The OpenTelemetry Generative AI semantic conventions define shared telemetry vocabulary for generative AI operations. They help runtime teams avoid one-off span names and metric fields that cannot be correlated across providers or applications.
At a glance
- Organization
- OpenTelemetry
- Runtime role
- Telemetry semantic conventions
- Category
- Observability and Evaluation
- Official documentation
- Visit official documentation opens in a new tab
Where it fits in the runtime stack
Cross-layer observability. These conventions support tracing and metrics around model requests, providers, usage, and runtime behavior.
Primary runtime role
Use the conventions as a baseline for span attributes, provider metadata, token usage, latency analysis, and cross-system observability.
Not the same as
Semantic conventions do not provide an observability backend, redaction policy, or evaluation system.
Integration notes
- Map runtime trace events to OpenTelemetry spans and attributes where the semantics match.
- Redact prompts, tool outputs, credentials, and personal data before exporting telemetry.
- Keep cost, token, model route, and policy-decision fields queryable for incident review.
Questions before production use
- Which spans represent model calls, retrieval, tools, routing, and human review?
- Which telemetry fields must be redacted or retained for shorter periods?
- Can traces correlate model events with application, database, and infrastructure spans?
Review and deprecation posture
This profile is reviewed as part of the aRuntime.com quarterly resource audit. If the official documentation moves, the project is archived, or the resource changes scope, this page should be updated with a dated status note rather than silently removed.
Sources and further reading
- Generative AI semantic conventions opens in a new tab — OpenTelemetry; official repository documentation; accessed 2026-06-20 UTC.
Last reviewed: .