CrewAI observability installation

Let AI instrument your LLM calls for you

Skip the manual setup — run this in your project and the wizard installs the SDK and wires up AI Observability for you.

Learn more
PostHog Wizard hedgehog

Contents

  1. Install the PostHog SDK

    Required

    Setting up analytics starts with installing the PostHog SDK. CrewAI uses LiteLLM under the hood, and PostHog integrates with LiteLLM's callback system.

    pip install posthog

  2. Install CrewAI

    Required

    Install CrewAI. PostHog instruments your LLM calls through LiteLLM's callback system that CrewAI uses natively.

    pip install crewai litellm

  3. Configure PostHog with LiteLLM

    Required

    Set your PostHog project token and host as environment variables, then configure LiteLLM to use PostHog as a callback handler. You can find your project token in your project settings.

    import os
    import litellm
    from crewai import Agent, Task, Crew, LLM
    

    # Set PostHog environment variables
    os.environ["POSTHOG_API_KEY"] = "<ph_project_token>"
    os.environ["POSTHOG_API_URL"] = "https://us.i.posthog.com"
    

    # Enable PostHog callbacks in LiteLLM
    litellm.success_callback = ["posthog"]
    litellm.failure_callback = ["posthog"]
    How this works

    CrewAI can route LLM calls either through its own provider clients or through LiteLLM. PostHog hooks into LiteLLM's callback system, so you need is_litellm=True on the LLM you pass to your agents. With it, PostHog captures every call as an $ai_generation event, without proxying your calls.

  4. Run your crew

    Required

    Run your CrewAI agents as normal. PostHog automatically captures an $ai_generation event for each LLM call. LiteLLM's callback does not see the tools your agents call. Capture a tool's own execution as a span from inside the tool itself instead, as my_tool does below.

    from posthog import Posthog
    from crewai.tools import tool
    import time, uuid
    

    posthog = Posthog("<ph_project_token>", host="https://us.i.posthog.com")
    

    trace_id = str(uuid.uuid4())
    

    @tool
    def my_tool(query: str) -> str:
        """Describe what your tool does."""
        start = time.time()
        result = run_tool(query)
    

        posthog.capture(
            distinct_id="user_123",
            event="$ai_span",
            properties={
                "$ai_trace_id": trace_id,
                "$ai_session_id": "conversation-abc",
                "$ai_span_id": str(uuid.uuid4()),
                "$ai_span_name": "my_tool",
                "$ai_input_state": {"query": query},
                "$ai_output_state": result,
                "$ai_latency": time.time() - start,
            },
        )
        return result
    

    # is_litellm=True routes calls through LiteLLM so the PostHog
    # callback fires. Without it, CrewAI uses its own provider client
    # and no events are captured.
    llm = LLM(
        model="gpt-4o-mini",
        is_litellm=True,
        metadata={
            "user_id": "user_123",
            "$ai_session_id": "conversation-abc",
            "$ai_trace_id": trace_id,
        },
    )
    

    researcher = Agent(
        role="Researcher",
        goal="Find the weather in a city",
        backstory="You are an expert wildlife researcher.",
        llm=llm,
        tools=[my_tool],
    )
    

    task = Task(
        description="Find the weather in Paris.",
        expected_output="The weather in Paris.",
        agent=researcher,
    )
    

    crew = Crew(agents=[researcher], tasks=[task])
    result = crew.kickoff()
    print(result)

    You can expect captured $ai_generation events to have the following properties:

    PropertyDescription
    $ai_modelThe specific model, like gpt-5-mini or claude-4-sonnet
    $ai_latencyThe latency of the LLM call in seconds
    $ai_time_to_first_tokenTime to first token in seconds (streaming only)
    $ai_toolsTools and functions available to the LLM
    $ai_inputList of messages sent to the LLM
    $ai_input_tokensThe number of tokens in the input (often found in response.usage)
    $ai_output_choicesList of response choices from the LLM
    $ai_output_tokensThe number of tokens in the output (often found in response.usage)
    $ai_total_cost_usdThe total cost in USD (input + output)
    [...]See full list of properties

  5. Verify traces and generations

    Recommended
    Confirm LLM events are being sent to PostHog

    Let's make sure LLM events are being captured and sent to PostHog. Under AI Observability, you should see rows of data appear in the Traces and Generations tabs.


    LLM generations in PostHog
    Check for LLM events in PostHog

  6. Next steps

    Recommended

    Now that you're capturing AI conversations, continue with the resources below to learn what else AI Observability enables within the PostHog platform.

    ResourceDescription
    BasicsLearn the basics of how LLM calls become events in PostHog.
    GenerationsRead about the $ai_generation event and its properties.
    TracesExplore the trace hierarchy and how to use it to debug LLM calls.
    SpansReview spans and their role in representing individual operations.
    Anaylze LLM performanceLearn how to create dashboards to analyze LLM performance.

Community questions

Was this page useful?