Supporting custom fetch in options for http proxy.

[WeiAnAn]

Is your feature request related to a problem? Please describe.

The current @google/genai SDK does not offer a direct method to configure an HTTP/HTTPS proxy for its API requests. This makes it challenging to use the SDK's clients in common corporate or restricted network environments that require a proxy.

---

Describe the solution you'd like

I propose adding a fetchImplementation (or alternatively, fetch) property to the configuration object passed to the SDK's client constructors. This would allow developers to supply their own fetch implementation, making it easy to integrate libraries like http-proxy-agent to correctly route traffic.

---

Describe alternatives you've considered

1. Direct Proxy Options: Adding explicit proxy fields (e.g., { proxy: { host: '...' } }) is an option, but it's less flexible than fetchImplementation for handling complex authentication, custom certificates, or different agent behaviors.

2. Relying on Environment Variables: Using a global environment variable like HTTPS_PROXY is a significant drawback because it affects the entire Node.js process. This forces all outbound HTTPS requests from the application through the proxy, not just requests from this SDK. An explicit fetchImplementation option provides safe, instance-level control without causing unintended side effects.

---

Additional context

Here is a conceptual code example illustrating how the proposed solution would work.

import { GoogleGenAI } from '@google/genai';
import { fetch } from "undici";
import { ProxyAgent } from "proxy-agent";

const agent = new ProxyAgent();
const customFetchWithProxy = (url, options) => {
  return fetch(url, { ...options, dispatcher: agent });
};

// Pass the custom fetch function during client initialization
const ai = new GoogleGenAI({
  vertexai: true,
  project: 'your_project_id',
  location: 'us-central1',
  fetchImplementation: customFetchWithProxy,
});

[Kludex]

Please call this parameter fetch.

[shmishra99]

Hi @WeiAnAn, Thank you for your feedback. We appreciate you taking the time to share your thoughts with us, and we'll be filing a feature request.

[yapnel]

+1 to this feature request. It is impossible to use this SDK in corporate environment without the ability to set proxy configuration.

Thanks

[Myriad-Dreamin]

I made a hacking script to use proxy before the PR #1215 merged.

/**
 * A wrapper around the GoogleGenAI class that uses a custom fetch implementation.
 */
class GenAI extends GoogleGenAI {
  constructor(
    apiKey: string,
    private readonly fetcher: (
      url: string,
      requestInit?: RequestInit,
    ) => Promise<Response>,
  ) {
    super({
      apiKey,
    });
    if (!(this.apiClient as any).apiCall) {
      throw new Error("apiCall cannot be patched");
    }
    // Monkey patches apiCall to use our fetch.
    (this.apiClient as any).apiCall = async (url: string, requestInit: any) => {
      return this.fetcher(url, requestInit).catch((e) => {
        throw new Error(`exception ${e} sending request`);
      });
    };
  }
}

async function genAiExample() {
  /**
   * The fetch should be optionally given by the classes who constructs `GenAI`
   */
  const genAi = new GenAI(
    apiKey,
    new FetchWithProxy(https_proxy).createFetcher(),
  );

  // generate hello world

  const response = await genAi.models.generateContent({
    model: "gemini-2.5-flash",
    contents: ["Hello, world!"],
  });

  console.log(response.text);
}

[etaloner]

Add fetchOptions also at the per-request level (in the generateContent method). In the OpenAI npm package, this feature is available.

[etaloner]

@Myriad-Dreamin Do you have similar code for per-request level calls? (generateContent method)

[vaibhavshn]

+1, need this for cloudflare/ai#389.

Other SDKs have a way for passing custom fetch implementations, would appreciate this.