Using Vector Stores#

LlamaIndex offers multiple integration points with vector stores / vector databases:

  1. LlamaIndex can use a vector store itself as an index. Like any other index, this index can store documents and be used to answer queries.

  2. LlamaIndex can load data from vector stores, similar to any other data connector. This data can then be used within LlamaIndex data structures.

Using a Vector Store as an Index#

LlamaIndex also supports different vector stores as the storage backend for VectorStoreIndex.

A detailed API reference is found here.

Similar to any other index within LlamaIndex (tree, keyword table, list), VectorStoreIndex can be constructed upon any collection of documents. We use the vector store within the index to store embeddings for the input text chunks.

Once constructed, the index can be used for querying.

Default Vector Store Index Construction/Querying

By default, VectorStoreIndex uses an in-memory SimpleVectorStore that’s initialized as part of the default storage context.

from llama_index import VectorStoreIndex, SimpleDirectoryReader

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

# Query index
query_engine = index.as_query_engine()
response = query_engine.query("What did the author do growing up?")

Custom Vector Store Index Construction/Querying

We can query over a custom vector store as follows:

from llama_index import VectorStoreIndex, SimpleDirectoryReader, StorageContext
from llama_index.vector_stores import DeepLakeVectorStore

# construct vector store and customize storage context
storage_context = StorageContext.from_defaults(
    vector_store=DeepLakeVectorStore(dataset_path="<dataset_path>")
)

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

# Query index
query_engine = index.as_query_engine()
response = query_engine.query("What did the author do growing up?")

Below we show more examples of how to construct various vector stores we support.

Apache Cassandra®

from llama_index.vector_stores import CassandraVectorStore
import cassio

# To use an Astra DB cloud instance through CQL:
cassio.init(database_id="1234abcd-...", token="AstraCS:...")

# For a Cassandra cluster:
from cassandra.cluster import Cluster

cluster = Cluster(["127.0.0.1"])
cassio.init(session=cluster.connect(), keyspace="my_keyspace")

# After the above `cassio.init(...)`, create a vector store:
vector_store = CassandraVectorStore(
    table="cass_v_table", embedding_dimension=1536
)

Astra DB

from llama_index.vector_stores import AstraDBVectorStore

astra_db_store = AstraDBVectorStore(
    token="AstraCS:xY3b...",  # Your Astra DB token
    api_endpoint="https://012...abc-us-east1.apps.astra.datastax.com",  # Your Astra DB API endpoint
    collection_name="astra_v_table",  # Table name of your choice
    embedding_dimension=1536,  # Embedding dimension of the embeddings model used
)

Azure Cognitive Search

from azure.core.credentials import AzureKeyCredential
from llama_index.vector_stores import AzureAISearchVectorStore

search_service_api_key = "YOUR-AZURE-SEARCH-SERVICE-ADMIN-KEY"
search_service_endpoint = "YOUR-AZURE-SEARCH-SERVICE-ENDPOINT"
search_service_api_version = "2023-11-01"
credential = AzureKeyCredential(search_service_api_key)

# Index name to use
index_name = "llamaindex-vector-demo"

client = SearchIndexClient(
    endpoint=search_service_endpoint,
    credential=credential,
)

vector_store = AzureAISearchVectorStore(
    search_or_index_client=client,
    index_name=index_name,
    embedding_dimensionality=1536,
)

Chroma

import chromadb
from llama_index.vector_stores import ChromaVectorStore

# Creating a Chroma client
# EphemeralClient operates purely in-memory, PersistentClient will also save to disk
chroma_client = chromadb.EphemeralClient()
chroma_collection = chroma_client.create_collection("quickstart")

# construct vector store
vector_store = ChromaVectorStore(
    chroma_collection=chroma_collection,
)

DashVector

import dashvector
from llama_index.vector_stores import DashVectorStore

# init dashvector client
client = dashvector.Client(
    api_key="your-dashvector-api-key",
    endpoint="your-dashvector-cluster-endpoint",
)

# creating a DashVector collection
client.create("quickstart", dimension=1536)
collection = client.get("quickstart")

# construct vector store
vector_store = DashVectorStore(collection)

DeepLake

import os
import getpath
from llama_index.vector_stores import DeepLakeVectorStore

os.environ["OPENAI_API_KEY"] = getpath.getpath("OPENAI_API_KEY: ")
os.environ["ACTIVELOOP_TOKEN"] = getpath.getpath("ACTIVELOOP_TOKEN: ")
dataset_path = "hub://adilkhan/paul_graham_essay"

# construct vector store
vector_store = DeepLakeVectorStore(dataset_path=dataset_path, overwrite=True)

DocArray

from llama_index.vector_stores import (
    DocArrayHnswVectorStore,
    DocArrayInMemoryVectorStore,
)

# construct vector store
vector_store = DocArrayHnswVectorStore(work_dir="hnsw_index")

# alternatively, construct the in-memory vector store
vector_store = DocArrayInMemoryVectorStore()

Elasticsearch

First, you can start Elasticsearch either locally or on Elastic cloud.

