opentelemetry-util-genai

What it is and what it does

OpenTelemetry Util for GenAI is a standardization layer for instrumenting generative AI libraries. It wraps LLM and embedding invocations to automatically generate spans, metrics, and events following OpenTelemetry semantic conventions. The package provides a TelemetryHandler that manages the lifecycle of AI operations, capturing request parameters, response metadata, token usage, and optionally message content (controlled via environment variables). It also supports uploading prompts and completions to external storage via fsspec-compatible paths.

The package is designed to reduce boilerplate when adding observability to GenAI applications. Instead of manually creating spans and recording metrics for each LLM call, developers use context managers or manual lifecycle methods provided by TelemetryHandler. It sets standard span attributes (operation name, provider, model, server details, token counts) and supports both chat and embedding operations. Message content capture is opt-in and configurable, and event emission can be controlled independently.

Use it for:

• Add observability to applications calling OpenAI, Anthropic, or other LLM providers without writing custom instrumentation code.
• Track token usage and latency across LLM invocations for cost monitoring and performance analysis.
• Archive prompts and completions to cloud storage (GCS, S3, local filesystem) for audit, fine-tuning, or debugging.
• Emit structured events with message content to a LoggerProvider for downstream processing or analysis.
• Standardize telemetry across multiple GenAI libraries using OpenTelemetry semantic conventions.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Provides standardized instrumentation utilities for generative AI libraries, managing telemetry (spans, metrics, events) around LLM and embedding invocations via OpenTelemetry.

Yes, if you are building observability into a GenAI application and want to follow OpenTelemetry standards. The package is actively maintained, has no known vulnerabilities, and low install friction. It is in beta status (1.0b0), so expect the API to stabilize; verify compatibility with your specific GenAI providers and exporters before production use.

Install

opentelemetry-util-genai on PyPI

pip install opentelemetry-util-genai

uv add opentelemetry-util-genai

poetry add opentelemetry-util-genai

Installing opentelemetry-util-genai

Before you install

Low install friction with four lightweight runtime dependencies (opentelemetry-api, opentelemetry-instrumentation, opentelemetry-semantic-conventions, wrapt). Package is actively maintained with recent commits and is in beta status.

License in practice

Licensed under Apache-2.0 (permissive), allowing commercial and private use with minimal restrictions.

Quickstart

pip install opentelemetry-util-genai

from opentelemetry.util.genai.handler import TelemetryHandler
from opentelemetry.util.genai.types import InferenceInvocation

handler = TelemetryHandler()
with handler.chat():
    # Your LLM invocation here
    pass

Requires Python 3.10 or later. OpenTelemetry API and instrumentation packages must be installed and configured with appropriate exporters for telemetry to be collected.

Verify before relying

• Whether the package works with all major GenAI providers or only specific ones (OpenAI shown in examples).
• Performance overhead of telemetry collection at scale.
• Compatibility with custom OpenTelemetry exporters beyond the standard setup.

Package facts

Evidence: opentelemetry_util_genai-1.0b0-py3-none-any.whl

Tags

More Monitoring packages