# Deploying AWS Bedrock Agents with Pulumi

## Introduction

This guide shows how to deploy AWS Bedrock Agents using Pulumi and TypeScript.

## Prerequisites

* AWS Account with Bedrock access
    
* Node.js and npm installed
    
* Pulumi CLI installed
    

## Project Setup

1. Generate Project with Pulumi  
    `pulumi new aws-typescript`  
    `cd bedrock-agent-project`
    

This creates a complete TypeScript project with:

* package.json with all dependencies
    
* Pulumi.yaml configuration
    
* index.ts entry point
    
* TypeScript configuration
    

2. Project Structure
    
    ```bash
    bedrock-agent-project/
    ├── package.json
    ├── Pulumi.yaml
    ├── index.ts
    ├── tsconfig.json
    └── agent/
    ├── agent-deploy.ts
    └── instructions.txt
    ```
    

## Pulumi Configuration

Complete the file Pulumi.yaml with the following content:

```yaml
name: My first agent
description: "Stack used to deploy my agent"
config:
aws_region: us-east-1
aws_accountId: "123456789012"
```

Replace `123456789012` with your actual AWS account ID and the region by the one you want to use

```typescript
agent/agent-deploy.ts
import  as pulumi from "@pulumi/pulumi";
import  as aws from "@pulumi/aws";

// Configuration
const config = new pulumi.Config();
const awsRegion = config.require("aws_region");
const accountId = config.require("aws_accountId");
const stack = pulumi.getStack();

// Create Bedrock Agent
const bedrockAgent = new aws.bedrock.AgentAgent(my-agent-${stack}, {
agentName: my-agent-${stack},
description: "A helpful AI assistant",
agentResourceRoleArn: "arn:aws:iam::YOUR_ACCOUNT_ID:role/YOUR_BEDROCK_EXECUTION_ROLE", // Replace with your IAM role ARN
idleSessionTtlInSeconds: 500,
foundationModel: arn:aws:bedrock:${awsRegion}:${accountId}:inference-profile/eu.claude-3-5-sonnet-20241022-v2:0,
instruction: "You are a helpful assistant"
});

export const agentId = bedrockAgent.id;
export const agentArn = bedrockAgent.agentArn;
```

```typescript
index.ts
Replace the default content with:
import { agentId, agentArn } from "./agent/agent-deploy";
export { agentId, agentArn };
```

## **Required IAM Role**

Important: You must create the IAM role before deploying the agent, otherwise the deployment will fail.

**Create IAM Role**

1. Go to AWS IAM Console
    
2. Create a new role with these settings:
    
    * Trusted entity: AWS service
        
    * Service: Bedrock
        
    * Use case: Bedrock - Agents
        
3. Attach this policy to the role:
    

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

4. Name your role (for example, MyBedrockAgentRole)
    
5. Update the agentResourceRoleArn in your code with the full ARN
    

## Deployment

Deploy your Bedrock Agent using this command  
`pulumi up`

## Testing

After deployment:

1. Go to AWS Bedrock console
    
2. Navigate to Agents
    
3. Find your deployed agent
    
4. Test using the built-in chat interface
    

## Cleaning

To clean up, don’t forget to run the command `pulumi down`

## Conclusion

That’s it!  
You now have a deployed Bedrock Agent that can have conversations with users. The agent uses the Claude 3.5 Sonnet model and follows the instructions you provided.

Useful link: [https://www.pulumi.com/registry/packages/aws/api-docs/bedrock/agentagent/](https://www.pulumi.com/registry/packages/aws/api-docs/bedrock/agentagent/)