To start Elasticsearch locally with docker, run the following command:

docker run -p 9200:9200 \
  -e "discovery.type=single-node" \
  -e "xpack.security.enabled=false" \
  -e "xpack.security.http.ssl.enabled=false" \
  -e "xpack.license.self_generated.type=trial" \
  docker.elastic.co/elasticsearch/elasticsearch:8.9.0

Then connect and use Elasticsearch as a vector database with LlamaIndex

from llama_index.vector_stores import ElasticsearchStore

vector_store = ElasticsearchStore(
    index_name="llm-project",
    es_url="http://localhost:9200",
    # Cloud connection options:
    # es_cloud_id="<cloud_id>",
    # es_user="elastic",
    # es_password="<password>",
)

This can be used with the VectorStoreIndex to provide a query interface for retrieval, querying, deleting, persisting the index, and more.

Epsilla

from pyepsilla import vectordb
from llama_index.vector_stores import EpsillaVectorStore

# Creating an Epsilla client
epsilla_client = vectordb.Client()

# Construct vector store
vector_store = EpsillaVectorStore(client=epsilla_client)

Note: EpsillaVectorStore depends on the pyepsilla library and a running Epsilla vector database. Use pip/pip3 install pyepsilla if not installed yet. A running Epsilla vector database could be found through docker image. For complete instructions, see the following documentation: https://epsilla-inc.gitbook.io/epsilladb/quick-start

Faiss

import faiss
from llama_index.vector_stores import FaissVectorStore

# create faiss index
d = 1536
faiss_index = faiss.IndexFlatL2(d)

# construct vector store
vector_store = FaissVectorStore(faiss_index)

...

# NOTE: since faiss index is in-memory, we need to explicitly call
#       vector_store.persist() or storage_context.persist() to save it to disk.
#       persist() takes in optional arg persist_path. If none give, will use default paths.
storage_context.persist()

txtai

import txtai
from llama_index.vector_stores import TxtaiVectorStore

# create txtai index
txtai_index = txtai.ann.ANNFactory.create(
    {"backend": "numpy", "dimension": 512}
)

# construct vector store
vector_store = TxtaiVectorStore(txtai_index)

Jaguar

from llama_index.schema import TextNode
from llama_index.vector_stores.types import VectorStoreQuery
from jaguardb_http_client.JaguarHttpClient import JaguarHttpClient
from llama_index.vector_stores.jaguar import JaguarVectorStore


# construct vector store client
url = "http://127.0.0.1:8080/fwww/"
pod = "vdb"
store = "llamaindex_rag_store"
vector_index = "v"
vector_type = "cosine_fraction_float"
vector_dimension = 3

# require JAGUAR_API_KEY environment variable or file $HOME/.jagrc to hold the
# jaguar API key to connect to jaguar store server
vector_store = JaguarVectorStore(
    pod, store, vector_index, vector_type, vector_dimension, url
)

# login to jaguar server for security authentication
vector_store.login()

# create a vector store on the back-end server
metadata_fields = "author char(32), category char(16)"
text_size = 1024
vector_store.create(metadata_fields, text_size)

# store some text
node = TextNode(
    text="Return of King Lear",
    metadata={"author": "William", "category": "Tragedy"},
    embedding=[0.9, 0.1, 0.4],
)
vector_store.add(nodes=[node], use_node_metadata=True)

# make a query
qembedding = [0.4, 0.2, 0.8]
vsquery = VectorStoreQuery(query_embedding=qembedding, similarity_top_k=1)
query_result = vector_store.query(vsquery)

# make a query with metadata filter (where condition)
qembedding = [0.6, 0.1, 0.4]
vsquery = VectorStoreQuery(query_embedding=qembedding, similarity_top_k=3)
where = "author='Eve' or (author='Adam' and category='History')"
query_result = vector_store.query(vsquery, where=where)

# make a query ignoring old data (with time cutoff)
qembedding = [0.3, 0.3, 0.8]
vsquery = VectorStoreQuery(query_embedding=qembedding, similarity_top_k=3)
args = "day_cutoff=180"  # only search recent 180 days data
query_result = vector_store.query(vsquery, args=args)

# check if a vector is anomalous
text = ("Gone With The Wind",)
embed_of_text = [0.7, 0.1, 0.2]
node = TextNode(text=text, embedding=embed_of_text)
true_or_false = vector_store.is_anomalous(node)

# llama_index RAG application
from llama_index.embeddings import OpenAIEmbedding
from llama_index.storage.storage_context import StorageContext
from llama_index import VectorStoreIndex, ServiceContext

question = "What did the author do growing up?"

