Amazon Bedrock AgentCore
AWS's fully managed platform for deploying, securing, and operating AI agents at enterprise scale.
About
Bedrock AgentCore is AWS's answer to the hardest part of agents: production. Instead of replacing frameworks, it supplies the infrastructure around them — a serverless Runtime with session isolation for long-running agents, Memory services for short- and long-term context, an Identity layer for OAuth-based access on behalf of users, a Gateway that turns APIs and Lambda functions into MCP-compatible tools, plus Code Interpreter, a cloud Browser tool, and Observability with OpenTelemetry export.
It went generally available in October 2025 and works with agents written in Strands, LangGraph, CrewAI, or plain custom code, against any model inside or outside Bedrock. Pricing is pay-per-use per module (runtime vCPU-hours, memory events, gateway invocations). The honest framing: it is the enterprise-grade, AWS-native way to run agents — powerful, but you are buying into the AWS billing and IAM model along with it.
What it can do
- AgentCore Runtime: serverless execution with session isolation
- Memory: managed short- and long-term agent memory
- Gateway: APIs and Lambda functions as MCP-compatible tools
- Identity: OAuth-based auth for agents acting on behalf of users
- Observability, Code Interpreter, and cloud Browser tool built in
Setup
In the AWS console enable Bedrock AgentCore, package your agent code for AgentCore Runtime, then attach Memory, Gateway, and Identity modules via the console, CLI, or CloudFormation.
Prerequisites
An AWS account with Bedrock access; agent code in Python (or any framework AgentCore supports).
Use cases
Works with
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: AWS managed service, GA 2025-10-13; no OSS star count applicable; officialUrl https://aws.amazon.com/bedrock/agentcore/ curl-verified 200; Runtime/Memory/Gateway/Identity/Observability/Code Interpreter/Browser modules verified against AWS announcement and independent coverage (VentureBeat, TechRadar). Manual review: framework-agnostic positioning (Strands/LangGraph/CrewAI) and MCP+A2A support are AWS-stated; pricing is pay-per-use per module. License: commercial (AWS).
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.