Integrate PolarDB-X with AI development frameworks

Step 1: Install dependencies

Install the package that matches the framework you use.

• langchain-polardbx requires Python 3.9 or later.

• llama-index-vector-stores-polardbx requires Python 3.10 or later.

Step 2: Create a vector table and insert data

1. Enable vector indexes: Before you create a vector table, make sure that the PolarDB-X instance supports the vector index feature. Vector indexes are disabled by default in PolarDB-X. You must run the following command with an administrator account to enable them. The setting takes effect immediately for new connections, and no instance restart is required.

   SET GLOBAL vidx_disabled = OFF;

2. Example:

   Note

   • In the following examples, replace host, port, user, password, and database with the actual connection information of your instance.

   • distance_strategy specifies the distance strategy. Valid values are cosine, euclidean, and inner_product. The inner_product distance strategy and advanced features such as index monitoring require a relatively recent version of PolarDB-X. The integration packages automatically detect the available capabilities and adapt.

   LangChain

   The following example calls add_texts to insert text. The integration package automatically calls the embedding model to generate vectors and writes them to PolarDB-X.

   from langchain_polardbx import PolarDBXVectorStore
   from langchain_openai import OpenAIEmbeddings
   
   embeddings = OpenAIEmbeddings()
   
   vectorstore = PolarDBXVectorStore(
       host="<polardb-x-host>",
       port=<polardb-x-port>,
       user="<polardb-x-username>",
       password="<polardb-x-password>",
       database="<polardb-x-databasename>",
       embedding=embeddings,
       table_name="<polardb-x-tablename>",
       distance_strategy="cosine",  # or "euclidean", "inner_product"
   )
   
   # Insert text (automatically calls the embedding model to generate vectors and writes them to PolarDB-X)
   vectorstore.add_texts(["Hello world", "PolarDB-X supports vector search"])

   LlamaIndex

   The following example creates an index from documents. Chunking, embedding, and writing are performed automatically.

   from llama_index.vector_stores.polardbx import PolarDBXVectorStore
   from llama_index.core import StorageContext, VectorStoreIndex, Settings
   from llama_index.embeddings.openai import OpenAIEmbedding
   
   Settings.embed_model = OpenAIEmbedding()
   
   vector_store = PolarDBXVectorStore(
       host="<polardb-x-host>",
       port=<polardb-x-port>,
       user="<polardb-x-username>",
       password="<polardb-x-password>",
       database="<polardb-x-databasename>",
       table_name="<polardb-x-tablename>",
       embed_dim=1536,
       distance_strategy="cosine",
   )
   
   # Create an index from documents (automatic chunking, embedding, and writing to PolarDB-X)
   from llama_index.core import Document
   documents = [Document(text="PolarDB-X supports vector search")]
   
   storage_context = StorageContext.from_defaults(vector_store=vector_store)
   index = VectorStoreIndex.from_documents(documents, storage_context=storage_context)

Step 3: Run a similarity search

After data is written, run a similarity search.

results = vectorstore.similarity_search("PolarDB-X", k=2)
for doc in results:
    print(doc.page_content)

query_engine = index.as_query_engine()
response = query_engine.query("What is PolarDB-X?")
print(response)

Step 4 (optional): Enable SQL queries

Both packages provide the PolarDBXSQLDatabase component, which wraps DDL reflection compatibility handling for PolarDB-X and serves as the base integration layer for the SQL capabilities of the frameworks. The create_partitioned_table helper function is also provided. It supports creating HASH/KEY/RANGE/LIST partitioned tables and broadcast tables.

from langchain_polardbx import PolarDBXSQLDatabase

db = PolarDBXSQLDatabase.from_uri(
    "mysql+pymysql://<polardb-x-username>:<polardb-x-password>@<polardb-x-host>:<polardb-x-port>/<polardb-x-databasename>"
)

# List available tables
tables = db.get_usable_table_names()

# Get table schema information
info = db.get_table_info(["your_table"])

# Run SQL
result = db.run("SELECT COUNT(*) FROM your_table")

from llama_index.vector_stores.polardbx import PolarDBXSQLDatabase

db = PolarDBXSQLDatabase.from_uri(
    "mysql+pymysql://<polardb-x-username>:<polardb-x-password>@<polardb-x-host>:<polardb-x-port>/<polardb-x-databasename>"
)

# List available tables
tables = db.get_usable_table_names()

# Get table schema information
info = db.get_single_table_info("your_table")

# Run SQL
result = db.run_sql("SELECT COUNT(*) FROM your_table")

PolarDBXSQLDatabase connects directly to the SQL agent capabilities of the frameworks (such as create_sql_agent in LangChain and NLSQLTableQueryEngine in LlamaIndex) to convert natural language into SQL queries.