Proposal: genai-telemetry - GenAI/LLM Observability SDK

[kamal-s-bisht]

Project Proposal: genai-telemetry

Requested maturity level: Sandbox

Summary

genai-telemetry is a platform-agnostic observability SDK for tracing GenAI/LLM applications. It enables developers to export traces, token usage, costs, and performance metrics to multiple observability backends including Splunk, Elasticsearch, OpenTelemetry, Datadog, Prometheus, Grafana Loki, and AWS CloudWatch.

Key Features

• Simple decorator-based instrumentation for LLM calls, embeddings, retrievers, tools, chains, and agents
• Multi-platform export support
• Automatic token extraction from OpenAI and Anthropic responses
• Lightweight design with minimal dependencies

Links

• GitHub: https://github.com/genai-telemetry/genai-telemetry
• PyPI: https://pypi.org/project/genai-telemetry/
• License: Apache 2.0

Alignment with LF AI & Data

This project supports the foundation's mission by democratizing GenAI observability, promoting interoperability across platforms, and enabling responsible AI through comprehensive monitoring.

Collaboration Opportunities

Identified integrations with OpenLineage, MLflow, Trusted AI, Flyte, and ONNX.

I am available to present to the TAC at your convenience.
Signed-off-by: kamal-s-bisht k.bisht7@gmail.com