Unified observability in Amazon OpenSearch Service: metrics, traces, and AI agent debugging in a single interface

This article announces unified observability capabilities in Amazon OpenSearch Service, integrating metrics, traces, and AI agent debugging into a single interface.

• OpenSearch UI now combines Prometheus metrics, logs, and traces in one observability workspace
• AI agent tracing shows full reasoning chains, LLM calls, tool invocations, and token consumption
• OpenTelemetry gen_ai semantic conventions enable AI-specific observability signals
• Application Map visualizes system topology with service health and latency indicators
• PromQL queries run natively alongside logs and traces without data duplication
• Drill-down capability traces issues from service-level health to specific failing spans
• Python SDK provides one-line setup with decorator-based tracing for AI agents
• Auto-instrumentation supports OpenAI, Anthropic, Bedrock, LangChain, and LlamaIndex
• Metrics remain in Prometheus, logs/traces in OpenSearch; UI federates queries at runtime
• OpenSearch Ingestion routes telemetry signals to appropriate backends via OTLP

This release enables comprehensive troubleshooting for both AI agents and traditional microservices through a unified interface built on open standards.