
Managed platform for building, connecting, and operating AI agents.
AgentCore provides a fully managed runtime, gateway, identity, and memory layer, removing infrastructure management and scaling from zero to thousands of isolated agent sessions.
It accepts frameworks and models from multiple providers rather than only Amazon's own, and charges compute only while agents actively consume resources.
AgentCore runs only on AWS infrastructure with no self-hosted option, and its observability and storage defaults tie deployments to CloudWatch, ECR, and S3.
Metered gateway, policy, and search charges stack on top of compute, and the instances option adds a management fee above EC2 prices.
Amazon announced Bedrock AgentCore in preview on July 16, 2025, introducing managed runtime, memory, identity, gateway, browser, and code-interpreter services for deploying AI agents at scale.
The service reached general availability on October 13, 2025, in nine AWS Regions with its SDK passing one million downloads, and later added payments, policy, and evaluation capabilities.
AgentCore uses consumption-based pricing with no upfront commitments or minimum fees, billing agent runtime microVMs per second of active consumption while leaving input-output wait time unbilled.
As of September 2026, metered add-ons cover gateway invocations, web search, identity requests, memory events, and policy authorization checks, and the instances option carries a management fee above EC2 pricing.
Amazon Bedrock AgentCore is a fully managed platform for building, deploying, and operating production AI agents with any framework and any foundation model.
It pairs a session-isolated agent runtime with managed gateway, identity, memory, browser, and code-interpreter services, and supports frameworks including CrewAI, LangGraph, LlamaIndex, and the OpenAI Agents SDK.