Skip to main content

​
Overview

Quraite provides a built-in adapter for OpenAI Agents. This adapter allows you to integrate OpenAI Agents with Quraite and run evaluations on them. This adapter uses OpenInference instrumentation to capture the trajectory of the agent.

​
Prerequisites

  • Agent dependencies: openai-agents package, OpenAI API key, and any tools you want to use.
  • Quraite dependencies: quraite[openai-agents-oi] package.

​
Define Your OpenAI Agent

You create a normal OpenAI Agent first, exactly as you would in any OpenAI Agents project. 
from agents import Agent
from dotenv import load_dotenv
from prompts import SYSTEM_PROMPT
from tools import (
    book_flight,
    cancel_ticket,
    fetch_user_flight_information,
    lookup_policy,
    search_flights,
    update_ticket_to_new_flight,
)


load_dotenv(override=True)


flight_booking_agent = Agent(
    name="Flight Booking Agent",
    instructions=SYSTEM_PROMPT,
    tools=[
        book_flight,
        cancel_ticket,
        fetch_user_flight_information,
        lookup_policy,
        search_flights,
        update_ticket_to_new_flight,
    ],
    model="gpt-4o-mini",
)

​
Set up Tracing

Quraite provides a tracing setup helper. For OpenAI Agents, pass the Framework.OPENAI_AGENTS enum: 
from quraite.tracing import Framework, setup_tracing


tracer_provider = setup_tracing([Framework.OPENAI_AGENTS])

​
Wrap the OpenAI Agent

Inject the tracer_provider into the OpenaiAgentsAdapter so Quraite can capture spans and metrics from the agent. 
from quraite.adapters import OpenaiAgentsAdapter


openai_agents_adapter = OpenaiAgentsAdapter(
    agent=flight_booking_agent,
    agent_name="Flight Booking Agent",
    tracer_provider=tracer_provider,
)

​
Run the Agent

Once wrapped, you expose the adapter using Quraite’s run_agent helper. 
if __name__ == "__main__":
    from dotenv import load_dotenv
    from quraite import run_agent

    load_dotenv(override=True)

    run_agent(
        openai_agents_adapter,
        port=8080,
        host="0.0.0.0",
        tunnel="cloudflare",
    )

​
End-to-End Examples