AWS Bedrock Converse API Example

Created by Wayne Workman

This Terraform module deploys an AWS Lambda function that uses Amazon Bedrock to detect prompt injection attempts in user input. The module implements the security principles outlined in this hands-on demo.

A minimal Python script demonstrating how to use AWS Bedrock's Converse API with conversation context tracking.

Purpose

This script serves as a quick reference implementation showing:

• How to use the AWS Bedrock Converse API
• How to maintain conversation context across multiple interactions
• Proper message format for the Converse API
• Basic error handling for API calls

Features

• Interactive chat interface with AWS Bedrock Nova model
• Maintains full conversation history for contextual responses
• Simple configuration through constants
• Clean error handling

Default Configuration

The script is configured by default to use:

• Region: us-east-2
• Model: Nova Lite (us.amazon.nova-lite-v1:0)

About Nova Lite

As of August 14, 2025, Nova Lite costs:

• Input: $0.00006 per 1K tokens
• Output: $0.00024 per 1K tokens

To put this in perspective: generating a standard 80,000-word novel (~100K tokens) would cost approximately $0.024 (2.4 cents) in output tokens. Nova Lite is one of the cheapest models available on AWS Bedrock, making it ideal for learning purposes and very light workloads.

Requirements

• Python 3.x
• boto3 library
• AWS credentials configured with access to Bedrock
• Important: You must enable the Nova Lite model in AWS Bedrock before using this script

AWS Permissions

The minimum IAM permissions required to run this script:

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "bedrock:InvokeModel",
                "bedrock:InvokeModelWithResponseStream"
            ],
            "Resource": "arn:aws:bedrock:us-east-2:*:foundation-model/us.amazon.nova-lite-v1:0"
        }
    ]
}

Usage

python converse_example.py

Type quit, exit, or bye to end the conversation.

Configuration

The script uses the following configurable constants:

• MODEL_NAME: Display name for the model (default: 'Nova')
• MODEL_ID: AWS Bedrock model identifier (default: Nova Lite)
• AWS_REGION: AWS region for Bedrock service (default: us-east-2)
• INFERENCE_CONFIG: Model inference parameters
  • maxTokens: 4096 (maximum response length)
  • temperature: 0.7 (creativity/randomness level)
  • topP: 0.9 (nucleus sampling parameter)

License

See the LICENSE file for licensing information.