OpenAI Agents Integration

Updated on Aug 14, 2026

Use this integration when your OpenAI Agents application is implemented in Python.

Setup

1. Install SDK

bash

pip install progress-observability

2. Instrument your app

python

Observability.instrument(
    app_name=os.getenv("OBSERVABILITY_APP_NAME"),
    api_key=os.getenv("OBSERVABILITY_API_KEY")
)

3. Complete example

python

import os

from dotenv import load_dotenv
from agents import Agent, Runner, function_tool

from progress.observability import Observability

load_dotenv()

# LLM INSTRUMENTATION
Observability.instrument(
    app_name=os.getenv("OBSERVABILITY_APP_NAME"),
    api_key=os.getenv("OBSERVABILITY_API_KEY")
)

model = os.getenv("OPENAI_MODEL", "gpt-4o-mini")


@function_tool
def get_weather(location: str) -> str:
    """Get the weather for a given location."""
    return f"The weather in {location} is cloudy with a high of 15°C."


agent = Agent(
    name="Weather Assistant",
    model=model,
    tools=[get_weather],
    instructions="You are a helpful AI assistant."
)

result = Runner.run_sync(agent, "What's the weather in Paris?")

Observability.shutdown()

Avoiding duplicate spans

When using Observability.instrument() with the OpenAI Agents SDK, each model call may be captured twice — once by the SDK's native tracing hooks (OpenAIAgentsInstrumentor) and once by the lower-level OpenAI client patch (OpenAIInstrumentor). This doubles reported token usage and cost.

To fix it, block the redundant instrumentor:

python

from progress.observability.enums import ObservabilityInstruments

Observability.instrument(
    app_name=os.getenv("OBSERVABILITY_APP_NAME"),
    api_key=os.getenv("OBSERVABILITY_API_KEY"),
    block_instruments={ObservabilityInstruments.OPENAI},
)