storage_context = StorageContext.from_defaults(vector_store=vector_store)
embed_model = OpenAIEmbedding()
embed_of_question = [0.7, 0.1, 0.2]
service_context = ServiceContext.from_defaults(embed_model=embed_model)
db_documents = vector_store.load_documents(embed_of_question, 10)
index = VectorStoreIndex.from_documents(
    db_documents,
    storage_context=storage_context,
    service_context=service_context,
)

query_engine = index.as_query_engine()
print(f"Question: {question}")
response = query_engine.query(question)
print(f"Answer: {str(response)}")

# logout to clean up resources
vector_store.logout()

Note: Client(requires jaguardb-http-client) <–> Http Gateway <–> JaguarDB Server Client side needs to run: “pip install -U jaguardb-http-client”

Milvus

  • Milvus Index offers the ability to store both Documents and their embeddings.

import pymilvus
from llama_index.vector_stores import MilvusVectorStore

# construct vector store
vector_store = MilvusVectorStore(
    uri="https://localhost:19530", overwrite="True"
)

Note: MilvusVectorStore depends on the pymilvus library. Use pip install pymilvus if not already installed. If you get stuck at building wheel for grpcio, check if you are using python 3.11 (there’s a known issue: https://github.com/milvus-io/pymilvus/issues/1308) and try downgrading.

MongoDBAtlas

# Provide URI to constructor, or use environment variable
import pymongo
from llama_index.vector_stores.mongodb import MongoDBAtlasVectorSearch
from llama_index.indices.vector_store.base import VectorStoreIndex
from llama_index.storage.storage_context import StorageContext
from llama_index.readers.file.base import SimpleDirectoryReader

# mongo_uri = os.environ["MONGO_URI"]
mongo_uri = (
    "mongodb+srv://<username>:<password>@<host>?retryWrites=true&w=majority"
)
mongodb_client = pymongo.MongoClient(mongo_uri)

# construct store
store = MongoDBAtlasVectorSearch(mongodb_client)
storage_context = StorageContext.from_defaults(vector_store=store)
uber_docs = SimpleDirectoryReader(
    input_files=["../data/10k/uber_2021.pdf"]
).load_data()

# construct index
index = VectorStoreIndex.from_documents(
    uber_docs, storage_context=storage_context
)

MyScale

import clickhouse_connect
from llama_index.vector_stores import MyScaleVectorStore

# Creating a MyScale client
client = clickhouse_connect.get_client(
    host="YOUR_CLUSTER_HOST",
    port=8443,
    username="YOUR_USERNAME",
    password="YOUR_CLUSTER_PASSWORD",
)


# construct vector store
vector_store = MyScaleVectorStore(myscale_client=client)

Neo4j

  • Neo4j stores texts, metadata, and embeddings and can be customized to return graph data in the form of metadata.

from llama_index.vector_stores import Neo4jVectorStore

# construct vector store
neo4j_vector = Neo4jVectorStore(
    username="neo4j",
    password="pleaseletmein",
    url="bolt://localhost:7687",
    embed_dim=1536,
)

Pinecone

import pinecone
from llama_index.vector_stores import PineconeVectorStore

# Creating a Pinecone index
api_key = "api_key"
pinecone.init(api_key=api_key, environment="us-west1-gcp")
pinecone.create_index(
    "quickstart", dimension=1536, metric="euclidean", pod_type="p1"
)
index = pinecone.Index("quickstart")

# construct vector store
vector_store = PineconeVectorStore(pinecone_index=index)

Qdrant

import qdrant_client
from llama_index.vector_stores import QdrantVectorStore

# Creating a Qdrant vector store
client = qdrant_client.QdrantClient(
    host="<qdrant-host>", api_key="<qdrant-api-key>", https=True
)
collection_name = "paul_graham"

# construct vector store
vector_store = QdrantVectorStore(
    client=client,
    collection_name=collection_name,
)

Redis

First, start Redis-Stack (or get url from Redis provider)

docker run --name redis-vecdb -d -p 6379:6379 -p 8001:8001 redis/redis-stack:latest

Then connect and use Redis as a vector database with LlamaIndex

from llama_index.vector_stores import RedisVectorStore

vector_store = RedisVectorStore(
    index_name="llm-project",
    redis_url="redis://localhost:6379",
    overwrite=True,
)

This can be used with the VectorStoreIndex to provide a query interface for retrieval, querying, deleting, persisting the index, and more.

