AWS Bedrock AgentCore
By Amazon Web Services
Amazon Bedrock AgentCore is an AWS service layer for deploying and operating AI agents at production scale — providing runtime, memory, identity, and tool-orchestration infrastructure on top of Amazon Bedrock's foundation models.
Definition
Amazon Bedrock AgentCore is an AWS service layer for deploying and operating AI agents at production scale — providing runtime, memory, identity, and tool-orchestration infrastructure on top of Amazon Bedrock's foundation models.
Overview
AWS Bedrock AgentCore extends Amazon Bedrock beyond simple model inference into the infrastructure needed to run autonomous or semi-autonomous AI agent workloads reliably in production. Where Bedrock's core service focuses on hosting and calling foundation models, AgentCore addresses the operational concerns that appear once an agent needs to take multi-step actions: a durable runtime for long-running agent sessions, persistent memory across interactions, identity and permission management for the tools an agent can call, and observability into what an agent did and why. This matters because production agents typically need to call external APIs, query internal databases, or trigger downstream workflows — not just generate text. AgentCore provides managed building blocks for that orchestration layer, similar in spirit to how Azure AI Foundry and Vertex AI Agent Builder address the same problem on their respective clouds. It's designed to work with agents built using popular open-source agent frameworks, rather than locking developers into a single proprietary agent-authoring syntax. Because agentic systems raise distinct security and reliability questions compared to single-turn chatbots — for example, ensuring an agent only accesses the tools and data it's authorized for, and can be audited after the fact — AgentCore packages identity and access controls specifically for agent tool calls, which is a newer and less mature area of the generative AI stack. Teams building on AWS often pair AgentCore with the broader agentic design patterns taught in AI Agents & Agentic Workflows. As with much of the agent-infrastructure space, AWS has continued to iterate on AgentCore's specific feature set and naming since its introduction, so developers should check current AWS documentation for the latest capabilities before architecting around it.
Key Features
- Managed runtime for long-running, stateful AI agent sessions
- Persistent memory layer so agents retain context across interactions
- Identity and access management scoped to individual agent tool calls
- Framework-agnostic support for popular open-source agent-building libraries
- Observability and tracing for auditing multi-step agent decisions
- Built on top of Amazon Bedrock's foundation model catalog