How osu-nlp-group/hipporag works
HippoRAG 2 is a persistent memory and retrieval-augmented generation framework. The principal workflow is to index documents, then answer queries using retrieval over the resulting memory; the README demonstrates index(docs) followed by rag_qa(queries). The graph also preserves the documented standard-RAG, model-provider, storage, evaluation, and MCP capabilities. Some subsystem links are based on SOURCE INDEX references rather than inspected call sites, so those links are marked dashed with no evidence path.
Main entry points: HippoRAG.py HippoRAG engine · base.py LLM interface · openie_openai.py OpenIE extraction · embedding_store.py Embedding store
Components 18
Memory indexing
- HippoRAG engine
src/hipporag/HippoRAG.py - Text preprocessing
src/hipporag/preprocessing.py - OpenIE extraction
src/hipporag/information_extraction/openie_openai.py - Offline OpenIE
src/hipporag/information_extraction/openie_vllm_offline.py - Entity linking
src/hipporag/prompts/linking.py
Retrieval and answers
- Standard RAG
src/hipporag/StandardRAG.py - Reranking
src/hipporag/rerank.py - MCP server
src/hipporag/mcp_server.py
Model integrations
- Prompt templates
src/hipporag/prompts/prompt_template_manager.py - Embedding models
src/hipporag/embedding_model/base.py - LLM interface
src/hipporag/llm/base.py - LLM providers
src/hipporag/llm/openai_gpt.py - LLM gateways
src/hipporag/llm/gateways.py
Storage and evaluation
- Embedding store
src/hipporag/embedding_store.py - Vector backends
src/hipporag/vector_stores - Evaluation
src/hipporag/evaluation/qa_eval.py - Configuration
src/hipporag/utils/config_utils.py
Other components
- Application user
Connections 18
- Application user → HippoRAG engine: indexes and queries
- HippoRAG engine → Text preprocessing: uses
src/hipporag/HippoRAG.py - HippoRAG engine → OpenIE extraction: extracts facts
- HippoRAG engine → Entity linking: uses
src/hipporag/HippoRAG.py - HippoRAG engine → Embedding models: embeds text
- HippoRAG engine → Embedding store: stores vectors
src/hipporag/HippoRAG.py - HippoRAG engine → Reranking: reranks results
src/hipporag/HippoRAG.py - HippoRAG engine → Evaluation: evaluates retrieval
src/hipporag/HippoRAG.py - HippoRAG engine → Configuration: loads settings
src/hipporag/HippoRAG.py - Standard RAG → Embedding store: uses
src/hipporag/StandardRAG.py - Standard RAG → LLM interface: generates answers
- OpenIE extraction → Prompt templates: renders prompts
src/hipporag/information_extraction/openie_openai.py - OpenIE extraction → LLM providers: requests extraction
src/hipporag/information_extraction/openie_openai.py - Offline OpenIE → LLM interface: uses offline model
src/hipporag/information_extraction/openie_vllm_offline.py - LLM providers → LLM interface: implements
src/hipporag/llm/openai_gpt.py - LLM gateways → LLM interface: routes through
- Embedding store → Vector backends: delegates storage
src/hipporag/embedding_store.py - MCP server → HippoRAG engine: exposes operations