langchain-mariadb

Released under the MIT license, LangChain's MariaDB integration (langchain-mariadb) provides vector capabilities for working with MariaDB version 11.7.1 and above. Users can use the provided implementations as-is or customize them for specific needs. Key features include:

• Built-in vector similarity search
• Support for cosine and euclidean distance metrics
• Robust metadata filtering options
• Performance optimization through connection pooling
• Configurable table and column settings

Getting Started

Setting Up MariaDB

Launch a MariaDB Docker container with:

docker run --name mariadb-container -e MARIADB_ROOT_PASSWORD=langchain -e MARIADB_DATABASE=langchain -p 3306:3306 -d mariadb:11.7

Installing the Package

The package uses SQLAlchemy but works best with the MariaDB connector, which requires C/C++ components:

# Debian, Ubuntu
sudo apt install libmariadb3 libmariadb-dev

# CentOS, RHEL, Rocky Linux
sudo yum install MariaDB-shared MariaDB-devel

# Install Python connector
pip install --quiet -U mariadb

Then install langchain-mariadb package

pip install -U langchain-mariadb

VectorStore works along with an LLM model, here using langchain-openai as example.

pip install langchain-openai
export OPENAI_API_KEY=...

Creating a Vector Store

from langchain_openai import OpenAIEmbeddings
from langchain_mariadb import MariaDBStore
from langchain_core.documents import Document

# connection string
url = f"mariadb+mariadbconnector://myuser:mypassword@localhost/langchain"

# Initialize vector store
vectorstore = MariaDBStore(
    embeddings=OpenAIEmbeddings(),
    embedding_length=1536,
    datasource=url,
    collection_name="my_docs"
)

Adding Data

You can add data as documents with metadata:

# adding documents
docs = [
    Document(page_content='there are cats in the pond', metadata={"id": 1, "location": "pond", "topic": "animals"}),
    Document(page_content='ducks are also found in the pond', metadata={"id": 2, "location": "pond", "topic": "animals"}),
    # More documents...
]
vectorstore.add_documents(docs)

Or as plain text with optional metadata:

texts = ['a sculpture exhibit is also at the museum', 'a new coffee shop opened on Main Street',]
metadatas = [
    {"id": 6, "location": "museum", "topic": "art"},
    {"id": 7, "location": "Main Street", "topic": "food"},
]

vectorstore.add_texts(texts=texts, metadatas=metadatas)

Searching

# Basic similarity search
results = vectorstore.similarity_search("Hello", k=2)

# Search with metadata filtering
results = vectorstore.similarity_search(
    "Hello",
    filter={"category": "greeting"}
)

Filter Options

The system supports various filtering operations on metadata:

• Equality: $eq
• Inequality: $ne
• Comparisons: $lt, $lte, $gt, $gte
• List operations: $in, $nin
• Text matching: $like, $nlike
• Logical operations: $and, $or, $not

Example:

# Search with simple filter
results = vectorstore.similarity_search('kitty', k=10, filter={
    'id': {'$in': [1, 5, 2, 9]}
})

# Search with multiple conditions (AND)
results = vectorstore.similarity_search('ducks', k=10, filter={
    'id': {'$in': [1, 5, 2, 9]},
    'location': {'$in': ["pond", "market"]}
})

Configuration Options

The MariaDBStore can be configured with various options to customize its behavior. Here are all available options:

Basic Configuration

Parameter	Type	Default	Description
embeddings	Embeddings	Required	The embeddings model to use for creating vector embeddings
embedding_length	int	1536	Length of the embedding vectors
datasource	Union[Engine, str]	Required	Database connection string or SQLAlchemy engine
collection_name	str	"langchain"	Name of the collection to store vectors
collection_metadata	Optional[dict]	None	Optional metadata for the collection
distance_strategy	DistanceStrategy	COSINE	Strategy for computing distances (COSINE or EUCLIDEAN)
logger	Optional[logging.Logger]	None	Optional logger instance for debugging
relevance_score_fn	Optional[Callable]	None	Optional function to override relevance score calculation
engine_args	Optional[dict]	None	Additional arguments passed to SQLAlchemy engine creation
lazy_init	bool	False	Whether to delay table creation until first use

Table and Column Configuration

You can customize table and column names using the MariaDBStoreSettings class:

from langchain_mariadb import MariaDBStoreSettings, TableConfig, ColumnConfig

config = MariaDBStoreSettings(
    tables=TableConfig(
        embedding_table="custom_embeddings",  # Default: "langchain_embedding"
        collection_table="custom_collections"  # Default: "langchain_collection"
    ),
    columns=ColumnConfig(
        # Embedding table columns
        embedding_id="doc_id",        # Default: "id"
        embedding="vector",           # Default: "embedding"
        content="text_content",       # Default: "content"
        metadata="doc_metadata",      # Default: "metadata"
        
        # Collection table columns
        collection_id="coll_id",      # Default: "id"
        collection_label="name",      # Default: "label"
        collection_metadata="meta"    # Default: "metadata"
    ),
    pre_delete_collection=False       # Whether to delete existing collection
)

vectorstore = MariaDBStore(
    embeddings=embeddings,
    datasource=url,
    config=config
)

Search Options

When performing searches, you can use these additional parameters:

Parameter	Type	Default	Description
k	int	4	Number of results to return
fetch_k	int	20	Number of documents to fetch before selecting top-k (for MMR search)
lambda_mult	float	0.5	Balance between relevance and diversity for MMR search (0-1)
filter	Optional[dict]	None	Optional metadata filter
score_threshold	Optional[float]	None	Optional minimum score threshold for results

Distance Strategies

The vector store supports two distance strategies:

• DistanceStrategy.COSINE (default): Uses cosine similarity
• DistanceStrategy.EUCLIDEAN: Uses Euclidean distance

from langchain_mariadb import DistanceStrategy

vectorstore = MariaDBStore(
    embeddings=embeddings,
    datasource=url,
    distance_strategy=DistanceStrategy.EUCLIDEAN
)

Chat Message History

The package also provides a way to store chat message history in MariaDB:

import uuid
from langchain_core.messages import SystemMessage, AIMessage, HumanMessage
from langchain_mariadb import MariaDBChatMessageHistory

# Set up database connection
url = f"mariadb+mariadbconnector://myuser:mypassword@localhost/chatdb"

# Create table (one-time setup)
table_name = "chat_history"
MariaDBChatMessageHistory.create_tables(url, table_name)

# Initialize chat history manager
chat_history = MariaDBChatMessageHistory(
    table_name,
    str(uuid.uuid4()), # session_id
    datasource=pool
)

# Add messages to the chat history
chat_history.add_messages([
    SystemMessage(content="Meow"),
    AIMessage(content="woof"),
    HumanMessage(content="bark"),
])

print(chat_history.messages)