Skip to main content

​
Overview

This guide shows you how to integrate Latitude Telemetry into an application built with CrewAI (crewai). Latitude includes dedicated instrumentation for CrewAI, so crew kickoffs, agent steps, model generations, and tool calls appear as traces.
You’ll keep building crews exactly as you do today. Telemetry observes agent steps, model calls, and tool calls as they happen.
CrewAI instrumentation is available in the Python SDK only.

​
Requirements

  • A Latitude account and API key
  • A Latitude project slug
  • A project that uses CrewAI (crewai)
  • A model provider key — CrewAI uses OpenAI by default (OPENAI_API_KEY)

​
Steps

1

Install

pip install latitude-telemetry crewai
2

Initialize and use

import crewai
from crewai import Agent, Crew, Task

from latitude_telemetry import Latitude, capture

latitude = Latitude(
    api_key="your-api-key",
    project="your-project-slug",
    instrumentations={"crewai": crewai},
)

def crew_run():
    researcher = Agent(
        role="Researcher",
        goal="Summarize topics concisely",
        backstory="You provide brief, accurate summaries.",
        llm="gpt-4o-mini",
    )
    task = Task(
        description="Explain what OpenTelemetry is in one sentence.",
        expected_output="A single sentence.",
        agent=researcher,
    )
    return Crew(agents=[researcher], tasks=[task]).kickoff().raw

capture("crew-run", crew_run)

latitude.shutdown()

​
What you get

Each crew kickoff shows up as a trace with nested spans:
  • Crew / agent spans — agent roles, goals, and configured tools
  • Generation spans — model, input/output messages, and token usage
  • Tool spans — tool calls with input arguments and output
Wrap a request or job with capture() to attach a user_id, session_id, tags, or metadata to every span produced inside.

​
Seeing Your Traces

Once connected, traces appear automatically in Latitude:
  1. Open your project in the Latitude dashboard
  2. Each crew run shows the full hierarchy of agent → generation → tool calls
  3. Token usage and latency are aggregated at every level