AI Observability
The Observability Stack provides specialized tooling for observing AI agent workflows, built on the OpenTelemetry GenAI semantic conventions.
Capabilities
Section titled “Capabilities”- Agent tracing: Visualize LLM agent execution as trace trees, DAG graphs, and timelines
- GenAI semantic conventions: Standard
gen_ai.*attributes for model, tokens, tools, and conversations - MCP server: Query OpenSearch from AI agents via the built-in Model Context Protocol server
- Python and JavaScript SDKs: Instrument your AI applications with purpose-built SDKs
How it works
Section titled “How it works”AI agent traces use the same OpenTelemetry infrastructure as service traces. Your application emits spans with gen_ai.* attributes, which flow through the OTel Collector and Data Prepper into OpenSearch. The Agent Traces plugin in OpenSearch Dashboards provides purpose-built views for exploring these traces.
Getting started
Section titled “Getting started”- Agent Tracing — the Agent Traces UI in OpenSearch Dashboards
- Agent Graph & Path — DAG visualization, trace tree, and timeline views
- MCP Server — query OpenSearch from AI agents via MCP
- Python SDK — instrument Python AI applications
- JavaScript SDK — instrument JavaScript AI applications