OpenInference OpenAI Agents Instrumentation

Warnings

• breaking This instrumentation is designed for `openai` library versions 1.0.0 and above. Using `openai` versions 0.x.x will lead to `AttributeError` or incorrect instrumentation as the API structure changed significantly.
  Fix: Upgrade your `openai` library to version `1.0.0` or higher (`pip install 'openai>=1.0.0,<2'`).
• gotcha The OpenTelemetry `TracerProvider` must be configured and set *before* calling `OpenAIInstrumentor().instrument()`. If the order is reversed, the instrumentation might not pick up the `TracerProvider`, resulting in no traces being emitted.
  Fix: Always ensure your OpenTelemetry SDK setup (e.g., `trace.set_tracer_provider(...)`) is executed prior to calling `OpenAIInstrumentor().instrument()`.
• gotcha By default, the instrumentation captures full inputs and outputs, which may include sensitive or personally identifiable information (PII). Be mindful of data privacy requirements.
  Fix: To prevent sensitive data from being exported, configure `OpenAIInstrumentor` with `hide_inputs=True` and/or `hide_outputs=True` when initializing, e.g., `OpenAIInstrumentor(hide_inputs=True, hide_outputs=True).instrument()`.

Install

• pip install openinference-instrumentation-openai-agents openai opentelemetry-sdk Install library and core dependencies

Imports

• OpenAIInstrumentor

  from openinference.instrumentation.openai_agents import OpenAIInstrumentor

Quickstart

This quickstart demonstrates how to instrument an OpenAI Assistants API workflow. It sets up an OpenTelemetry TracerProvider, instruments the OpenAI library, creates an assistant, thread, message, and runs the assistant, capturing traces for each step. Remember to set your `OPENAI_API_KEY` environment variable.

import os
import time
from opentelemetry import trace
from opentelemetry.sdk.resources import Resource
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor, ConsoleSpanExporter
from openinference.instrumentation.openai_agents import OpenAIInstrumentor
import openai

# 1. Setup OpenTelemetry (before instrumentation)
resource = Resource.create({"service.name": "openai-agents-quickstart"})
tracer_provider = TracerProvider(resource=resource)
tracer_provider.add_span_processor(SimpleSpanProcessor(ConsoleSpanExporter()))
trace.set_tracer_provider(tracer_provider)

# 2. Instrument OpenAI Agents
OpenAIInstrumentor().instrument()

# 3. Initialize OpenAI Client
# Ensure OPENAI_API_KEY is set in your environment
client = openai.OpenAI(api_key=os.environ.get("OPENAI_API_KEY", ""))

try:
    # 4. Interact with OpenAI Assistants API
    assistant = client.beta.assistants.create(
        name="Math Tutor",
        instructions="You are a personal math tutor. Answer questions briefly.",
        model="gpt-4o",
    )
    print(f"Created Assistant: {assistant.id}")

    thread = client.beta.threads.create()
    print(f"Created Thread: {thread.id}")

    message = client.beta.threads.messages.create(
        thread_id=thread.id,
        role="user",
        content="What is 1 + 1?",
    )
    print(f"Added Message: {message.id}")

    run = client.beta.threads.runs.create(
        thread_id=thread.id,
        assistant_id=assistant.id,
        instructions="Please address the user as Professor."
    )
    print(f"Started Run: {run.id}")

    # 5. Wait for the run to complete (simulated polling)
    while run.status in ['queued', 'in_progress', 'cancelling']:
        time.sleep(1)
        run = client.beta.threads.runs.retrieve(
            thread_id=thread.id,
            run_id=run.id
        )
        print(f"Run status: {run.status}")

    # 6. Retrieve messages after the run
    messages = client.beta.threads.messages.list(
        thread_id=thread.id
    )
    print("\nConversation:")
    for m in reversed(messages.data):
        print(f"{m.role}: {m.content[0].text.value}")

except openai.APIAssistantError as e:
    print(f"OpenAI API Error: {e}")
    print("Please ensure you have an OpenAI API key set and have access to the Assistants API.")
except Exception as e:
    print(f"An unexpected error occurred: {e}")
finally:
    # Clean up resources
    try:
        if 'thread' in locals() and thread.id:
            client.beta.threads.delete(thread.id)
            print(f"Deleted Thread: {thread.id}")
        if 'assistant' in locals() and assistant.id:
            client.beta.assistants.delete(assistant.id)
            print(f"Deleted Assistant: {assistant.id}")
    except Exception as e:
        print(f"Error during cleanup: {e}")