Skip to main content

CrewAI Integration

Foxhound integrates with CrewAI via FoxCrewTracer, which hooks into step and task callbacks to capture agent activity.

Installation​

pip install foxhound-ai[crewai]

Usage​

Synchronous​

from crewai import Agent, Crew, Task
from foxhound import FoxhoundClient
from foxhound.integrations.crewai import FoxCrewTracer

fox = FoxhoundClient(api_key="fox_...", endpoint="https://your-foxhound-instance.com")
tracer = FoxCrewTracer.from_client(fox, agent_id="my-crew")

researcher = Agent(role="Researcher", goal="Research topics", backstory="...")
task = Task(description="Research AI safety", agent=researcher)

crew = Crew(
    agents=[researcher],
    tasks=[task],
    step_callback=tracer.on_step,
    task_callback=tracer.on_task,
)

result = tracer.kickoff(crew, inputs={"topic": "AI safety"})
tracer.flush_sync()

Async​

result = await tracer.kickoff_async(crew, inputs={"topic": "AI safety"})
await tracer.flush()

What gets traced​

  • Per-agent steps with tool calls and outputs
  • Task lifecycle (start, complete, fail)
  • Inter-agent handoffs
  • LLM calls from each agent
  • Errors and retries

Why use the native integration?​

FoxCrewTracer has direct access to raw CrewAI event payloads, including per-turn tool call arguments that are not available through the generic OTel bridge.