[BUG] OpenAIModel does not work with Gemini OpenAI-compatible endpoints

[maximelebastard]

Checks

• I have updated to the lastest minor and patch version of Strands
• I have checked the documentation and this is not expected behavior
• I have searched ./issues and there are no duplicates of my issue

Strands Version

0.1.2

Node.js Version

23.10.0

Operating System

macOS

Installation Method

npm

Steps to Reproduce

Here is a reproduction code.
You have to get an OpenAI api key and a Google AI Studio api key - and provide them as env vars.

import { Agent, tool } from "@strands-agents/sdk";
import { OpenAIModel } from "@strands-agents/sdk/openai";
import OpenAI from "openai";
import { z } from "zod";

/**
 * Standalone repro for Strands SDK tool handling with two backends:
 * 1. Native OpenAI (expected to succeed)
 * 2. Google Gemini 2.5 Flash via the OpenAI-compatible endpoint (expected to fail)
 *
 * Run from the examples package:
 *   OPENAI_API_KEY=... GOOGLE_API_KEY=... pnpm tsx src/reproduce-strands-openapi-tool-issue.ts
 */

const getWeather = tool({
  name: "get_weather",
  description: "Get weather information for a city.",
  inputSchema: z.object({
    location: z
      .string()
      .describe("City or region the user is interested in."),
  }),
  async callback({ location }) {
    return {
      location,
      temperature: 72,
      conditions: "sunny",
    };
  },
});

const systemPrompt = [
  "You are a weather routing agent.",
  "You MUST call the `get_weather` tool exactly once before answering.",
  "Do not fabricate weather data yourself.",
].join(" ");

async function main(): Promise<void> {
  const openaiClient = new OpenAI({
    apiKey: process.env.OPENAI_API_KEY ?? "set-your-openai-api-key",
    baseURL: process.env.OPENAI_BASE_URL,
  });
  const geminiClient = new OpenAI({
    apiKey: process.env.GOOGLE_API_KEY ?? "set-your-google-api-key",
    baseURL:
      process.env.GEMINI_OPENAI_BASE ??
      "https://generativelanguage.googleapis.com/v1beta/openai/",
  });

  const scenarios = [
    {
      label: "OpenAI (gpt-4o-mini)",
      client: openaiClient,
      modelId: process.env.OPENAI_MODEL_ID ?? "gpt-4o-mini",
    },
    {
      label: "Gemini via OpenAI proxy (gemini-2.5-flash)",
      client: geminiClient,
      modelId: process.env.GEMINI_MODEL_ID ?? "gemini-2.5-flash",
    },
  ] as const;

  const results: Array<{
    label: string;
    success: boolean;
    error?: unknown;
  }> = [];

  for (const scenario of scenarios) {
    const summary = `[${scenario.label}]`;
    console.log(`${summary} invoking agent...`);

    const model = new OpenAIModel({
      client: scenario.client,
      modelId: scenario.modelId,
      temperature: 0,
    });

    const agent = new Agent({
      systemPrompt,
      model,
      tools: [getWeather],
    });

    try {
      await agent.invoke("Can you get the weather for San Francisco?");
      console.log(`${summary} ✅ succeeded`);
      results.push({ label: scenario.label, success: true });
    } catch (error) {
      console.error(`${summary} ❌ threw an error`, error);
      results.push({ label: scenario.label, success: false, error });
    }
  }
}

main().catch((error) => {
  console.error(error);
  process.exitCode = 1;
});

Expected Behavior

Tools use are working both with OpenAI and Google OpenAI compatible endpoints

Actual Behavior

It works with OpenAI
With Gemini, it throws an error about a missing tool_use block in the messages collection

Error: Model indicated toolUse but no tool use blocks found in message

Additional Context

No response

Possible Solution

No response

Related Issues

No response

[maximelebastard]

The issue may come from the OpenAI adapter issued by Gemini.
Trying to solve that by providing a direct Gemini model adapter in #331