How do you use pgvector with LangChain for building AI applications?

# pip install langchain langchain-postgres langchain-openai

from langchain_postgres import PGVector
from langchain_openai import OpenAIEmbeddings
from langchain_core.documents import Document

# Connection string:
CONNECTION = "postgresql+psycopg://user:pass@localhost/mydb"

# Initialise the vector store (creates table and extension if needed)
vectorstore = PGVector(
    connection=CONNECTION,
    collection_name="documents",
    embeddings=OpenAIEmbeddings(model="text-embedding-3-small"),
    use_jsonb=True,           # store metadata in JSONB column
)

# Add documents
docs = [
    Document(page_content="pgvector enables vector search in PostgreSQL",
             metadata={"source": "docs", "category": "database"}),
    Document(page_content="HNSW is the recommended index for most use cases",
             metadata={"source": "docs", "category": "indexing"}),
]
vectorstore.add_documents(docs)

# Similarity search:
results = vectorstore.similarity_search(
    "What index should I use for fast search?",
    k=3,
)
for doc in results:
    print(doc.page_content, doc.metadata)

# Search with score:
results_with_score = vectorstore.similarity_search_with_score(
    "fast nearest neighbour search",
    k=3,
)
for doc, score in results_with_score:
    print(f"score={score:.4f}: {doc.page_content}")

# Use as a LangChain retriever:
retriever = vectorstore.as_retriever(search_kwargs={"k": 5})