OpenAI Agents SDK Integration

Route OpenAI Agents SDK calls through the Curate-Me gateway for cost tracking, rate limiting, PII scanning, and governance. The Agents SDK uses an OpenAI client internally — just set the base_url and register it as the default.

Prerequisites


# Python
pip install openai-agents
 
# TypeScript
npm install @openai/agents openai

Python Setup

Create a gateway-routed client and register it as the default for all agents:


from openai import AsyncOpenAI
from agents import Agent, Runner, set_default_openai_client
 
client = AsyncOpenAI(
    api_key="sk-your-openai-key",
    base_url="https://api.curate-me.ai/v1/openai",
    default_headers={
        "X-CM-API-Key": "cm_sk_xxx",
        "X-CM-Tags": "project=my-agents,env=production",
    },
)
 
# All agents now route through the gateway
set_default_openai_client(client)

Defining Agents

Define agents as normal. The gateway is transparent — agents don’t need to know about it:


triage_agent = Agent(
    name="Triage",
    instructions="You route questions to the right specialist. For technical "
    "questions, hand off to the technical agent. For general questions, "
    "answer directly.",
    model="gpt-4o",
)
 
technical_agent = Agent(
    name="Technical Advisor",
    instructions="You are a technical advisor specializing in AI infrastructure. "
    "Give concise, actionable answers.",
    model="gpt-4o",
)
 
# Wire up handoffs
triage_agent.handoffs = [technical_agent]

Running Agents


import asyncio
 
async def main():
    # Single agent
    result = await Runner.run(
        triage_agent,
        input="What is an AI gateway and why do I need one?",
    )
    print(f"Agent: {result.last_agent.name}")
    print(f"Response: {result.final_output}")
 
    # With handoff -- triage routes to technical agent
    result = await Runner.run(
        triage_agent,
        input="How do I set up rate limiting for my LLM API calls?",
    )
    print(f"Agent: {result.last_agent.name}")
    print(f"Response: {result.final_output}")
 
asyncio.run(main())

Every LLM call made by every agent is routed through the gateway. The dashboard shows per-agent cost breakdowns when you use X-CM-Tags.

TypeScript Setup


import OpenAI from "openai";
import { Agent, run, setDefaultOpenAIClient } from "@openai/agents";
 
const client = new OpenAI({
  apiKey: "sk-your-openai-key",
  baseURL: "https://api.curate-me.ai/v1/openai",
  defaultHeaders: {
    "X-CM-API-Key": "cm_sk_xxx",
    "X-CM-Tags": "project=my-agents,env=production",
  },
});
 
setDefaultOpenAIClient(client);
 
const technicalAgent = new Agent({
  name: "Technical Advisor",
  instructions: "You are a technical advisor. Give concise answers.",
  model: "gpt-4o",
});
 
const triageAgent = new Agent({
  name: "Triage",
  instructions: "Route technical questions to the technical agent.",
  model: "gpt-4o",
  handoffs: [technicalAgent],
});
 
const result = await run(triageAgent, {
  input: "How do I set up rate limiting for my LLM API calls?",
});
console.log(`Agent: ${result.lastAgent.name}`);
console.log(`Response: ${result.finalOutput}`);

Governance with Multi-Agent Handoffs

When agents hand off to each other, each agent’s LLM calls are individually governed:

Rate limiting applies per-org across all agents (shared quota)
Cost tracking records each call separately with the agent name in metadata
PII scanning checks every message at every handoff boundary
Budget enforcement applies the same daily/monthly budget across the entire org

Use X-CM-Tags to track costs per agent role:


# Per-agent cost tags
client = AsyncOpenAI(
    api_key="sk-your-key",
    base_url="https://api.curate-me.ai/v1/openai",
    default_headers={
        "X-CM-API-Key": "cm_sk_xxx",
        "X-CM-Tags": "agent=triage,project=support-bot",
    },
)

Environment Variables


CM_GATEWAY_URL=https://api.curate-me.ai
CM_API_KEY=cm_sk_xxx
PROVIDER_KEY=sk-your-openai-key


import os
 
client = AsyncOpenAI(
    api_key=os.environ["PROVIDER_KEY"],
    base_url=f"{os.environ['CM_GATEWAY_URL']}/v1/openai",
    default_headers={
        "X-CM-API-Key": os.environ["CM_API_KEY"],
    },
)
set_default_openai_client(client)

Next Steps

Gateway Quickstart — full setup walkthrough
Cost Attribution — track costs by agent, project, team
Orchestration Patterns — DAG-based multi-agent workflows
Framework Integrations — other SDK integrations