CrewAI

1. Emitting — CrewAI does not emit OTel spans through your application’s tracer provider. (Its built-in telemetry runs on a private, internal provider that reports product usage to CrewAI’s own servers — deliberately never installed globally, so nothing from it ever reaches your exporter.) Span emission comes from the OpenInference instrumentors.
2. Receiving — decimalai.init(crewai=True) wires DecimalAI’s OTel exporter as the global tracer provider and activates those instrumentors against it, which turns their spans into DecimalAI traces.

​

Install

pip install decimalai crewai openinference-instrumentation-crewai openinference-instrumentation-openai

​

Use

import decimalai
decimalai.init(api_key="dai_sk_...", crewai=True)  # exporter + the installed instrumentors

from crewai import Agent, Crew, Task

researcher = Agent(
    role="Support researcher",
    goal="Answer questions about order status",
    backstory="You have read every order-status runbook.",
    llm="openai/gpt-5-mini",
)
crew = Crew(
    agents=[researcher],
    tasks=[Task(
        description="Where is order 4512?",
        expected_output="One sentence.",
        agent=researcher,
    )],
)

# Crew tasks and agent interactions are now traced
crew.kickoff()

​

What gets captured

• Each task in the crew as a span (CrewAIInstrumentor)
• Agent-to-agent conversations and tool calls with inputs and outputs (CrewAIInstrumentor)
• LLM generations underneath each agent’s reasoning step, with the model name, token counts and the messages themselves — from the provider instrumentor, not from CrewAIInstrumentor, which emits no LLM span

​

Caveats

• Both halves are required, and it is the packages that decide whether you have both. init(crewai=True) turns the emitters on for you, but only the ones that are installed, and a missing one is a log line rather than an error — so an install line short a package produces a quiet, partial trace rather than a failure. Without openinference-instrumentation-crewai it captures nothing (an exporter with no spans to export) and says so at WARNING. Without a provider instrumentor it captures the crew shape — the kickoff span and each task’s span — with no LLM call, no model and no token counts, and says so at INFO.
• LiteLLM is an optional CrewAI extra now, and init() does not reach it. Older CrewAI routed every model string through LiteLLM, where LiteLLMInstrumentor emitted the LLM spans. On the version we measured (1.15.16) litellm ships only with pip install "crewai[litellm]", and llm="openai/…" uses CrewAI’s own OpenAI provider, which calls the openai SDK directly — which is why openinference-instrumentation-openai is the package in the install line. If your setup does route through LiteLLM, init(crewai=True) will not pick it up: it auto-activates the openai, anthropic and google-genai instrumentors and no others. Install openinference-instrumentation-litellm and call LiteLLMInstrumentor().instrument() yourself — it binds to the tracer provider init() already set, and the LLM spans land.
• Manifest detection captures tool names only — schemas aren’t always exposed through CrewAI’s API. Use register_manifest() for schema-aware tracking.
• Multi-crew setups: each crew is its own DecimalAI “agent” by default. Pass agent_name= to init() if you want them unified.

​

What’s next