openGauss Vector Store for LangChain

openGauss integration for LangChain providing scalable vector storage and search capabilities, powered by openGauss.

Features

• 🚀 Multi-Index Support - HNSW and IVFFLAT vector indexing algorithms
• 📐 Multiple Distance Metrics - EUCLIDEAN/COSINE/MANHATTAN/NEGATIVE_INNER_PRODUCT
• 🔧 Auto-Schema Management - Automatic table creation and validation
• 🧮 Dimension Validation - Type-safe dimension constraints for different vector types
• 🛡️ ACID Compliance - Transaction-safe operations with connection pooling
• 🔀 Hybrid Search - Combine vector similarity with metadata filtering
• 😀 openGauss age Graph Support - Graph store implementation for openGauss age

Installation

pip install langchain-opengauss

Prerequisites:

• openGauss >= 7.0.0
• Python 3.8+
• psycopg2-binary

Quick Start

1. Start openGauss Container

docker run --name opengauss \
  --privileged=true \
  -d \
  -e GS_PASSWORD=MyStrongPass@123 \
  -p 8888:5432 \
  opengauss/opengauss-server:latest

2. Basic Usage

from langchain_opengauss import OpenGauss, OpenGaussSettings
from langchain_core.documents import Document
from langchain_openai import OpenAIEmbeddings

# Configuration with validation
config = OpenGaussSettings(
    table_name="research_papers",
    embedding_dimension=1536,
    index_type="HNSW",
    distance_strategy="COSINE",
)

# Initialize with OpenAI embeddings
embeddings = OpenAIEmbeddings(model="text-embedding-3-large")
vector_store = OpenGauss(embedding=embeddings, config=config)

# Insert documents
docs = [
    Document(page_content="Quantum computing basics", metadata={"field": "physics"}),
    Document(page_content="Neural network architectures", metadata={"field": "ai"})
]
vector_store.add_documents(docs)

# Semantic search
results = vector_store.similarity_search("deep learning models", k=1)
print(f"Found {len(results)} relevant documents")

Configuration Guide

Connection Settings

Parameter	Default	Description
host	localhost	Database server address
port	8888	Database connection port
user	gaussdb	Database username
password	-	Complex password string
database	postgres	Default database name
min_connections	1	Connection pool minimum size
max_connections	5	Connection pool maximum size
table_name	langchain_docs	Name of the table for storing vector data and metadata
index_type	IndexType.HNSW	Vector index algorithm type. Options: HNSW or IVFFLAT\nDefault is HNSW.
vector_type	VectorType.vector	Type of vector representation to use. Default is Vector.
distance_strategy	DistanceStrategy.COSINE	Vector similarity metric to use for retrieval. Options: euclidean (L2 distance), cosine (angular distance, ideal for text embeddings), manhattan (L1 distance for sparse data), negative_inner_product (dot product for normalized vectors).\n Default is cosine.
embedding_dimension	1536	Dimensionality of the vector embeddings.

Vector Configuration

class OpenGaussSettings(BaseModel):
    index_type: IndexType = IndexType.HNSW  # HNSW or IVFFLAT
    vector_type: VectorType = VectorType.vector  # Currently supports float vectors
    distance_strategy: DistanceStrategy = DistanceStrategy.COSINE
    embedding_dimension: int = 1536  # Max 2000 for vector type

Supported Combinations

Vector Type	Dimensions	Index Types	Supported Distance Strategies
vector	≤2000	HNSW/IVFFLAT	COSINE/EUCLIDEAN/MANHATTAN/INNER_PROD

Advanced Usage

Hybrid Search with Metadata

# Filter by metadata with vector search
results = vector_store.similarity_search(
    query="machine learning",
    k=3,
    filter={"publish_year": 2023, "category": "research"},
)

Index Management

# Create optimized HNSW index
vector_store.create_hnsw_index(
    m=24,  # Number of bi-directional links
    ef_construction=128,  # Search scope during build
    ef=64,  # Search scope during queries
)

API Reference

Core Methods

Method	Description
add_documents	Insert documents with automatic embedding
similarity_search	Basic vector similarity search
similarity_search_with_score	Return (document, similarity_score) tuples
delete	Remove documents by ID list
drop_table	Delete entire collection

Performance Tips

1. Index Tuning

HNSW Index Optimization

• m (max connections per layer)

  • Default: 16
  • Range: 2~100
  • Tradeoff: Higher values improve recall but increase index build time and memory usage

• ef_construction (construction search scope)

  • Default: 64
  • Range: 4~1000 (must ≥ 2*m)

