Skip to content
GraphRAG Toolkit

S3 Vectors Vector Store

Topics

Section titled “Topics”

Overview

Section titled “Overview”

You can use Amazon S3 Vectors as a vector store.

Creating an S3 Vectors vector store

Section titled “Creating an S3 Vectors vector store”

Use the VectorStoreFactory.for_vector_store() static factory method to create an instance of an Amazon S3 Vectors vector store.

To create an Amazon S3 Vectors store, supply a connection string in the following format:

s3vectors://<bucket_name>[/<index_prefix>]

For example:

from graphrag_toolkit.lexical_graph.storage import VectorStoreFactory


s3_vectors_connection_info = 's3vectors://my-s3-vectors-bucket/app1'


with VectorStoreFactory.for_vector_store(s3_vectors_connection_info) as vector_store:
    ...

Connection string parameters

Section titled “Connection string parameters”

The connection string includes two parameters:

bucket_name
Section titled “bucket_name”

Mandatory. Name of an Amazon S3 vector bucket in the same AWS Region as the application running the graphrag-toolkit. If the vector bucket does not already exist, the indexing process will create a new bucket.

index_prefix
Section titled “index_prefix”

Optional. Prefix to be attached to the name of each index created by the indexing process. Prefixes allow you to store indexes created by different graphrag-toolkit applications in the same vector bucket.

Imagine an application with two tenants - the default tenant, and an admin tenant - and a connection to an S3 Vectors vector store that uses the following connection string:

s3vectors://my-s3-vectors-bucket

Because the vector store connection string is configured with a bucket name only, the application will create the following chunk indexes:

  • chunk
  • chunk-admin

If the connection string includes a prefix, like this -

s3vectors://my-s3-vectors-bucket/app1

the application will create the following chunk indexes:

  • app1.chunk
  • app1.chunk-admin

IAM permissions required to use Amazon S3 Vectors as a vector store

Section titled “IAM permissions required to use Amazon S3 Vectors as a vector store”

Indexing

Section titled “Indexing”

The identity under which the graphrag-toolkit’s indexing process runs requires the following IAM permissions:

  • s3Vectors:GetVectorBucket
  • s3Vectors:CreateVectorBucket
  • s3Vectors:GetIndex
  • s3Vectors:CreateIndex
  • s3Vectors:DeleteVectors
  • s3Vectors:GetVectors
  • s3Vectors:PutVectors

Querying

Section titled “Querying”

The identity under which the graphrag-toolkit’s querying process runs requires the following IAM permissions:

  • s3Vectors:GetVectorBucket
  • s3Vectors:GetIndex
  • s3Vectors:QueryVectors
  • s3Vectors:GetVectors

See Identity and Access management in S3 Vectors for more details on AWS security best practices for S3 Vectors.