5-based deduplication for incremental inserts.
About kg-rag
VectorInstitute/kg-rag
This project implements a comprehensive framework for Knowledge Graph Retrieval Augmented Generation (KG-RAG). It focuses on financial data from SEC 10-Q filings and explores how knowledge graphs can improve information retrieval and question answering compared to baseline approaches.
Implements multiple retrieval strategies including entity-based embedding matching with beam search, Cypher queries against Neo4j, and hierarchical community detection (GraphRAG-style), enabling direct comparison of knowledge graph approaches versus traditional vector similarity and chain-of-thought baselines. Built as a modular Python package with Chroma vector stores, OpenAI LLM integration, and comprehensive evaluation pipelines including hyperparameter search across methods.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work