OpenAI Agents

2

Install the OpenAI Agents SDK

Required

Install the OpenAI Agents SDK. PostHog instruments your agent runs by registering a tracing processor. The PostHog SDK does not proxy your calls.

pip install openai-agents

Proxy note

These SDKs do not proxy your calls. They only fire off an async call to PostHog in the background to send the data. You can also use LLM analytics with other SDKs or our API, but you will need to capture the data in the right format. See the schema in the manual capture section (opens in a new tab) for more details.

3

Initialize PostHog tracing

Required

Initialize PostHog with your project token and host from your project settings (opens in a new tab), then call instrument() to register PostHog tracing with the OpenAI Agents SDK. This automatically captures all agent traces, spans, and LLM generations.

from posthog import Posthog
from posthog.ai.openai_agents import instrument
posthog = Posthog(
    "<ph_project_token>",
    host="https://us.i.posthog.com"
)
instrument(
    client=posthog,
    distinct_id="user_123", # optional
    privacy_mode=False, # optional
    groups={"company": "company_id_in_your_db"}, # optional
    properties={"conversation_id": "abc123"}, # optional
)

Note: If you want to capture LLM events anonymously, don’t pass a distinct ID to instrument(). See our docs on anonymous vs identified events (opens in a new tab) to learn more.

4

Run your agents

Required

Run your OpenAI agents as normal. PostHog automatically captures $ai_generation events for LLM calls and $ai_span events for agent execution, tool calls, and handoffs.

from agents import Agent, Runner
agent = Agent(
    name="Assistant",
    instructions="You are a helpful assistant.",
)
result = Runner.run_sync(agent, "Tell me a fun fact about hedgehogs")
print(result.final_output)

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

5

Multi-agent and tool usage

Optional

PostHog captures the full trace hierarchy for complex agent workflows including handoffs and tool calls.

from agents import Agent, Runner, function_tool
@function_tool
def get_weather(city: str) -> str:
    """Get the weather for a city."""
    return f"The weather in {city} is sunny, 72F"
weather_agent = Agent(
    name="WeatherAgent",
    instructions="You help with weather queries.",
    tools=[get_weather]
)
triage_agent = Agent(
    name="TriageAgent",
    instructions="Route weather questions to the weather agent.",
    handoffs=[weather_agent]
)
result = Runner.run_sync(triage_agent, "What's the weather in San Francisco?")

This captures:

• Agent spans for TriageAgent and WeatherAgent
• Handoff spans showing the routing between agents
• Tool spans for get_weather function calls
• Generation spans for all LLM calls

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 LLM analytics, you should see rows of data appear in the Traces and Generations tabs.

Check for LLM events in PostHog (opens in a new tab)

6

Next steps

Recommended

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