Using Gemini with OpenAI Agents SDK

API
,
1

Hello everyone,

I’m trying to use Gemini API with OpenAI’s Agents SDK TS but it doesn’t work. Maybe you know a way to make it work?

import { Agent, Runner } from '@openai/agents';
import { aisdk } from '@openai/agents-extensions';
import { google } from '@ai-sdk/google';

 new Agent<AgentContextStorefront, typeof AgentsResponseStorefront>({
        name: cfg.name,
        instructions: promptWithHandoffInstructionsAndPersonality(cfg.instructions, multiAgentSystem.systemInstructions, context),
        model: aisdk(google('gemini-2.5-flash')),
        tools: usedTools,
        handoffs: [],              
        handoffDescription: cfg.handoffDescription,
        outputType: AgentsResponseStorefront,
        toolUseBehavior: "run_llm_again",
        modelSettings: {
          toolChoice: cfg.modelSettings?.toolChoice !== undefined ? cfg.modelSettings?.toolChoice : 'auto',
          parallelToolCalls: cfg.modelSettings?.parallelToolCalls !== undefined ? cfg.modelSettings?.parallelToolCalls : false,

        }
      });

When I run, I get:

AI_APICallError: Function calling with a response mime type: 'application/json' is unsupported
    at <anonymous> (/Users/bakikucukcakiroglu/Workspace/lookfor/backend/node_modules/@ai-sdk/provider-utils/src/response-handler.ts:59:16)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at postToApi (/Users/bakikucukcakiroglu/Workspace/lookfor/backend/node_modules/@ai-sdk/provider-utils/src/post-to-api.ts:111:28)
    at GoogleGenerativeAILanguageModel.doStream (/Users/bakikucukcakiroglu/Workspace/lookfor/backend/node_modules/@ai-sdk/google/src/google-generative-ai-language-model.ts:312:50)
    at AiSdkModel.getStreamedResponse (/Users/bakikucukcakiroglu/Workspace/lookfor/backend/node_modules/@openai/agents-extensions/src/aiSdk.ts:615:26)
    at Runner.#runStreamLoop (/Users/bakikucukcakiroglu/Workspace/lookfor/backend/node_modules/@openai/agents-core/src/run.ts:731:28)
2

I don’t see where you are specifying the different base URL?

Looks like you can specify it in your environment variables:

1 Like
3

I don’t really know how the problem is related to baseURL but here you should be able to see where it is set.

4

How? This is not your code or environment it’s just documentation. How does that prove you’ve followed it?

Of all the things I might see, your environment variables are (rightly) not amongst them :slight_smile:

Did you include them in the OP? No.

So how did you expect me to know what you included? For all I knew you hadn’t set it.

You reported this problem, and this is one of the obvious potential issues.

5

Below is the code that worked for me

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

// Load environment variables from .env file
config();

if (!process.env.BASE_URL || !process.env.API_KEY || !process.env.MODEL_NAME) {
  throw new Error(
    "Please set EXAMPLE_BASE_URL, EXAMPLE_API_KEY, EXAMPLE_MODEL_NAME via env var or code."
  );
}

/**
 * This example uses a custom provider for all requests by default. We do three things:
 * 1. Create a custom client.
 * 2. Set it as the default OpenAI client, and don't use it for tracing.
 * 3. Set the default API as Chat Completions, as most LLM providers don't yet support Responses API.
 *
 * Note that in this example, we do not set up tracing, under the assumption that you don't have an API key
 * from platform.openai.com. If you do have one, you can set the `OPENAI_API_KEY` env var for tracing.
 */

// Create a custom OpenAI client and provider
const openaiClient = new OpenAI({
  apiKey: process.env.API_KEY,
  baseURL: process.env.BASE_URL,
});
const modelProvider = new OpenAIProvider({
  openAIClient: openaiClient,
});
setDefaultOpenAIClient(openaiClient); // Pass the OpenAI client instance
setOpenAIAPI("chat_completions");
setTracingDisabled(true);

// Tool definition
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) {
    // input: { city: string }
    console.log(`[debug] getting weather for ${input.city}`);
    return `The weather in ${input.city} is sunny.`;
  },
});

async function main() {
  const agent = new Agent({
    name: "Assistant",
    instructions:
      "You only respond in short sentences. Mention temperature in Farenheit 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 Washington D.C?"
  );
  console.log(result.finalOutput);
}

main();

.env file

BASE_URL=https://generativelanguage.googleapis.com/v1beta/openai/
API_KEY=<YourGemini-Key>
MODEL_NAME=gemini-2.5-flash
1 Like