OSU-NLP-Group/HippoRAG

[NeurIPS'24] HippoRAG is a novel RAG framework inspired by human long-term memory that enables LLMs to continuously integrate knowledge across external documents. RAG + Knowledge Graphs + Personalized PageRank.

What it solves

HippoRAG 2 addresses the limitations of standard Retrieval-Augmented Generation (RAG) systems in handling complex, multi-hop reasoning and continuous knowledge integration. While traditional RAG often struggles to connect disparate pieces of information across large datasets, HippoRAG 2 uses a graph-based memory framework to improve associativity and sense-making, allowing LLMs to better synthesize information as if they had a more human-like long-term memory.

How it works

HippoRAG 2 functions as a memory framework that builds a structured graph of knowledge from input documents. Unlike standard vector-only RAG, it extracts entities and relationships to create a knowledge graph, enabling the system to perform multi-hop retrieval—connecting facts that are not directly related in the raw text but are logically linked. It is designed to be efficient in both online inference and offline indexing, requiring fewer computational resources than other graph-based RAG alternatives.

Who it’s for

This project is for developers and researchers building advanced RAG applications who need their LLMs to perform complex reasoning over large, evolving knowledge bases. It is particularly useful for those who need to integrate persistent, queryable memory into AI agents or applications using the Model Context Protocol (MCP).

Highlights

  • Improved Reasoning: Specifically optimized for multi-hop retrieval and complex sense-making tasks.
  • Resource Efficient: Offers a more lightweight indexing process compared to other graph-based RAG methods like GraphRAG or RAPTOR.
  • MCP Integration: Includes a built-in Model Context Protocol (MCP) server, allowing it to act as a persistent memory layer for agents in tools like Claude Desktop or Cursor.
  • Flexible Backend: Supports multiple vector store backends, including local Parquet files, Qdrant, ChromaDB, and Milvus.

Related

  • 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
    hhy-huang/HiRAG

    HiRAG is a Retrieval-Augmented Generation framework that uses hierarchical knowledge to improve the quality and comprehensiveness of LLM responses compared to standard RAG.

  • Project
    iMoonLab/Hyper-RAG

    A hypergraph-driven Retrieval-Augmented Generation (RAG) framework that reduces LLM hallucinations by capturing complex, beyond-pairwise relationships in domain-specific knowledge.

  • Project
    LHRLAB/HyperGraphRAG

    A Retrieval-Augmented Generation framework that uses hypergraph-structured knowledge representation to better capture complex relationships between entities.

  • Project
    Alibaba-NLP/VRAG

    A multi-turn multimodal RAG framework that uses agentic reinforcement learning to enable VLMs to reason and retrieve information across text, images, and videos.