LangGraph ReAct Agent on AWS Bedrock AgentCore - Pulumi Infrastructure

Overview

This Pulumi program provisions production-ready infrastructure for deploying a LangGraph ReAct agent to AWS Bedrock AgentCore. It creates all necessary AWS resources using infrastructure as code, including Knowledge Bases, Lambda functions, and containerized agent runtime.

Getting Started

Prerequisites

• AWS CLI configured with appropriate credentials
• Docker installed and running
• Pulumi CLI installed
• Python 3.12+
• UV package manager (installation guide)

Setup

1. Deploy CloudFormation stack (Knowledge Bases, Lambdas, Guardrails):

   aws cloudformation create-stack \
     --stack-name pet-store-kb \
     --template-body file://pet_store_knowledge_bases.yaml \
     --capabilities CAPABILITY_IAM

2. Sync Knowledge Bases:

   • Navigate to AWS Console → Bedrock → Knowledge Bases
   • Sync both Product Info and Pet Care knowledge bases

3. Install dependencies:

   uv sync

4. Configure Pulumi with CloudFormation outputs:

   uv run python setup_config.py pet-store-kb

5. Login to Pulumi (if needed):

   pulumi login

6. Preview infrastructure:

   pulumi preview

7. Deploy infrastructure:

   pulumi up

Architecture

The infrastructure includes:

Core Components

• AWS Bedrock AgentCore Runtime: Containerized execution environment for the LangGraph agent
• ECR Repository: Stores the ARM64 Docker image for the agent
• Knowledge Bases (2):
  • Product Information KB
  • Pet Care KB
• Lambda Functions (2):
  • Inventory Management System
  • User Management System
• Guardrails: Content filtering and safety controls
• IAM Roles & Policies: Secure access control for all components

Key Features

• ARM64 Architecture: Optimized for AWS Graviton processors
• OpenTelemetry Integration: Full observability with CloudWatch GenAI dashboard
• Automated Docker Build: Uses Pulumi Docker Build provider for image building
• Production Security: Proper IAM roles, policies, and guardrails
• UV Package Manager: Fast, reliable dependency management

Resources Created

1. ECR Repository: Container image storage with lifecycle policy
2. Docker Image: LangGraph agent with OpenTelemetry instrumentation
3. AgentCore Runtime: Serverless runtime integrated with Knowledge Bases, Lambda, and Guardrails
4. Configuration: Pulumi stack configuration from CloudFormation outputs

Deployment

Prerequisites

• CloudFormation stack deployed and Knowledge Bases synced
• AWS CLI configured with appropriate credentials
• Docker installed and running
• Pulumi CLI installed
• UV package manager (already configured)

Deploy

# Preview changes
pulumi preview

# Deploy infrastructure
pulumi up

# Get outputs
pulumi stack output

Key Outputs

• agent_runtime_arn: Full ARN for agent invocation
• agent_runtime_name: Runtime name for evaluation
• ecr_repository_url: Container image repository

Agent Capabilities

The deployed agent handles:

• Product Queries: Retrieves product info from knowledge base
• Inventory Checks: Real-time inventory via Lambda
• User Management: Customer lookup by ID and email
• Pet Care Advice: Provides advice from knowledge base
• Order Processing: Calculates pricing, discounts, and shipping
• Business Rules:
  • 15% discount on orders over $300
  • 10% multi-unit discount
  • Dynamic shipping costs
  • Inventory replenishment alerts

Observability

The agent includes full OpenTelemetry instrumentation:

• Service Name: langgraph-agent
• Dashboard: CloudWatch GenAI Observability
• Traces: Full request/response tracing
• Metrics: Performance and usage metrics

Testing

Invoke the agent runtime:

import boto3
import json
import uuid

client = boto3.client('bedrock-agentcore')
response = client.invoke_agent_runtime(
    agentRuntimeArn='<from-pulumi-output>',
    qualifier='DEFAULT',
    traceId=str(uuid.uuid4()),
    contentType='application/json',
    payload=json.dumps({
        "prompt": "What's the price of Doggy Delights?"
    })
)

Troubleshooting

• "Knowledge Base not found": Ensure KB synced in AWS Console
• "Docker build failed": Check Docker daemon and ECR permissions
• "Runtime creation failed": Verify IAM role permissions and CloudFormation outputs

Cleanup

pulumi destroy

Notes

• The agent code is located in the pet_store_agent directory
• Docker build uses ARM64 platform for optimal AgentCore performance
• All resources follow AWS best practices
• Configuration is automatically populated from CloudFormation stack outputs