Amazon Bedrock Integration

This guide walks you through integrating Amazon Bedrock with NudgeBee's LLM Server and RAG Server applications.

Overview

Amazon Bedrock is a fully managed service that offers a choice of high-performing foundation models (FMs) from leading AI companies through a unified API. Bedrock simplifies the process of building generative AI applications while maintaining privacy and security.

Prerequisites

AWS account with Amazon Bedrock access enabled
IAM user with appropriate Bedrock permissions
AWS Access Key ID and Secret Access Key

Setting up Amazon Bedrock Credentials

Request Access to Models:
- Navigate to the Amazon Bedrock console in AWS
- Go to "Model access" in the navigation pane
- Request access to the foundation models you want to use
- Wait for approval (typically immediate for most models)

Create IAM User with Bedrock Permissions:

aws iam create-user --user-name bedrock-user

Attach one of the following policies to grant Bedrock access:

Option A: Custom Inline Policy (Recommended — Least Privilege)

Create a policy file bedrock-policy.json:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Sid": "VisualEditor0",
            "Effect": "Allow",
            "Action": [
                "bedrock:InvokeModel",
                "bedrock:InvokeModelWithResponseStream"
            ],
            "Resource": "*"
        }
    ]
}

Attach the inline policy:

aws iam put-user-policy --user-name bedrock-user --policy-name BedrockInvokeAccess --policy-document file://bedrock-policy.json

Option B: AWS Managed Policy

aws iam attach-user-policy --user-name bedrock-user --policy-arn arn:aws:iam::aws:policy/AmazonBedrockLimitedAccess

Generate Access Keys:
```
aws iam create-access-key --user-name bedrock-user
```
Save the returned Access Key ID and Secret Access Key securely.

Model ID vs Inference Profile

When calling models on Amazon Bedrock, the model name you provide depends on your throughput setup:

Inference Profile (default): If you have not purchased dedicated/provisioned throughput, you can use an inference profile ID (recommended) or a bare model ID as the model name. Inference profiles are prefixed with the region shorthand (e.g., us., eu.).
- Example LLM (Meta Llama): us.meta.llama3-8b-instruct-v1:0
- Example LLM (Anthropic Claude): us.anthropic.claude-sonnet-4-6-20250514-v1:0
- Example Embeddings: us.amazon.titan-embed-text-v2:0
- You can find available inference profile IDs in the Bedrock console under Inference profiles, or by running:
```
aws bedrock list-inference-profiles --region <your-region>
```
Dedicated/Provisioned Throughput: If you have purchased provisioned throughput for a model, use the provisioned model ARN as the model name.
- Example: arn:aws:bedrock:<region>:<account-id>:provisioned-model/<model-name>

Recommended: While bare model IDs (e.g., meta.llama3-8b-instruct-v1:0) work for on-demand inference within the same region, using inference profile IDs is recommended for cross-region routing and better availability. Some newer models may require inference profiles.

Integrating with LLM Server

Configure Bedrock in LLM Server:

Add the following configuration to your LLM Server settings:

LLM_PROVIDER=bedrock
LLM_PROVIDER_REGION=<AWS_Region> # e.g., us-west-2
LLM_MODEL_NAME=<Inference_Profile_ID_or_Provisioned_ARN> # e.g., us.meta.llama3-8b-instruct-v1:0

Integrating with RAG Server

Configure Bedrock in RAG Server:

Add the following configuration to your RAG Server settings:

EMBEDDINGS_PROVIDER=bedrock
EMBEDDINGS_PROVIDER_REGION=<AWS_Region> # e.g., us-west-2
EMBEDDINGS_MODEL_NAME=<Inference_Profile_ID_or_Provisioned_ARN> # e.g., us.amazon.titan-embed-text-v2:0

Overview​

Prerequisites​

Setting up Amazon Bedrock Credentials​

Model ID vs Inference Profile​

Integrating with LLM Server​

Integrating with RAG Server​

To deploy NudgeBee AI models on AWS Bedrock and integrate NudgeBee Model Deployment​