LangChain 的 MariaDB 集成 (langchain-mariadb) 为使用 MariaDB 11.7.1 及以上版本提供向量功能，采用 MIT 许可证分发。用户可直接使用提供的实现，也可根据特定需求进行自定义。 主要特性包括：

• * 内置向量相似性搜索
• * 支持余弦和欧几里得距离度量
• * 强大的元数据过滤选项
• * 通过连接池实现性能优化
• * 可配置的表和列设置

设置

使用以下命令启动 MariaDB Docker 容器：

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

安装包

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 -U mariadb

然后安装 langchain-mariadb 包

pip install -U langchain-mariadb

VectorStore 与 LLM 模型配合使用，这里使用 langchain-openai 作为示例。

pip install langchain-openai

初始化

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

# 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",
)

管理向量存储

添加数据

您可以将数据作为带元数据的文档添加：

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)

或作为纯文本并附带可选元数据：

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)

查询向量存储

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

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

过滤选项

系统支持多种元数据过滤操作：

• * 相等：$eq
• * 不相等：$ne
• * 比较：$lt、$lte、$gt、$gte
• * 列表操作：$in、$nin
• * 文本匹配：$like、$nlike
• * 逻辑操作：$and、$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"]}},
)

API 参考

参阅 langchain-mariadb API 文档 了解更多详情。