CrewAI

CrewAI is a Python framework for orchestrating role-playing autonomous agents — agents collaborate on tasks, hand off work, and call tools. Arize AX captures every CrewAI run — Crew / Agent / Task spans plus the underlying LLM calls — via the openinference-instrumentation-crewai package, paired with an LLM-level instrumentor matching the provider CrewAI routes to.

CrewAI Tracing Tutorial (Google Colab)

Prerequisites

Python 3.10+
An Arize AX account (sign up)
An OPENAI_API_KEY from the OpenAI Platform
CrewAI 1.0+ (this guide uses the native-SDK routing model introduced in 1.0; for 0.x see Troubleshooting)

Launch Arize AX

Sign in to your Arize AX account.
From Space Settings, copy your Space ID and API Key. You will set them as ARIZE_SPACE_ID and ARIZE_API_KEY below.

Install

pip install arize-otel \
  openinference-instrumentation-crewai \
  openinference-instrumentation-openai \
  crewai

CrewAI 1.0+ routes LLM calls based on the model-string prefix: openai/... and bare model names go to the native OpenAI SDK, anthropic/... goes to Anthropic, etc. The LLM-level instrumentor must match the provider CrewAI is actually calling. The example below uses openai/gpt-5.5 so it pairs openinference-instrumentation-crewai with openinference-instrumentation-openai. Other prefixes need a different pairing — see Which instrumentor do I need? below.

Which instrumentor do I need?

CrewAI 1.0+ routes LLM calls to different provider SDKs based on your model string. Match your LLM-level instrumentor to the provider CrewAI is actually calling:

Model string	Provider SDK used	Instrumentor package
`openai/gpt-5.5`, `gpt-5.5`, or no prefix	OpenAI (native)	`openinference-instrumentation-openai`
`anthropic/claude-sonnet-4-6`	Anthropic (native)	`openinference-instrumentation-anthropic`
Groq, Together, or other providers without a native CrewAI SDK	LiteLLM (fallback)	`openinference-instrumentation-litellm`

The CrewAI instrumentor (openinference-instrumentation-crewai) is always required regardless of provider — it captures Crew, Agent, and Task spans.

Configure credentials

export ARIZE_SPACE_ID="<your-space-id>"
export ARIZE_API_KEY="<your-api-key>"
export ARIZE_PROJECT_NAME="crewai-tracing-example"
export OPENAI_API_KEY="<your-openai-api-key>"

Setup tracing

# instrumentation.py
import os

from arize.otel import register
from openinference.instrumentation.crewai import CrewAIInstrumentor
from openinference.instrumentation.openai import OpenAIInstrumentor

tracer_provider = register(
    space_id=os.environ["ARIZE_SPACE_ID"],
    api_key=os.environ["ARIZE_API_KEY"],
    project_name=os.environ["ARIZE_PROJECT_NAME"],
)

# Both instrumentors monkey-patch their target libraries — they must run
# before any `from crewai import ...` or any LLM client construction.
CrewAIInstrumentor().instrument(tracer_provider=tracer_provider)
OpenAIInstrumentor().instrument(tracer_provider=tracer_provider)
print("Arize AX tracing initialized for CrewAI.")

CrewAI ships its own built-in telemetry, which emits extra OpenTelemetry spans (Crew Created, Task Created, Task Execution, Environment Context) to the same tracer provider you register above. Those spans carry no openinference.span.kind, so they land in Arize AX as kind-less spans mixed in with the Crew, Agent, and LLM spans you instrumented. Opt out of CrewAI’s telemetry to keep them out of your project — it leaves the OpenInference spans exported to Arize AX untouched:

export CREWAI_DISABLE_TELEMETRY=true

Set it before running your app. On older CrewAI releases the variable is CREWAI_DISABLE_TRACKING=true. Don’t reach for OTEL_SDK_DISABLED=true here — CrewAI honors it too, but it turns off the entire OpenTelemetry SDK and stops the Arize AX export along with the CrewAI spans.

