How to use Gemini with the OpenAI SDK (JavaScript)

Google’s Gemini models can be used through OpenAI compatible endpoints. That means you can keep using your existing OpenAI SDK or the new OpenAI Agents SDK, and simply point it at Gemini by swapping the base URL and API key.

This guide will show you step by step how to connect Gemini to the OpenAI SDK in JavaScript, and then how to run an agent using the OpenAI Agents SDK.

1. Prerequisites

Before you begin, make sure you have:

1. Node.js v18 or higher

2. A Gemini API key from Google AI Studio (copy this key and keep it safe)

3. Dependencies installed

    npm install openai @openai/agents dotenv zod
   

2. Project setup

Create a .env file in your project root:

BASE_URL=https://generativelanguage.googleapis.com/v1beta/openai/
API_KEY=your-gemini-api-key
MODEL_NAME=gemini-2.5-flash

Here

• BASE_URL points to the Gemini OpenAI compatible endpoint

• API_KEY is your Gemini API key

• MODEL_NAME is the Gemini model you want to use

3. Wiring Gemini into OpenAI SDK

We now set up a custom OpenAI client that uses Gemini’s endpoint.

import OpenAI from "openai";
import 'dotenv/config';

config();

if (!process.env.BASE_URL || !process.env.API_KEY || !process.env.MODEL_NAME) {
  throw new Error("Please set BASE_URL, API_KEY, and MODEL_NAME in .env");
}

const openai = new OpenAI({
  apiKey: process.env.API_KEY,
  baseURL: process.env.BASE_URL,
});

// Simple test call without Agents SDK
async function simpleChat() {
  const res = await openai.chat.completions.create({
    model: process.env.MODEL_NAME,
    messages: [{ role: "user", content: "Write a short poem about JavaScript" }],
  });

  console.log(res.choices[0].message.content);
}

simpleChat();

If you run this file, you should get a response from Gemini.

This proves your Gemini key and endpoint are working. Only after confirming this step should you integrate with the Agents SDK.

4. Using Gemini with the OpenAI Agents SDK

The OpenAI Agents SDK lets you build agents with tools, function calling, and structured workflows. To make it work with Gemini, we set up a custom provider.

Here is a working example:

import {
  Agent,
  Runner,
  setTracingDisabled,
  tool,
  OpenAIProvider,
  setDefaultOpenAIClient,
  setOpenAIAPI,
} from "@openai/agents";
import OpenAI from "openai";
import 'dotenv/config';
import { z } from "zod";

// Validate env
if (!process.env.BASE_URL || !process.env.API_KEY || !process.env.MODEL_NAME) {
  throw new Error("Missing BASE_URL, API_KEY, or MODEL_NAME in .env");
}

// 1. Create a custom OpenAI client pointing to Gemini
const openaiClient = new OpenAI({
  apiKey: process.env.API_KEY,
  baseURL: process.env.BASE_URL,
});

// 2. Set up provider and defaults
const modelProvider = new OpenAIProvider({ openAIClient: openaiClient });
setDefaultOpenAIClient(openaiClient);
setOpenAIAPI("chat_completions"); // Gemini supports chat completions API
setTracingDisabled(true);

// 3. Example tool
const getWeather = tool({
  name: "get_weather",
  description: "Get the weather for a city",
  parameters: z.object({
    city: z.string().describe("The city to get weather for"),
  }),
  async execute(input) {
    console.log(`[debug] getting weather for ${input.city}`);
    return `The weather in ${input.city} is sunny.`;
  },
});

// 4. Run agent
async function main() {
  const agent = new Agent({
    name: "Assistant",
    instructions:
      "You only respond in short sentences. Always mention temperature in Fahrenheit and wind speed as well.",
    model: process.env.MODEL_NAME,
    tools: [getWeather],
  });

  const runner = new Runner({ modelProvider });
  const result = await runner.run(agent, "What's the weather in Tokyo?");
  console.log(result.finalOutput);
}

main();

5. Key Points to Remember

• Always use the chat completions API for Gemini with OpenAI SDK.

• baseURL must be set to https://generativelanguage.googleapis.com/v1beta/openai/

• The MODEL_NAME must match exactly the Gemini model you want (gemini-2.5-flash, gemini-2.0-pro, etc).

• Disable tracing unless you have a real OpenAI API key from platform.openai.com

6. Debugging Common Issues

1. 401 Unauthorized
   Check that your Gemini API key is valid and correctly set in .env.

2. Model not found
   Double check the MODEL_NAME. If you mistype it, Gemini will reject the request.

3. Wrong API type
   If you forget setOpenAIAPI("chat_completions"), the SDK may try to use the newer responses API which Gemini does not yet support.

Conclusion

With just a few tweaks, you can connect Gemini to the OpenAI SDK and run agents through the OpenAI Agents SDK. The key is setting the baseURL, apiKey, and model correctly. Start small with direct chat calls, then move to agents with tools once your setup is confirmed.