LoongSuite Semantic Conventions is an open-source semantic conventions repository from Alibaba that provides enhanced GenAI (Generative AI) observability standards, building upon the foundation of OpenTelemetry Semantic Conventions.
This repository supplements the OpenTelemetry Semantic Conventions with additional GenAI-specific semantic conventions derived from Alibaba's internal observability practices. It defines a comprehensive set of semantic attributes, spans, metrics, and events specifically designed for tracing and monitoring Generative AI applications and services.
- GenAI-Focused: Specialized semantic conventions for LLM applications, model interactions, and AI service observability
- Production-Tested: Based on real-world experience from Alibaba's internal AI infrastructure
- OTel-Compatible: Built on OpenTelemetry standards, ensuring seamless integration with the OTel ecosystem
- Community-Driven: Open-sourced to benefit the broader GenAI observability community
- Tracing LLM API calls (OpenAI, Alibaba Cloud Bailian, Spring AI, etc.)
- Monitoring token usage and model performance
- Tracking prompt engineering and response quality
- Observing multi-modal AI interactions
- Measuring AI service latency and costs
The human-readable semantic conventions documentation is available in the docs folder. These Markdown documents are generated from YAML model definitions located in the model folder, following the OpenTelemetry Semantic Conventions toolchain approach.
├── model/ # YAML definitions for semantic conventions
├── docs/ # Generated documentation (Markdown)
├── templates/ # Weaver templates for code generation
└── schemas/ # JSON schemas for validation
LoongSuite Semantic Conventions extends and supplements the official OpenTelemetry Semantic Conventions. While OTel provides foundational semantic conventions for distributed tracing, metrics, and logs, LoongSuite focuses specifically on GenAI observability patterns that emerged from Alibaba's production AI workloads.
- Extended GenAI Attributes: Additional attributes for AI model characteristics, prompt engineering, and response quality
- Alibaba Cloud Integration: Specific conventions for Alibaba Cloud Bailian and other cloud AI services
- Multi-Modal Support: Conventions for image, video, and audio AI interactions
- Cost Tracking: Detailed attributes for tracking AI service costs and resource consumption
- Quality Metrics: Conventions for measuring AI output quality, relevance, and safety
We welcome contributions from the community! Whether you're adding new GenAI conventions, improving documentation, or fixing issues, your input is valuable.
See CONTRIBUTING.md for detailed guidelines on:
- How to propose new semantic conventions
- Code of conduct
- Development workflow
- Testing and validation requirements
This project is licensed under the Apache License 2.0
This project builds upon the excellent work of the OpenTelemetry Semantic Conventions project and the broader OpenTelemetry community. We are grateful for their foundational contributions to observability standardization.
Maintained by Alibaba Observability Team