LlamaIndex Vector Store - LanceDB

0.5.0 verified Fri May 01 auth: no python

LanceDB vector store integration for LlamaIndex. Current version 0.5.0. Provides a LanceDBVectorStore class for storing and querying vector embeddings. Release cadence is irregular; follows LlamaIndex package structure. Supports hybrid search and column filtering in LanceDB.

pip install llama-index-vector-stores-lancedb

Common errors

error ModuleNotFoundError: No module named 'llama_index.vector_stores.lancedb' ↓

cause Package not installed correctly or import path is wrong for version < 0.4.0.

fix

Install the package: pip install llama-index-vector-stores-lancedb. For versions older than 0.4.0, use: from llama_index.vector_stores.lancedb.base import LanceDBVectorStore

error TypeError: LanceDBVectorStore.__init__() got an unexpected keyword argument 'connection' ↓

cause The 'connection' parameter was removed in v0.4.0; constructor now expects 'uri' and 'table_name'.

fix

Change to: LanceDBVectorStore(uri='your_uri', table_name='your_table')

error lancedb.error.LanceError: Invalid argument: Dimension mismatch: expected 768 but got 1536 ↓

cause Embedding dimension provided to LanceDB table does not match the dimension of the vectors being inserted.

fix

Set the dimension explicitly when creating the vector store or table: LanceDBVectorStore(uri=uri, table_name=table_name, dim=1536)

Warnings

breaking LanceDBVectorStore no longer accepts 'connection' parameter; use 'uri' instead. As of v0.4.0, the constructor requires 'uri' and 'table_name' directly. ↓

fix Replace `LanceDBVectorStore(connection=db)` with `LanceDBVectorStore(uri=uri, table_name=table_name)`.

gotcha The vector dimension must match the embedding model dimension. LanceDB does not automatically validate dimensions; mismatches cause runtime errors. ↓

fix Ensure embedding dimension passed to LanceDBVectorStore matches your embedding model's dimension. Default is 1536 for OpenAI but can be set via 'dim' parameter.

gotcha Table must exist before using LanceDBVectorStore; the store does not automatically create tables. Use lancedb.connect(uri).create_table(table_name, data) first. ↓

fix Create the table explicitly using the LanceDB Python API before using the vector store.

Imports

LanceDBVectorStore

wrong

from llama_index.vector_stores.lancedb.base import LanceDBVectorStore

correct

from llama_index.vector_stores.lancedb import LanceDBVectorStore

Older versions used a submodule path; current canonical import is from the package directly.

LanceDBVectorStore

from llama_index.vector_stores.lancedb import LanceDBVectorStore

Correct import path.

Quickstart

Minimal working example: connect to LanceDB, create a vector store, build an index, and query.

import lancedb
from llama_index.core import SimpleDirectoryReader, StorageContext, VectorStoreIndex
from llama_index.vector_stores.lancedb import LanceDBVectorStore

# Create LanceDB connection and table
uri = "data/sample-lancedb"
db = lancedb.connect(uri)
table_name = "my_table"
table = db.create_table(table_name, data=[{"vector": [0.0]*768, "text": "placeholder"}], mode="overwrite")

# Initialize vector store
vector_store = LanceDBVectorStore(uri=uri, table_name=table_name)

# Create storage context
storage_context = StorageContext.from_defaults(vector_store=vector_store)

# Load documents and build index
documents = SimpleDirectoryReader("data").load_data()
index = VectorStoreIndex.from_documents(documents, storage_context=storage_context)

# Query
query_engine = index.as_query_engine()
response = query_engine.query("What is LanceDB?")
print(response)