Amazon Web Services has updated Amazon Bedrock AgentCore runtime to start agent sessions faster, support larger workloads and reduce memory use while agents wait. The managed service runs each session in an isolated environment, so developers can deploy agents without managing the underlying compute infrastructure.
AWS says sessions now start up to 15 times faster and in under two seconds, regardless of container image size or the number of concurrent sessions. A session can use up to 8 vCPUs and 32 GB of memory, four times the compute capacity of the previous version. Container images can be up to 10 GB.
Each session still runs in its own hardware-isolated Firecracker micro virtual machine, with its own identity, access policy and audit trail, according to AWS. The larger sessions are intended for work such as compiling codebases, running browser workflows and performing local model inference.
Idle agents scale to zero to reduce memory use
The runtime now reclaims memory within minutes after an agent releases it, and idle agents scale to zero. AWS says this can reduce memory consumed per session by up to 68 percent for agents that pause between interactions. For agents that run continuously, committed baseline pricing bills a minimum memory allocation and allows bursts of up to four times that amount. AWS did not give specific prices.

Agents can also mount a filesystem dynamically and save their work to it. AWS gives an Amazon S3 bucket with FUSE support as an example. The company says this keeps data within the session’s trust boundary throughout a job, under one set of controls.
The runtime is available now across multiple regions
The updated runtime is available now in AWS Regions across North America, South America, Europe and Asia Pacific. AWS says it plans to add per-session storage scoping by re:Invent, using session-labeled credentials to bind each session to its own folder.



