Amazon Bedrock Embeddings

Generate embeddings using Amazon Bedrock's embedding models. Supports multiple model options and flexible AWS authentication methods for enterprise-grade vector representations.

Amazon Bedrock Embeddings component interface and configuration

AWS Authentication Notice: Ensure your AWS credentials have the appropriate permissions to access Amazon Bedrock services. Using IAM roles with least privilege is recommended for production deployments.

Component Inputs

Model ID: The Bedrock model identifier for embeddings
Example: "amazon.titan-embed-text-v1", "cohere.embed-english-v3"
AWS Access Key ID: Your AWS access key identifier
Example: "AKIAIOSFODNN7EXAMPLE"
AWS Secret Access Key: Your AWS secret access key
Example: "wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY"
Credentials Profile Name: Alternative to direct credentials, uses a named profile
Example: "bedrock-profile"
Region Name: AWS region where Bedrock service is accessed
Example: "us-east-1", "us-west-2"
Endpoint URL: Custom endpoint for AWS Bedrock
Example: "https://bedrock.us-west-2.amazonaws.com"

Component Outputs

Embeddings: Vector representation of the input text
Example: [0.024, -0.032, 0.067, ...]
Metadata: Additional information about the embedding generation
Example: model: amazon.titan-embed-text-v1, dimensions: 1536

Model Comparison

amazon.titan-embed-text-v1

Amazon's proprietary embedding model designed for general text embedding tasks

Dimensions: 1536
Language Support: Multilingual
Ideal for: General-purpose text embeddings and semantic search

cohere.embed-english-v3

Cohere's English-optimized embedding model available through Bedrock

Dimensions: 1024
Language Support: English-focused
Ideal for: English content retrieval and similarity matching

cohere.embed-multilingual-v3

Cohere's multilingual embedding model with support for 100+ languages

Dimensions: 1024
Language Support: 100+ languages
Ideal for: Multilingual applications and cross-lingual retrieval

Implementation Example

const embeddor = new AmazonBedrockEmbeddor({
  modelId: "amazon.titan-embed-text-v1",
  awsAccessKeyId: process.env.AWS_ACCESS_KEY_ID,
  awsSecretAccessKey: process.env.AWS_SECRET_ACCESS_KEY,
  regionName: "us-west-2"
});

// Generate embeddings
const result = await embeddor.embed({
  input: "Your text to embed"
});

// Using profile-based auth
const profileEmbeddor = new AmazonBedrockEmbeddor({
  modelId: "cohere.embed-english-v3",
  credentialsProfileName: "my-aws-profile",
  regionName: "us-east-1"
});

// Generate embeddings with the profile-based embedder
const profileResult = await profileEmbeddor.embed({
  input: "Another text to embed"
});

console.log(result.embeddings);

Use Cases

AWS-Native RAG Pipelines: Build retrieval systems within AWS ecosystem
Semantic Search: Create vector databases for content retrieval
Document Classification: Categorize documents based on semantic content
Content Recommendation: Build recommendation systems for AWS-hosted content
Multi-Modal Applications: Combine with other AWS services for comprehensive solutions

Useful Resources

Best Practices

Use environment variables for AWS credentials in production environments
Implement AWS IAM roles with least privilege for production deployments
Consider caching embeddings for frequently used content
Choose the right model based on language requirements and use case
Monitor usage costs and implement appropriate rate limiting

Documentation