AWS Bedrock

Amazon Bedrock is AWS's managed AI service that provides access to foundation models from Amazon and other providers. If you're running on AWS and want embeddings that integrate with your existing infrastructure—IAM roles, VPCs, CloudWatch—Bedrock is the natural choice.

The default model is Amazon's Titan embedding model, but Bedrock also provides access to Cohere and other models through a unified API.

Setup

You'll need an AWS account with Bedrock enabled in your region and model access granted for the embedding models you want to use. Authentication uses standard AWS credentials.

Install the Bedrock SDK package:

bun add @ai-sdk/amazon-bedrock

Configure the provider in your unrag.config.ts:

import { defineUnragConfig } from "./lib/unrag/core";

export const unrag = defineUnragConfig({
  // ...
  embedding: {
    provider: "bedrock",
    config: {
      model: "amazon.titan-embed-text-v2:0",
      timeoutMs: 15_000,
    },
  },
} as const);

Configuration options

model specifies which Bedrock model to use. If not set, the provider checks the BEDROCK_EMBEDDING_MODEL environment variable, then falls back to amazon.titan-embed-text-v2:0.

timeoutMs sets the request timeout in milliseconds.

dimensions requests a specific number of dimensions for models that support it.

normalize controls whether the output vectors are normalized. Normalized vectors have unit length, which can improve similarity calculations.

embedding: {
  provider: "bedrock",
  config: {
    model: "amazon.titan-embed-text-v2:0",
    dimensions: 512,
    normalize: true,
    timeoutMs: 20_000,
  },
},

Available models

Amazon Titan Embed Text v2 (amazon.titan-embed-text-v2:0) is Amazon's latest embedding model. It supports multiple output dimensions (256, 512, or 1024) and optional normalization.

Cohere models are also available through Bedrock if you've enabled them in your AWS account. Use the Bedrock model IDs like cohere.embed-english-v3.

Check the AWS Bedrock console for the full list of available embedding models in your region.

Authentication

Bedrock uses standard AWS authentication. The SDK automatically uses the credential chain, which means:

On AWS: When running on EC2, Lambda, ECS, or other AWS services, the SDK uses the attached IAM role automatically.

Locally: Configure credentials via the AWS CLI (aws configure), environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY), or a credentials file.

No explicit API key is needed—authentication is handled through AWS IAM.

Environment variables

AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY: Standard AWS credentials when running outside of AWS.

AWS_REGION: The AWS region to use for Bedrock requests.

BEDROCK_EMBEDDING_MODEL (optional): Overrides the model specified in code.

# .env (when running outside AWS)
AWS_ACCESS_KEY_ID="..."
AWS_SECRET_ACCESS_KEY="..."
AWS_REGION="us-east-1"
BEDROCK_EMBEDDING_MODEL="amazon.titan-embed-text-v2:0"

When to use Bedrock

Choose Bedrock when you're already on AWS and want your AI workloads to integrate with your existing infrastructure. You get IAM-based access control, CloudWatch logging, and the ability to keep all your services within AWS.

If you're not on AWS, the setup overhead probably isn't worth it—use a simpler provider like OpenAI or Cohere directly.