postgresql_management_langchain_pgvector

Hybrid BM25-powered search & DB management toolkit for PostgreSQL/pgvector with LangChain integration

install

git clone https://github.com/BWKBH/postgresql_management_langchain_pgvector.git
cd postgresql_management_langchain_pgvector
pip install -e .
pip install -r requirements.txt

Install Required PostgreSQL Extensions

This project depends on the following PostgreSQL extensions:

• pgvector: for vector similarity search in PostgreSQL
• pg_search (Casecommons): full-text search functionality using PostgreSQL text search features

---

Create Extensions in Your PostgreSQL Database

After installing the extensions, connect to your PostgreSQL instance and run:

CREATE EXTENSION IF NOT EXISTS vector;
CREATE EXTENSION IF NOT EXISTS pg_search;

Features

• Seamless integration with PostgreSQL + pgvector
• Hybrid search support with BM25 + vector (Reciprocal Rank Fusion)
• Support for HNSW indexing and distance strategy configuration
• Easy control over DB schema, index creation, and deletion
• LangChain-compatible retriever for use in RAG pipelines

Quick Usage Example

from langchain_pgvector_searchkit import PGVectorController

controller = PGVectorController(...)
await controller.set_pgvector(...)
retriever = controller.vector_store.as_retriever()
docs = await retriever.ainvoke("한국어 문법에 대해 알려줘", k=4)

See more usage in test_pgvector_controller.ipynb.

Project Structure

langchain_pgvector_searchkit/
├── service/         # PGVectorController and DB management
├── db/              # Hybrid BM25 logic and vector store overrides
└── rrf/             # RRF ranking functions