What is Amazon Bedrock AgentCore?

AgentCore is a managed platform by AWS that allows developers to build, run, and monitor secure AI agents using any framework or model. It includes modular components like Runtime, Memory, Gateway, and Browser.

How does AgentCore differ from Bedrock Agents?

AgentCore provides session isolation, observability, API integrations, and identity support, which are missing from the basic Bedrock Agents. It's designed for production use in enterprise environments.

Can I use non-AWS models with AgentCore?

Yes, AgentCore supports external models and frameworks such as LangGraph and CrewAI. You're not locked into AWS models or services.

What is the AgentCore Gateway module?

The Gateway module exposes internal APIs securely for agent use, wrapping them in a standard interface using Model Context Protocol (MCP).

Amazon Bedrock AgentCore: Secure, Scalable AI Agents for Enterprise

2 min read

Amazon Bedrock AgentCore

Amazon Bedrock AgentCore (Preview) launched on July 16, 2025, marking a major evolution in how enterprises build and deploy AI agents at scale. With modular components, open framework support, and AWS-level security, AgentCore is engineered for production from day one.

What Is AgentCore?

Amazon’s fully managed AgentCore platform allows developers to build, deploy, and operate AI agents using any model or agent framework (like CrewAI, LangGraph, Strands Agents, etc.), even if the models are hosted outside Bedrock.

Core modules:

Runtime: Serverless agent execution with long-session support
Memory: Long-term and short-term memory with cost tracking
Gateway: Easily expose internal APIs as agent-ready tools
Browser: Headless browsing tool for secure web automation
Code Interpreter: Secure code execution in a sandbox
Identity: OAuth-based access to internal services
Observability: Real-time telemetry and debugging

💡

GPT-5 And AI-Agents for Business →

Why It Matters

From Prototype to Production

AgentCore takes the burden of infrastructure off your hands. No more cobbling together runtime environments, identity layers, or observability dashboards.

Full Flexibility, No Lock-In

Use open frameworks and models—inside or outside AWS—without rebuilding or replatforming.

Secure by Design

With session isolation, identity federation, and audit logging, AgentCore meets even the strictest security and compliance standards.

Marketplace Integration

Deploy agents directly from the AWS AI Agents Marketplace, which now includes over 800 agent templates ready to go.

Key Modules Explained

Runtime

Serverless containerized execution
Up to 8-hour agent sessions
Multi-step workflow support

Memory

Persistent memory
Event-based billing model
Use across sessions or teams

Gateway

Wraps internal APIs in a secure, agent-compatible interface
Model Context Protocol (MCP) ready

Identity

OAuth-based login
Integrates with AWS Cognito, Okta, etc.
Role-based permissioning

Observability

Tracks token usage, latency, and failure events
Compatible with CloudWatch and OpenTelemetry

Real-World Use Cases

Banking

Itaú Unibanco uses AgentCore to power multilingual customer support bots while complying with strict financial regulations.

Healthcare

Innovaccer exposes HIPAA-compliant APIs to agents using Gateway, enabling automated medical workflows securely.

Marketing

Epsilon uses Browser and Gateway modules to automate campaign creation and testing—cutting build time by 30%.

AgentCore vs Bedrock Agents vs DIY

Feature	AgentCore	Bedrock Agents	DIY Frameworks
Session isolation	✅ Yes	❌ No	❌ No
Built-in memory	✅ Yes	❌ Minimal	❌ Manual setup
API integration	✅ Gateway module	❌ Manual coding	❌ Custom tooling
Web automation	✅ Browser module	❌ No	❌ Custom scripting
Identity + auth	✅ OAuth, IAM	❌ None	❌ Self-managed
Monitoring & telemetry	✅ OpenTelemetry, CloudWatch	❌ Basic logs	❌ Custom pipelines

Quick Start (Simplified)

Install SDK:
pip install bedrock-agentcore
Build your agent logic using CrewAI or LangGraph
Wrap it:
BedrockAgentCoreApp()
Test locally:
agentcore launch --local
Deploy via console or:
genesis launch

Preview Pricing (until Sep 16, 2025)

Module	Price Estimate
Runtime	$0.0895 per vCPU/hr
Memory	$0.25–0.75 per 1K events
Gateway	$0.005 per 1K API calls
Identity	$0.010 per 1K tokens
Browser	$0.0895 per vCPU/hr

Billing is modular and tied to usage volume.

What’s Coming Next

VPC support for fully isolated environments
Agent2Agent communication across microservices
Multi-agent orchestration built-in
More marketplace integrations with partners

Final Thoughts

Amazon Bedrock AgentCore unlocks the infrastructure behind true agentic AI—not just simple chatbot interactions. If you’re scaling AI inside a secure enterprise context, AgentCore is the missing platform layer.

Ready to Build with AI Agents?

Run the AI Scan →
Identify automation opportunities across your workflows.