SingleStore

from llama_index.vector_stores import SingleStoreVectorStore
import os

# can set the singlestore db url in env
# or pass it in as an argument to the SingleStoreVectorStore constructor
os.environ["SINGLESTOREDB_URL"] = "PLACEHOLDER URL"
vector_store = SingleStoreVectorStore(
    table_name="embeddings",
    content_field="content",
    metadata_field="metadata",
    vector_field="vector",
    timeout=30,
)

Timescale

from llama_index.vector_stores import TimescaleVectorStore

vector_store = TimescaleVectorStore.from_params(
    service_url="YOUR TIMESCALE SERVICE URL",
    table_name="paul_graham_essay",
)

Upstash

from llama_index.vector_stores import UpstashVectorStore

vector_store = UpstashVectorStore(url="YOUR_URL", token="YOUR_TOKEN")

Weaviate

import weaviate
from llama_index.vector_stores import WeaviateVectorStore

# creating a Weaviate client
resource_owner_config = weaviate.AuthClientPassword(
    username="<username>",
    password="<password>",
)
client = weaviate.Client(
    "https://<cluster-id>.semi.network/",
    auth_client_secret=resource_owner_config,
)

# construct vector store
vector_store = WeaviateVectorStore(weaviate_client=client)

Zep

Zep stores texts, metadata, and embeddings. All are returned in search results.

from llama_index.vector_stores.zep import ZepVectorStore

vector_store = ZepVectorStore(
    api_url="<api_url>",
    api_key="<api_key>",
    collection_name="<unique_collection_name>",  # Can either be an existing collection or a new one
    embedding_dimensions=1536,  # Optional, required if creating a new collection
)

storage_context = StorageContext.from_defaults(vector_store=vector_store)

index = VectorStoreIndex.from_documents(
    documents, storage_context=storage_context
)

# Query index using both a text query and metadata filters
filters = MetadataFilters(
    filters=[ExactMatchFilter(key="theme", value="Mafia")]
)
retriever = index.as_retriever(filters=filters)
result = retriever.retrieve("What is inception about?")

Zilliz

  • Zilliz Cloud (hosted version of Milvus) uses the Milvus Index with some extra arguments.

import pymilvus
from llama_index.vector_stores import MilvusVectorStore


# construct vector store
vector_store = MilvusVectorStore(
    uri="foo.vectordb.zillizcloud.com",
    token="your_token_here",
    overwrite="True",
)

Example notebooks can be found here.

Loading Data from Vector Stores using Data Connector#

LlamaIndex supports loading data from a huge number of sources. See Data Connectors for more details and API documentation.

Chroma stores both documents and vectors. This is an example of how to use Chroma:

from llama_index.readers.chroma import ChromaReader
from llama_index.indices import SummaryIndex

# The chroma reader loads data from a persisted Chroma collection.
# This requires a collection name and a persist directory.
reader = ChromaReader(
    collection_name="chroma_collection",
    persist_directory="examples/data_connectors/chroma_collection",
)

query_vector = [n1, n2, n3, ...]

documents = reader.load_data(
    collection_name="demo", query_vector=query_vector, limit=5
)
index = SummaryIndex.from_documents(documents)

query_engine = index.as_query_engine()
response = query_engine.query("<query_text>")
display(Markdown(f"<b>{response}</b>"))

Qdrant also stores both documents and vectors. This is an example of how to use Qdrant:

from llama_index.readers.qdrant import QdrantReader

reader = QdrantReader(host="localhost")

# the query_vector is an embedding representation of your query_vector
# Example query_vector
# query_vector = [0.3, 0.3, 0.3, 0.3, ...]

query_vector = [n1, n2, n3, ...]

# NOTE: Required args are collection_name, query_vector.
# See the Python client: https;//github.com/qdrant/qdrant_client
# for more details

documents = reader.load_data(
    collection_name="demo", query_vector=query_vector, limit=5
)

NOTE: Since Weaviate can store a hybrid of document and vector objects, the user may either choose to explicitly specify class_name and properties in order to query documents, or they may choose to specify a raw GraphQL query. See below for usage.

# option 1: specify class_name and properties

# 1) load data using class_name and properties
documents = reader.load_data(
    class_name="<class_name>",
    properties=["property1", "property2", "..."],
    separate_documents=True,
)

# 2) example GraphQL query
query = """
{
    Get {
        <class_name> {
            <property1>
            <property2>
        }
    }
}
"""

documents = reader.load_data(graphql_query=query, separate_documents=True)

NOTE: Both Pinecone and Faiss data loaders assume that the respective data sources only store vectors; text content is stored elsewhere. Therefore, both data loaders require that the user specifies an id_to_text_map in the load_data call.

For instance, this is an example usage of the Pinecone data loader PineconeReader:

from llama_index.readers.pinecone import PineconeReader

reader = PineconeReader(api_key=api_key, environment="us-west1-gcp")

id_to_text_map = {
    "id1": "text blob 1",
    "id2": "text blob 2",
}

query_vector = [n1, n2, n3, ...]

documents = reader.load_data(
    index_name="quickstart",
    id_to_text_map=id_to_text_map,
    top_k=3,
    vector=query_vector,
    separate_documents=True,
)

Example notebooks can be found here.