This version is still in development and is not considered stable yet. For the latest snapshot version, please use Spring AI 1.1.2!

Amazon Bedrock Knowledge Base

This section walks you through setting up the Amazon Bedrock Knowledge Base VectorStore to perform similarity searches against a pre-configured Knowledge Base.

Amazon Bedrock Knowledge Bases is a fully managed RAG (Retrieval-Augmented Generation) capability that allows you to connect foundation models to your data sources. Unlike other vector stores, Bedrock Knowledge Base handles document ingestion, chunking, and embedding internally.

Prerequisites

  1. AWS Account with Bedrock access enabled

  2. A configured Bedrock Knowledge Base with at least one data source synced

  3. AWS credentials configured (via environment variables, AWS config file, or IAM role)

This vector store is read-only. Documents are managed through the Knowledge Base’s data source sync process, not through the add() or delete() methods.

Auto-configuration

Spring AI provides Spring Boot auto-configuration for the Bedrock Knowledge Base Vector Store. To enable it, add the following dependency to your project’s Maven pom.xml file:

<dependency>
    <groupId>org.springframework.ai</groupId>
    <artifactId>spring-ai-starter-vector-store-bedrock-knowledgebase</artifactId>
</dependency>

or to your Gradle build.gradle build file:

dependencies {
    implementation 'org.springframework.ai:spring-ai-starter-vector-store-bedrock-knowledgebase'
}
Refer to the Dependency Management section to add the Spring AI BOM to your build file.

Unlike other vector stores, Bedrock Knowledge Base does not require an EmbeddingModel bean. The Knowledge Base handles embeddings internally during data source synchronization.

To connect to your Knowledge Base, provide the Knowledge Base ID via Spring Boot’s application.properties:

spring.ai.vectorstore.bedrock-knowledge-base.knowledge-base-id=YOUR_KNOWLEDGE_BASE_ID
spring.ai.vectorstore.bedrock-knowledge-base.region=us-east-1

Or via environment variables:

export SPRING_AI_VECTORSTORE_BEDROCK_KNOWLEDGE_BASE_KNOWLEDGE_BASE_ID=YOUR_KNOWLEDGE_BASE_ID

Now you can auto-wire the Vector Store in your application:

@Autowired VectorStore vectorStore;

// ...

// Retrieve documents similar to a query
List<Document> results = vectorStore.similaritySearch(
    SearchRequest.builder()
        .query("What is the return policy?")
        .topK(5)
        .build());

Configuration Properties

You can use the following properties in your Spring Boot configuration to customize the Bedrock Knowledge Base vector store.

Property Description Default value

spring.ai.vectorstore.bedrock-knowledge-base.knowledge-base-id

The ID of the Bedrock Knowledge Base to query

-

spring.ai.vectorstore.bedrock-knowledge-base.region

AWS region for the Bedrock service

SDK default

spring.ai.vectorstore.bedrock-knowledge-base.top-k

Number of results to return

5

spring.ai.vectorstore.bedrock-knowledge-base.similarity-threshold

Minimum similarity score (0.0 to 1.0)

0.0

spring.ai.vectorstore.bedrock-knowledge-base.search-type

Search type: SEMANTIC or HYBRID

null (KB default)

spring.ai.vectorstore.bedrock-knowledge-base.reranking-model-arn

ARN of Bedrock reranking model

null (disabled)

Search Types

Bedrock Knowledge Base supports two search types:

  • SEMANTIC - Vector similarity search only (default)

  • HYBRID - Combines semantic search with keyword search

HYBRID search is only available with OpenSearch-based vector stores. S3 Vectors, Aurora PostgreSQL, and other vector store types only support SEMANTIC search.

 
spring.ai.vectorstore.bedrock-knowledge-base.search-type=HYBRID

Reranking

You can improve search relevance by enabling a Bedrock reranking model:

spring.ai.vectorstore.bedrock-knowledge-base.reranking-model-arn=arn:aws:bedrock:us-west-2::foundation-model/amazon.rerank-v1:0

Available reranking models:

  • Amazon Rerank 1.0 - Available in us-west-2, ap-northeast-1, ca-central-1, eu-central-1

  • Cohere Rerank 3.5 - Requires AWS Marketplace subscription

Metadata Filtering

You can leverage the generic, portable metadata filters with the Bedrock Knowledge Base store.

