The OpenSearch Observability Technical Advisory Group (TAG) marks its first anniversary, recapping a year of community-driven progress including OTLP-native ingestion work, Prometheus metrics support shipped in OpenSearch 3.5, the launch of the open-source OpenSearch Observability Stack (formerly ATLAS), AI agent observability tooling (Agent Health), Dashboards-as-Code, and cross-community collaboration with Perses, KEDA, and the CNCF Observability TAG. Contributors came from AWS, Apple, SAP, Uber, CERN, Hilti, Red Hat, and others across 10+ monthly meetings. The group outlines its focus areas for year two, including deeper Prometheus integration, expanded AI/agent observability, and further OpenTelemetry alignment.
Table of contents
What is the Observability TAG?A year of milestonesThe communityGet involvedLooking aheadAuthorsQuestions this post answers
When did OpenSearch add native Prometheus metrics support?
Prometheus metrics support launched in OpenSearch 3.5, following work guided by the OpenSearch Observability TAG on a new Discover Metrics experience. This included multi-query support, label filtering, and PromQL integration in OpenSearch Dashboards, with Prometheus alerting and AlertManager integration following shortly after. Developers tracking OpenSearch release changes can follow version-specific updates on daily.dev.
What is the OpenSearch Observability Stack and what does it include?
The OpenSearch Observability Stack is a complete, open-source, production-ready observability platform built on OpenTelemetry, OpenSearch, and Prometheus, originally incubated under the name ATLAS. It offers one-command deployment, APM with service maps, distributed tracing, log analytics with PPL, metrics dashboards, alerting, and AI agent observability, with zero vendor lock-in, hosted at observability.opensearch.org. Anyone comparing open-source observability platforms can keep up with stacks like this via daily.dev.
How does OpenSearch support observability for AI agents?
OpenSearch offers AI agent observability through a project called Agent Health, providing execution graph visualization, token usage tracking, tool-call tracing, and LLM evaluation frameworks. This work aligns with OpenTelemetry GenAI semantic conventions and grew out of RFC reviews covering agent tracing architectures and MCP server integration for the observability stack. Teams building AI agent monitoring can track tooling like Agent Health through daily.dev.