Amazon Bedrock AgentCore

Overview

Amazon Bedrock AgentCore is an agentic platform for building, deploying, and operating effective AI agents securely at scale—no infrastructure management needed. It provides framework-agnostic primitives that work with popular open-source frameworks (Strands, LangGraph, CrewAI, Autogen) and any model.

Purpose: Transform any AI agent into a production-ready application with enterprise-grade infrastructure

Pattern: Capabilities-based (6 independent service modules)

Key Principles (validated by AWS December 2025):

1. Framework Agnostic - Works with any agent framework or model
2. Zero Infrastructure - Fully managed, no ops overhead
3. Session Isolation - Complete data isolation between sessions
4. Enterprise Security - VPC, PrivateLink, identity integration
5. Composable Services - Use only what you need
6. Production Ready - Built for scale, reliability, and security

Quality Targets:

• Deployment: < 5 minutes from code to production
• Latency: Low-latency to 8-hour async workloads
• Observability: Full CloudWatch integration

---

When to Use

Use bedrock-agentcore when:

• Building production AI agents on AWS
• Need managed infrastructure for agent deployment
• Require session isolation and enterprise security
• Want to use existing agent frameworks (LangGraph, CrewAI, etc.)
• Need browser automation or code execution capabilities
• Integrating with existing identity providers

When NOT to Use:

• Simple Bedrock model invocations (use bedrock-runtime)
• Standard Bedrock Agents with action groups (use bedrock-agent)
• Non-AWS deployments

---

Prerequisites

Required

• AWS account with Bedrock access
• IAM permissions for AgentCore services
• Python 3.10+ (for SDK)

Recommended

• bedrock-agentcore-sdk-python installed
• bedrock-agentcore-starter-toolkit CLI
• Foundation model access enabled (Claude, etc.)

Installation

# Install SDK and CLI
pip install bedrock-agentcore strands-agents bedrock-agentcore-starter-toolkit

# Verify installation
agentcore --help

---

Core Services

1. AgentCore Runtime

Secure, session-isolated compute for running agent code.

Boto3 Client:

import boto3

# Data plane operations
client = boto3.client('bedrock-agentcore')

# Control plane operations
control = boto3.client('bedrock-agentcore-control')

Create Agent Runtime:

# Using starter toolkit
# agentcore configure -e main.py -n my-agent
# agentcore deploy

# Using boto3 control plane
response = control.create_agent_runtime(
    name='my-production-agent',
    description='Customer service agent',
    agentRuntimeArtifact={
        's3': {
            'uri': 's3://my-bucket/agent-package.zip'
        }
    },
    roleArn='arn:aws:iam::123456789012:role/AgentCoreExecutionRole',
    pythonRuntime='PYTHON_3_13',
    entryPoint=['main.py']
)
agent_runtime_arn = response['agentRuntimeArn']

Invoke Agent:

# Invoke deployed agent
response = client.invoke_agent_runtime(
    agentRuntimeArn='arn:aws:bedrock-agentcore:us-east-1:123456789012:agent-runtime/xxx',
    runtimeSessionId='session-123',
    payload={
        'prompt': 'What is my order status?',
        'context': {'user_id': 'user-456'}
    }
)

result = response['payload']
print(result)

Agent Entry Point Structure:

from bedrock_agentcore import BedrockAgentCoreApp
from strands import Agent

app = BedrockAgentCoreApp(debug=True)
agent = Agent()

@app.entrypoint
def invoke(payload):
    """Main agent entry point"""
    user_message = payload.get("prompt", "Hello!")
    app.logger.info(f"Processing: {user_message}")

    result = agent(user_message)
    return {"result": result.message}

if __name__ == "__main__":
    app.run()

---

2. AgentCore Gateway

Transforms existing APIs and Lambda functions into agent-compatible tools with semantic search discovery.

Create Gateway:

response = control.create_gateway(
    name='customer-service-gateway',
    description='Gateway for customer service tools',
    protocolType='REST'
)
gateway_arn = response['gatewayArn']

Add Gateway Target (Tool):

# Add an existing Lambda as a tool
response = control.create_gateway_target(
    gatewayId='gateway-xxx',
    name='GetOrderStatus',
    description='Retrieves order status by order ID',
    targetConfiguration={
        'lambdaTarget': {
            'lambdaArn': 'arn:aws:lambda:us-east-1:123456789012:function:GetOrder'
        }
    },
    toolSchema={
        'name': 'get_order_status',
        'description': 'Get the current status of a customer order',
        'inputSchema': {
            'type': 'object',
            'properties': {
                'order_id': {
                    'type': 'string',
                    'description': 'The unique order identifier'
                }
            },
            'required': ['order_id']
        }
    }
)

**

...