langchain-polardb-pg

LangChain integration for Alibaba Cloud PolarDB for PostgreSQL — providing in-database embeddings, vector store, and model management powered by the polar_ai extension.

Features

• In-Database Embeddings — compute text embeddings directly inside PolarDB, eliminating external API calls and network round-trips.
• Dual Embedding Modes — automatically selects the optimal path based on your cluster edition:
  • ai_text_embedding mode (Standard edition): inline embedding in a single SQL statement
  • call_model mode (Enterprise edition + AI node): embedding via ai_callmodel
• Vector Store — full-featured similarity search, MMR, filtering, built on community langchain-postgres.
• Model Management — register, configure, and manage AI models through Python APIs.
• Auto-Discovery — from_instance() resolves cluster endpoints and capabilities via Alibaba Cloud OpenAPI.

Installation

pip install langchain-polardb-pg

With cloud auto-discovery support (recommended):

pip install langchain-polardb-pg[cloud]

Quick Start

from langchain_polardb_pg import PolarDBPGEngine, PolarDBPGEmbeddings, PolarDBPGVector

# 1. Connect to your PolarDB instance
engine = PolarDBPGEngine.from_instance(
    cluster_id="pc-xxxxx",
    database="mydb",
    user="user",
    password="password",
    access_key_id="your-ak",
    access_key_secret="your-sk",
)

# 2. Create in-database embeddings (no external API needed!)
embeddings = PolarDBPGEmbeddings.create_sync(engine)

# 3. Initialize vector store table
engine.init_vectorstore_table("my_docs", vector_size=1536)

# 4. Create vector store
store = PolarDBPGVector.create_sync(engine, embeddings, table_name="my_docs")

# 5. Add documents & search
store.add_texts(["PolarDB is a cloud-native database", "LangChain is an AI framework"])
results = store.similarity_search("cloud database", k=2)

Embedding Modes

The mode is auto-resolved from your cluster attributes. You can also specify it explicitly:

from langchain_polardb_pg.embeddings import EmbeddingMode

embeddings = PolarDBPGEmbeddings.create_sync(engine, mode=EmbeddingMode.AI_TEXT_EMBEDDING)

Components

PolarDBPGEngine

Extends the community PGEngine with:

• from_instance() — auto-discover endpoints via Alibaba Cloud OpenAPI
• cluster_attribute — cached cluster metadata (region, edition, AI capabilities)
• get_ai_api_key() — unified AI key retrieval (from cluster or environment)

PolarDBPGEmbeddings

Implements LangChain Embeddings interface:

• embed_query(text) / embed_documents(texts) — standard embedding APIs
• embed_query_inline(text) — returns a SQL expression for inline embedding (ai_text_embedding mode only)

PolarDBPGVector

Inherits all capabilities from community PGVectorStore:

• add_documents() / add_texts() — with in-database embedding
• similarity_search() / similarity_search_with_score()
• max_marginal_relevance_search() — diversity-aware retrieval
• delete() — by IDs or metadata filter

PolarDBPGModelManager

Model lifecycle management:

• create_model() / alter_model() / drop_model()
• set_model_token() / call_model()
• list_models() / get_model()

Connection Methods

# Method 1: Auto-discovery (recommended)
engine = PolarDBPGEngine.from_instance(
    cluster_id="pc-xxxxx",
    database="mydb", user="user", password="pass",
    access_key_id="ak", access_key_secret="sk",
)

# Method 2: Direct connection string
engine = PolarDBPGEngine.from_connection_string(
    "postgresql+asyncpg://user:pass@host:5432/mydb"
)

# Method 3: From existing SQLAlchemy engine
engine = PolarDBPGEngine.from_engine(existing_async_engine)

Requirements

• Python >= 3.9
• PolarDB for PostgreSQL with polar_ai extension
• Dependencies: langchain-postgres, langchain-core, sqlalchemy[asyncio], asyncpg

Development

Install local development dependencies:

python -m pip install -U pip
python -m pip install -e ".[dev,cloud]"

Run unit tests:

make test

Run LangChain standard vector store integration tests with a live PolarDB instance:

make integration-test

Build and validate distribution metadata:

make build
make check-dist

License

Apache License 2.0