Run CrewAI

# example.py

# Importing instrumentation first ensures both instrumentors run before
# CrewAI is imported. Out-of-order imports silently drop spans.
from instrumentation import tracer_provider

from crewai import Agent, Crew, LLM, Task

# `openai/gpt-5.5` tells CrewAI 1.0+ to use the native OpenAI SDK, which
# OpenAIInstrumentor patches. It reads OPENAI_API_KEY from the env.
llm = LLM(model="openai/gpt-5.5")

researcher = Agent(
    role="Ocean researcher",
    goal="Explain ocean facts to a general audience.",
    backstory="You are a marine scientist who values brevity.",
    llm=llm,
    allow_delegation=False,
    verbose=False,
)

writer = Agent(
    role="Science writer",
    goal="Turn researcher notes into a clean two-sentence summary.",
    backstory="You write clear, concise prose for a general audience.",
    llm=llm,
    allow_delegation=False,
    verbose=False,
)

research_task = Task(
    description=(
        "Explain why the ocean is salty. List the two primary mechanisms."
    ),
    expected_output="A short factual paragraph naming the two mechanisms.",
    agent=researcher,
)

writeup_task = Task(
    description=(
        "Using the researcher's notes, produce exactly two sentences "
        "for a general reader. No preamble."
    ),
    expected_output="Exactly two sentences, no headings.",
    agent=writer,
)

crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, writeup_task],
    verbose=False,
)

result = crew.kickoff()
print(result)

Expected output

Arize AX tracing initialized for CrewAI.
The ocean is salty because rivers continuously dissolve mineral salts from rocks and soil and carry them to the sea, where they accumulate over millions of years. Water leaves the ocean through evaporation but the salts remain, steadily concentrating until reaching today's roughly 3.5% salinity.

Verify in Arize AX

Open your Arize AX space and select project crewai-tracing-example.
You should see a new trace within ~30 seconds with this shape: a Crew_<uuid>.kickoff root span (CHAIN) wraps a per-agent <role>._execute_core span (AGENT) for each task, and each _execute_core wraps a ChatCompletion LLM span (model gpt-5.5, with prompt, response, and token usage attached). If you left CrewAI’s built-in telemetry on, you’ll also see kind-less Crew Created / Task Created / Task Execution spans — see the note under Setup tracing to suppress them.
If no traces appear, see Troubleshooting.

Check from the skill, CLI, or SDK

Confirm spans are actually reaching your Arize AX project. Use whichever fits your workflow — the skill and CLI work for any framework; the SDK check is shown for each language.

Arize skill (agent)
AX CLI
SDK

Install the Arize Skills plugin and let your coding agent check for you:

npx skills add Arize-ai/arize-skills

Then prompt your agent:

Use the arize-trace skill to export and analyze recent traces from my project. Confirm spans are arriving, and summarize any errors or latency issues.

Export recent spans for your project — any rows mean traces are landing:

ax spans export "$ARIZE_PROJECT_NAME" --space "$ARIZE_SPACE_ID" \
  --limit 5 --stdout | jq 'length'

A non-zero count confirms spans reached Arize AX. Run ax auth login first if you have not authenticated. See the ax spans reference.

Query the project’s spans and check that at least one came back.

import os
from arize import ArizeClient

client = ArizeClient(api_key=os.environ["ARIZE_API_KEY"])
resp = client.spans.list(
    project=os.environ["ARIZE_PROJECT_NAME"],
    space=os.environ["ARIZE_SPACE_ID"],
    limit=5,
)
count = len(resp.spans)
print(
    f"{count} span(s) found" if count else "No spans yet — recheck setup"
)

