Skip to content

Search Databases

Flexible GraphRAG supports three full-text search options for the hybrid search pipeline. Set SEARCH_DB to select the store and SEARCH_BACKEND to choose the framework (llamaindex or langchain).

BM25 (Built-in)

Local in-memory BM25 full-text search with TF-IDF ranking. No external server required.

  • Best for: Local development, simple deployments
SEARCH_DB=bm25
BM25_SEARCH_DB_CONFIG={"persist_dir": "./bm25_index"}

Elasticsearch

Enterprise search engine with advanced analyzers, faceted search, and real-time analytics.

  • Dashboard: Kibana at http://localhost:5601
  • Docker: Uncomment includes/elasticsearch-dev.yaml in docker-compose.yaml
SEARCH_DB=elasticsearch
ELASTICSEARCH_SEARCH_DB_CONFIG={"hosts": ["http://localhost:9200"], "index_name": "hybrid_search"}

OpenSearch

AWS-led open-source fork with native hybrid scoring (vector + BM25) and k-NN algorithms.

  • Dashboard: OpenSearch Dashboards at http://localhost:5601
  • Docker: Uncomment includes/opensearch.yaml in docker-compose.yaml
SEARCH_DB=opensearch
OPENSEARCH_SEARCH_DB_CONFIG={"hosts": ["http://localhost:9201"], "index_name": "hybrid_search"}

For optimal hybrid search with OpenSearch, set up the hybrid search pipeline:

python scripts/create_opensearch_pipeline.py

Disabling Full-Text Search

SEARCH_DB=none