How do you integrate LanceDB with LangChain for RAG?

LangChain provides a LanceDB vector store class that wraps a LanceDB table behind LangChain's standard vector store interface, letting LanceDB slot into an existing LangChain RAG pipeline as the retrieval backend with minimal glue code.

from langchain_community.vectorstores import LanceDB
import lancedb
from langchain_openai import OpenAIEmbeddings

db = lancedb.connect("./my_lancedb")
embeddings = OpenAIEmbeddings()

vectorstore = LanceDB(connection=db, embedding=embeddings, table_name="documents")
vectorstore.add_texts(["LanceDB is an embedded vector database.", "It stores data in the Lance format."])

results = vectorstore.similarity_search("What is LanceDB built on?", k=2)

Under the hood, LangChain's wrapper handles converting between its own Document objects (text plus metadata) and LanceDB's table rows, calling the configured embedding model to generate vectors on add_texts, and translating LangChain's similarity_search calls into the equivalent LanceDB vector query.

Because the wrapper accepts an existing LanceDB connection and table name, a table created and populated outside of LangChain (using the native LanceDB SDK directly, perhaps with the embedding function registry) can also be opened and used from LangChain, which is useful when a team wants direct SDK control over ingestion, indexing, and schema, while still using LangChain's chain/agent orchestration on top.