// Reads ARIZE_API_KEY from the environment.
import { listSpans } from "@arizeai/ax-client";

const { data: spans } = await listSpans({
  project: process.env.ARIZE_PROJECT_NAME!,
  space: process.env.ARIZE_SPACE_ID!,
  limit: 5,
});
const count = spans.length;
console.log(
  count ? `${count} span(s) found` : "No spans yet — recheck setup",
);

client, err := arize.NewClient(
    arize.Config{APIKey: os.Getenv("ARIZE_API_KEY")},
)
if err != nil {
    log.Fatal(err)
}
resp, err := client.Spans.List(ctx, spans.ListRequest{
    Project: os.Getenv("ARIZE_PROJECT_NAME"),
    Space:   os.Getenv("ARIZE_SPACE_ID"),
    Limit:   5,
})
if err != nil {
    log.Fatal(err)
}
fmt.Printf("%d span(s) found\n", len(resp.Spans))

SDK span references: Python · TypeScript · Go.

Troubleshooting

No traces in Arize AX. Confirm ARIZE_SPACE_ID and ARIZE_API_KEY are set in the same shell that runs example.py. Enable OpenTelemetry debug logs with export OTEL_LOG_LEVEL=debug and re-run.
Crew / Agent / Task spans appear but no LLM spans. The LLM-level instrumentor doesn’t match the provider CrewAI is calling. CrewAI 1.0+ routes by model-string prefix:
- openai/<model> (or no prefix) → install openinference-instrumentation-openai
- anthropic/<model> → install openinference-instrumentation-anthropic
- Groq / Together / providers without a native CrewAI SDK → install openinference-instrumentation-litellm (CrewAI falls back to LiteLLM for these)
All spans missing — no Crew, no LLM. This is almost always import order. CrewAIInstrumentor().instrument(...) and the LLM-level instrumentor both monkey-patch class methods or module functions; if crewai (or the LLM library) is imported before instrument() runs, the patches land on stale references and never fire. Make sure instrumentation.py is the first import in your entry point — and watch for transitive imports (importing your own crew.py module that has from crewai import ... at the top has the same effect).
401 from OpenAI. Verify OPENAI_API_KEY is set and has access to gpt-5.5. Swap openai/gpt-5.5 in LLM(model=...) for a model your key can call.
Version-check error from CrewAIInstrumentor. If you’re on a CrewAI release candidate that’s outside the instrumentor’s supported range, escape-hatch with CrewAIInstrumentor().instrument(skip_dep_check=True, tracer_provider=tracer_provider).
You’re on CrewAI 0.x. 0.x routes every LLM call through LangChain or LiteLLM rather than provider-native SDKs. Pair the CrewAI instrumentor with both openinference-instrumentation-langchain and openinference-instrumentation-litellm instead of openinference-instrumentation-openai. The Setup tracing call list becomes CrewAIInstrumentor().instrument(...) + LangChainInstrumentor().instrument(...) + LiteLLMInstrumentor().instrument(...). Upgrade to 1.0+ when you can — the LiteLLM removal guide covers the migration.

OpenTelemetry

LLM Providers

Agent Frameworks

Coding Agents

Platforms

Orchestration

Evaluation

CrewAI Tracing Tutorial (Google Colab)

Prerequisites

Launch Arize AX

Install

Which instrumentor do I need?

Configure credentials

Setup tracing

Run CrewAI

Expected output

Verify in Arize AX

Check from the skill, CLI, or SDK

Troubleshooting

Resources

CrewAI Documentation

CrewAI LLM Provider Routing

OpenInference CrewAI Instrumentor

CrewAI GitHub

CrewAI Tracing Tutorial (Google Colab)

​Prerequisites

​Launch Arize AX

​Install

​Which instrumentor do I need?

​Configure credentials

​Setup tracing

​Run CrewAI

​Expected output

​Verify in Arize AX

​Check from the skill, CLI, or SDK

​Troubleshooting

​Resources

CrewAI Documentation

CrewAI LLM Provider Routing

OpenInference CrewAI Instrumentor

CrewAI GitHub

Prerequisites

Launch Arize AX

Install

Which instrumentor do I need?

Configure credentials

Setup tracing

Run CrewAI

Expected output

Verify in Arize AX

Check from the skill, CLI, or SDK

Troubleshooting

Resources