For example, you can use the text expression language:

vectorStore.similaritySearch(
    SearchRequest.builder()
        .query("travel policy")
        .topK(5)
        .similarityThreshold(0.5)
        .filterExpression("department == 'HR' && year >= 2024")
        .build());

or programmatically using the Filter.Expression DSL:

FilterExpressionBuilder b = new FilterExpressionBuilder();

vectorStore.similaritySearch(
    SearchRequest.builder()
        .query("travel policy")
        .topK(5)
        .filterExpression(b.and(
            b.eq("department", "HR"),
            b.gte("year", 2024)).build())
        .build());

Supported Filter Operators

Spring AI Bedrock Description

EQ

equals

Equal to

NE

notEquals

Not equal to

GT

greaterThan

Greater than

GTE

greaterThanOrEquals

Greater than or equal

LT

lessThan

Less than

LTE

lessThanOrEquals

Less than or equal

IN

in

Value in list

NIN

notIn

Value not in list

AND

andAll

Logical AND

OR

orAll

Logical OR

NOT

(negation)

Logical NOT

Metadata filtering requires documents in your Knowledge Base to have metadata attributes. For S3 data sources, create .metadata.json files alongside your documents.

Manual Configuration

If you prefer to configure the vector store manually, you can do so by creating the beans directly.

Add this dependency to your project:

<dependency>
    <groupId>org.springframework.ai</groupId>
    <artifactId>spring-ai-bedrock-knowledgebase-store</artifactId>
</dependency>
Refer to the Dependency Management section to add the Spring AI BOM to your build file.

Sample Code

@Bean
public BedrockAgentRuntimeClient bedrockAgentRuntimeClient() {
    return BedrockAgentRuntimeClient.builder()
        .region(Region.US_EAST_1)
        .build();
}

@Bean
public VectorStore vectorStore(BedrockAgentRuntimeClient client) {
    return BedrockKnowledgeBaseVectorStore.builder(client, "YOUR_KNOWLEDGE_BASE_ID")
        .topK(10)
        .similarityThreshold(0.5)
        .searchType(SearchType.SEMANTIC)
        .build();
}

Then use the vector store:

List<Document> results = vectorStore.similaritySearch(
    SearchRequest.builder()
        .query("What are the company holidays?")
        .topK(3)
        .build());

for (Document doc : results) {
    System.out.println("Content: " + doc.getText());
    System.out.println("Score: " + doc.getScore());
    System.out.println("Source: " + doc.getMetadata().get("source"));
}

Accessing the Native Client

The Bedrock Knowledge Base Vector Store provides access to the underlying native client through the getNativeClient() method:

BedrockKnowledgeBaseVectorStore vectorStore = context.getBean(BedrockKnowledgeBaseVectorStore.class);
Optional<BedrockAgentRuntimeClient> nativeClient = vectorStore.getNativeClient();

if (nativeClient.isPresent()) {
    BedrockAgentRuntimeClient client = nativeClient.get();
    // Use the native client for Bedrock-specific operations
}

Limitations

  • Read-only: The add() and delete() methods throw UnsupportedOperationException. Documents are managed through the Knowledge Base’s data source sync process.

  • HYBRID search: Only available with OpenSearch-based vector stores.

  • Reranking availability: Model availability varies by AWS region.

Supported Data Sources

Bedrock Knowledge Base supports multiple data source types. The source location is included in document metadata:

Data Source Metadata Field Example

S3

source

s3://bucket/path/document.pdf

Confluence

source

confluence.example.com/page/123

SharePoint

source

sharepoint.example.com/doc/456

Salesforce

source

salesforce.example.com/record/789

Web Crawler

source

example.com/page

Custom

source

Custom document ID