asopi tech OSS Developer
JapaneseEnglish
asopi tech
asopi tech OSS Developer

Graph RAG

HC GraphRAG

GraphRAG for AWS infrastructure

HC GraphRAG adapts Microsoft Research's GraphRAG for AWS. While the official implementation requires Azure OpenAI, HC GraphRAG uses AWS Bedrock + Anthropic models and deploys on EC2, Fargate, or other AWS services.

Microsoft GraphRAG for AWS (Bedrock + Anthropic)

GraphRAG OSS
Python (LlamaIndex) Java (ONNX)

Key points

  • Leiden algorithm for hierarchical community detection
  • Local, global, and hybrid search modes
  • Multi-format document processing
ai rag graphrag

Challenge

Beyond simple vector search

Standard RAG struggles with complex queries requiring multi-hop reasoning. Microsoft GraphRAG addresses this by building hierarchical knowledge graphs with community summarization.

  • Simple vector search misses relationships between entities
  • Difficult to answer questions requiring reasoning across documents
  • No structured knowledge representation

HC GraphRAG adapts Microsoft Research’s GraphRAG for AWS infrastructure. The official Microsoft GraphRAG requires Azure OpenAI, while HC GraphRAG uses AWS Bedrock + Anthropic models and can be deployed on EC2, Fargate, or other AWS services. Two implementations are available: Python (LlamaIndex) for rapid prototyping, and Java (ONNX) for production deployments.

Bug reports and feature requests are welcome via GitHub Issues.

GraphRAG on AWS

HC GraphRAG provides the Microsoft GraphRAG pipeline on AWS infrastructure: use Bedrock + Anthropic for entity extraction, Leiden algorithm for community detection, and deploy on EC2 or Fargate.

  • Entity and relationship extraction from documents
  • Hierarchical community detection and summarization
  • Multiple search strategies for different query types

Capabilities

Two implementations

Choose Python for rapid prototyping with LlamaIndex, or Java for production deployments with native binary support.

Python Implementation

LlamaIndex + Anthropic Claude

  • Document processing: TXT, CSV, PDF, DOCX, PPTX, HTML, EML
  • Entity/relationship extraction with Anthropic Claude
  • Vector search with LanceDB
  • Community detection with Leiden algorithm

Java Implementation

ONNX + AWS Bedrock

  • Document processing: PDF, Word, HTML, plain text
  • ONNX-based embeddings (no external API calls)
  • Local, global, and hybrid search modes
  • GraalVM Native Image support

AWS Deployment

Production-ready for AWS

  • EC2 / Fargate deployable
  • AWS Bedrock + Anthropic models
  • No Azure OpenAI dependency

Features

Core features

Both implementations provide the core GraphRAG pipeline.

Document Ingestion

Multi-format support with chunking and deduplication

Entity Extraction

LLM-powered entity and relationship extraction

Community Detection

Leiden algorithm for hierarchical graph clustering

Search Strategies

Local (entity-focused), global (community-level), and hybrid modes

Tech Stack

Technology stack

Different tech choices for different deployment scenarios.

Python Stack

Rapid prototyping and experimentation

LlamaIndex framework Anthropic Claude for LLM LanceDB for vector storage HuggingFace embeddings

Java Stack

Production deployments

Java 21+ with Maven ONNX Runtime for embeddings AWS Bedrock for LLM GraalVM Native Image

Data Format

Shared across implementations

Parquet for persistence JSON for configuration CLI interface

Resources

Resources

Source code available on GitHub.

公開リソース

最新の資料・コード・コミュニティ導線をまとめています。

graphrag-anthropic-llamaindexhc-graphrag/java-core

Get Started

Try HC GraphRAG

Choose the implementation that fits your needs.

Python Implementation Java Implementation GitHub Issues