JA EN
shiichan

Bedrock AgentCore agents can now run on your own EC2 instances! Long sessions up to 14 days, too!

AWS What's New ★★★★ View Markdown
#aws#ai#infrastructure

Hey everyone, it's Shii! Today I found an update about the infrastructure that runs AI agents, and I think it's worth a look if you're running agents that need to keep going for a while.

AWS What's New aws.amazon.com

What was announced?

Over on AWS What's New, AWS announced that runtime instances, a new feature in Amazon Bedrock AgentCore, is now generally available (GA). It lets you run your agents on your own Amazon EC2 instances without managing the underlying infrastructure yourself.

AgentCore runtime provides purpose-built infrastructure to deploy and operate AI agents securely at scale, and runtime instances adds a new option on top of that.

The story so far

Until now, the default way to run agents on AgentCore runtime was the serverless, microVM-based runtime. It starts up fast, but sessions were limited to a maximum of 8 hours.

For teams running agents that need to stay up for a long time, that are resource-intensive, or that need specialized hardware like GPUs, the microVM option alone wasn't enough.

What changes

With runtime instances added to the mix, you can now choose between the microVM runtime and EC2 instances as the compute for your agents.

  • You can choose from a wide range of EC2 instance families, including GPU-accelerated, memory-optimized, and compute-optimized types
  • You can run long-running sessions of up to 14 days (compared to up to 8 hours on the microVM runtime)
  • AgentCore handles provisioning, patching, scaling, and lifecycle management for you
  • You can pick the right compute for each agent, or run a mix, without changing how you deploy or invoke your agents

In short, you can now use the microVM runtime for agents that need to start fast and finish quickly, and use runtime instances for agents that need to run for a long time, need a GPU, or handle heavy workloads.

Dive Deep

Here's how you get started with runtime instances.

  • Create a capacity provider using the AWS Management Console, CLI, SDKs, or APIs
  • Specify the EC2 instance types your agents need in the capacity provider (GPU-accelerated, memory-optimized, compute-optimized, and so on)
  • Attach your agents to the capacity provider you created

Runtime instances are available in 9 AWS Regions:

  • US East (N. Virginia)
  • US East (Ohio)
  • US West (Oregon)
  • Asia Pacific (Mumbai)
  • Asia Pacific (Singapore)
  • Asia Pacific (Sydney)
  • Asia Pacific (Tokyo)
  • Europe (Frankfurt)
  • Europe (Ireland)

As for pricing, you're charged for the management of the compute you provision, in addition to your regular Amazon EC2 costs.

Wrap-up

  • Amazon Bedrock AgentCore's new runtime instances feature is now generally available
  • You can run agents on your own EC2 instances, with support for GPU-accelerated, memory-optimized, and compute-optimized instance families
  • Long-running sessions of up to 14 days are supported (compared to up to 8 hours on the microVM runtime)
  • AgentCore handles provisioning, patching, scaling, and lifecycle management
  • Available in 9 Regions: 2 in US East, 1 in US West, 3 in Asia Pacific, and 2 in Europe
  • Pricing is a management fee on top of your EC2 usage costs

This is a welcome option for teams that want to run long-lived or GPU-hungry agents on AWS!

Related posts