# Example HNSW configuration
vector_store.create_hnsw_index(
    m=16,  # Balance between recall and performance
    ef_construction=64,  # Ensure >2*m (48) and >ef_search
)

IVFFLAT Index Optimization

• lists
  • Calculation:

    # Recommended formula
    lists = min(int(math.sqrt(total_rows)) if total_rows > 1e6 else int(total_rows / 1000),
         2000,  # openGauss maximum
    )
    

  • Adjustment Guide:
    • Start with 1000 lists for 1M vectors
    • 2000 lists for 10M+ vectors
    • Monitor recall rate and adjust

2. Connection Pooling

OpenGaussSettings(
 min_connections=3,
 max_connections=20
)

Limitations

• Vector type bit and sparsevec currently under development

3. Start with openGaussAGEGraph

3.1. Create extension age in openGauss

#Enter docker container
docker exec -it opengauss bash

#Switch to omm user
su omm

#Connect to the database, and the OMM database is used by default
gsql -r

#Create the age plug-in on the OMM database
create extension age;

#Exit database connecting
\q

3.2. Basic Usage

from langchain_core.documents import Document
from langchain_experimental.graph_transformers import LLMGraphTransformer
from langchain_opengauss import openGaussAGEGraph, OpenGaussSettings
from langchain_community.llms import Tongyi
from langchain_core.prompts import PromptTemplate
from langchain.chains import GraphCypherQAChain
from langchain_core.output_parsers import StrOutputParser
import os

#set api-key
os.environ["DASHSCOPE_API_KEY"] = "sk-**"
graph_llm =Tongyi(model="qwen-plus", temperature=0, base_url="https://dashscope.aliyuncs.com/compatible-mode/v1")

llm_transformer = LLMGraphTransformer(
    llm=graph_llm,
    allowed_nodes=["Person", "Organization", "Location", "Award", "ResearchField"],
    allowed_relationships = ["SPOUSE", "AWARD", "FIELD_OF_RESEARCH", "WORKS_AT", "IN_LOCATION"],
)

text = """
Marie Curie, 7 November 1867 – 4 July 1934, was a Polish and naturalised-French physicist and chemist who conducted pioneering research on radioactivity.
She was the first woman to win a Nobel Prize, the first person to win a Nobel Prize twice, and the only person to win a Nobel Prize in two scientific fields.
Her husband, Pierre Curie, was a co-winner of her first Nobel Prize, making them the first-ever married couple to win the Nobel Prize and launching the Curie family legacy of five Nobel Prizes.
She was, in 1906, the first woman to become a professor at the University of Paris.
"""

documents = [Document(page_content=text)]
graph_documents = llm_transformer.convert_to_graph_documents(documents)

conf = OpenGaussSettings{
    database = "omm",				#Default database name
    user = "gaussdb",				#Database username
    password = "YourPassoword",	    #Password with complexity requirements
    host = "Your IP",				#Database server address
    port = 8888					#Database server port
}
graph=openGaussAGEGraph(graph_name='graphtest',conf=conf,create=True)
graph.add_graph_documents(graph_documents)
graph.refresh_schema()

cypher_prompt = PromptTemplate(
    template="""You are an expert in generating AGE Cypher queries.Use the following schema to generate a Cypher query to answer the given question.Do not include name, properties, or cypher.
    Schema:{schema}
    Question: {question}
    Cypher Query:""",
    input_variables=["schema", "question"],
)

chain = GraphCypherQAChain.from_llm(
    graph_llm, graph=graph, verbose=True, allow_dangerous_requests=True, cypher_validation=True, return_intermediate_steps=True,cypher_prompt=cypher_prompt
)

question = "Who get Nobel Prize ?"
result = chain.invoke({"query": question})

prompt = PromptTemplate(
    template="""You are an assistant for question-answering tasks. 
    Use the following pieces of retrieved context from a graph database to answer the question. If you don't know the answer, just say that you don't know. 
    Use two sentences maximum and keep the answer concise:
    Question: {question} 
    Graph Context: {graph_context}
    Answer: 
    """,
    input_variables=["question", "graph_context"],
)

composite_chain = prompt | graph_llm |StrOutputParser()

answer = composite_chain.invoke(
    {"question": question, "graph_context": result}
)
print(answer)

3.3 API Reference

Core Methods

Method	Description
__init__(graph_name, conf, create)	Create object of openGaussAGEGraph
_wrap_query(query: str, graph_name: str)	Convert a Cyper query to an openGauss Age compatible Sql Query.
add_graph_documents(graph_documents, include_source)	insert a list of graph documents into the graph
refresh_schema()	Refresh the graph schema information by updating the available labels, relationships, and properties