AWS Bedrock AgentCore adds observability for multi-cloud and on-premises AI agents

Amazon Bedrock AgentCore Observability is expanding its coverage to include AI agents running outside AWS, enabling operators to centralize telemetry in the same dashboard used for AWS-hosted agents.

Agents built with frameworks such as Strands Agents, LangGraph, and CrewAI can now send session traces, span metrics, and token usage to the AgentCore Observability dashboard even when deployed on Google Cloud Platform, Microsoft Azure, on-premises Kubernetes, or developer workstations.

Previously, AgentCore Observability only natively supported agents running on the AgentCore runtime within AWS. To extend observability to non-AWS environments, teams must configure the AWS Distro for OpenTelemetry (ADOT) auto-instrumentation in those environments and route telemetry to the dashboard using IAM credentials.

The setup involves configuring ADOT in non-AWS environments, validating the telemetry pipeline end to end, and then using the AgentCore Observability dashboard to monitor performance across heterogeneous deployments.

The post emphasizes that this capability is intended to reduce observability fragmentation by allowing a single pane of glass for AI agents regardless of their runtime environment.