openinference-instrumentation-crewai

What it is and what it does

This package wraps CrewAI agent execution to emit OpenTelemetry spans, capturing the structure and behavior of crews, agents, and tool calls. It integrates with the OpenInference semantic conventions to standardize how LLM agent traces are recorded, then exports them to any OpenTelemetry-compatible backend—most commonly Arize Phoenix for local inspection or Arize AX for cloud-hosted analysis.

The instrumentation works in two modes: the default wrapper-based approach, which intercepts Python method calls on CrewAI objects, and an event-listener mode for low-code or AMP environments where execution flows through CrewAI's event bus instead. Once enabled, it requires minimal setup—typically just instantiating the instrumentor with a tracer provider—and automatically captures agent roles, task descriptions, tool invocations, and LLM calls without modifying your crew code.

Use it for:

• Debug crew execution by inspecting traces in Phoenix to see which agents ran, what tools they called, and where latency occurred.
• Monitor production CrewAI applications by exporting traces to Arize AX for alerting, performance analysis, and cost tracking.
• Validate agent behavior during development by replaying traces and checking that agents followed expected task sequences.
• Correlate crew performance with LLM provider costs and latency by analyzing span metadata and LLM call counts.
• Troubleshoot tool failures by examining detailed traces showing which tools were invoked and what data they received.

Worth the install?

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

Automatically instruments CrewAI agent applications with OpenTelemetry tracing, sending execution traces to monitoring backends like Arize Phoenix or Arize AX for observability.

Yes. The package is actively maintained, has no known vulnerabilities, carries a permissive license, and solves a real need—CrewAI applications are difficult to debug and monitor without instrumentation. If you run CrewAI in production or need visibility into agent behavior, this is the standard path. Install it if you have an OpenTelemetry collector available; skip it only if your crews are purely local experiments with no observability requirement.

Install

openinference-instrumentation-crewai on PyPI

pip install openinference-instrumentation-crewai

uv add openinference-instrumentation-crewai

poetry add openinference-instrumentation-crewai

Installing openinference-instrumentation-crewai

Before you install

Low friction: pure Python wheel with seven runtime dependencies, all from the OpenTelemetry and OpenInference ecosystems. Actively maintained—last commit 2026-08-14, release 7 days old.

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-crewai

from openinference.instrumentation.crewai import CrewAIInstrumentor
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor
from opentelemetry.exporter.otlp.proto.http.trace_exporter import OTLPSpanExporter

trace_provider = TracerProvider()
trace_provider.add_span_processor(SimpleSpanProcessor(OTLPSpanExporter("http://127.0.0.1:6006/v1/traces")))
CrewAIInstrumentor().instrument(tracer_provider=trace_provider)

Requires an OpenTelemetry collector (e.g., Arize Phoenix) running and accessible at the configured endpoint; CrewAI and its dependencies must be installed separately.

Verify before relying

• Whether event-listener mode works with all CrewAI execution patterns or only specific low-code/AMP setups.
• Performance overhead of instrumentation on large crews or high-frequency agent executions.
• Compatibility with custom CrewAI extensions or third-party agent frameworks beyond standard CrewAI.

Package facts

Evidence: openinference_instrumentation_crewai-1.1.12-py3-none-any.whl

Tags

More Monitoring packages

Further reading