openinference-instrumentation

What it is and what it does

OpenInference Instrumentation is a utility library that bridges typed annotation and evaluation objects into OpenTelemetry span attributes following the OpenInference specification. It provides helper functions to flatten Annotation objects (with fields like name, score, label, explanation, annotator_kind, and metadata) into span attributes at different scopes (span, trace, or session), and context managers to attach session IDs, user IDs, custom metadata, tags, and prompt template information to spans during execution.

The package is designed for teams instrumenting LLM applications and AI systems that need to attach structured evaluation results, annotations, and operational context to their traces. It depends on opentelemetry-api, opentelemetry-sdk, and openinference-semantic-conventions to ensure compatibility with standard observability tooling. The library handles JSON serialization of metadata and manages scope-specific prefixes automatically, reducing boilerplate in tracing code.

Use it for:

• Attach LLM evaluation scores and hallucination labels to spans for post-hoc analysis of model outputs.
• Track multi-turn conversations by wrapping code blocks with session and user context managers.
• Log prompt templates with versions and variable bindings to spans for prompt management and debugging.
• Add custom metadata and tags to traces for filtering and grouping related requests in observability platforms.
• Standardize annotation attributes across multiple instrumentation points using the Annotation model.

Worth the install?

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

Provides utility functions to convert typed annotation and evaluation objects into OpenInference span attributes, and context managers to attach session, user, metadata, and prompt template information to traced spans.

Yes. The package is actively maintained, has low install friction, carries a permissive Apache-2.0 license, and provides a well-designed abstraction for a common need in LLM observability. Use it if you are already using OpenTelemetry and need to attach structured annotations or evaluation data to spans in a standardized way.

Install

openinference-instrumentation on PyPI

pip install openinference-instrumentation

uv add openinference-instrumentation

poetry add openinference-instrumentation

Installing openinference-instrumentation

Before you install

Low friction installation with a pure-Python wheel. Actively maintained with a release 7 days ago and 1147 repository stars. Depends on opentelemetry-api, opentelemetry-sdk, and openinference-semantic-conventions, all standard observability packages.

License in practice

Apache-2.0 permissive license allows commercial and private use with minimal restrictions; suitable for most projects.

Quickstart

pip install openinference-instrumentation

from openinference.instrumentation import (
    Annotation,
    get_annotation_attributes,
)

span_annotations = get_annotation_attributes(
    annotations=[Annotation(name="hallucination", label="factual")]
)
span.set_attributes(span_annotations)

Requires Python 3.10 or later (supports up to 3.14).

Verify before relying

• Whether the package works with OpenTelemetry versions older than those in the current SDK release.
• Performance impact of context manager overhead in high-throughput tracing scenarios.

Package facts

Evidence: openinference_instrumentation-0.1.57-py3-none-any.whl

Tags

More Monitoring packages