OpenTelemetry GenAI Semantic Conventions
OpenTelemetry GenAI Semantic Conventions is a repository that defines semantic conventions for generative AI, covering spans, metrics, and events for GenAI clients, MCP, and provider-specific conventions like OpenAI. It extends the core OpenTelemetry Semantic Conventions using Weaver. This is intended for developers building observability tooling for GenAI applications.
✨ Key features
- Defines semantic conventions for GenAI spans, metrics, and events.
- Covers GenAI clients, MCP, and provider-specific conventions (e.g., OpenAI).
- Extends core OpenTelemetry Semantic Conventions using Weaver.
- Includes human-readable docs generated from YAML models.
- Provides reference implementations and compliance matrices.
- Offers contribution guidelines for the reference project.
🎯 Use cases
- Instrumenting GenAI applications to emit standardized telemetry.
- Building OpenTelemetry-compatible exporters for GenAI providers.
- Creating dashboards and alerts based on GenAI semantic conventions.
- Ensuring interoperability of GenAI observability tools.
- Contributing to the evolution of GenAI semantic conventions.
⚠️ Good to know
The repository is in early stages; the schema URL is marked as TODO, and some docs are generated from YAML definitions.
❓ FAQ
What does this repository contain?
It contains semantic conventions for GenAI, including spans, metrics, and events, plus docs and reference implementations.
How does it relate to the core OpenTelemetry Semantic Conventions?
It extends the core conventions with GenAI-specific ones, using Weaver to manage dependencies.
Where can I find the human-readable documentation?
The docs are in the 'docs' folder, generated from YAML definitions in the 'model' folder.
Are there reference implementations?
Yes, reference implementations and tooling are under 'reference', with a Python compliance matrix in reference/README.md.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.