Monitor on-premises and multi-cloud AI agents with AgentCore Observability

This article explains how to set up observability for AI agents running outside AWS using AWS Distro for OpenTelemetry (ADOT) to route telemetry to Amazon Bedrock AgentCore Observability.

• Use ADOT auto-instrumentation to capture agent traces, spans, and logs from non-AWS environments
• Configure IAM credentials and OpenTelemetry environment variables for SigV4 authentication with CloudWatch
• Route telemetry through CloudWatch OTLP endpoint to AgentCore Observability dashboard
• Supports agents on-premises, GCP, Azure, and other cloud providers with internet access
• Validated with Strands Agents framework running on Google Cloud Shell and simulated on-premises environments
• Provides unified visibility into agent reasoning, tool invocations, token usage, and latency across all deployment environments

ADOT-based observability enables cross-platform AI agent monitoring with identical telemetry to AgentCore runtime-hosted agents, requiring only pip install and environment variable configuration.