Skip to content
Source
Sign inSign up
TanStack/ai

ai-gemini: `httpOptions.fetch` is ignored since 1.1.1, so requests go to the global `fetch` #2112

Send to my agent
Closedkurochenko(GitHub profile, opens in a new tab) opened this issue 0 comments

TanStack AI version

@tanstack/ai-gemini 1.2.0 (also 1.1.1). The last good version is 1.1.0. @tanstack/ai 1.2.0.

Framework/Library version

Node.js 24.22.1, @google/genai 2.32.0 (what ^2.10.0 installs today). Server code, no UI framework.

Describe the bug and the steps to reproduce it

With an apiKey, the Gemini adapters ignore httpOptions.fetch. Every request goes to the global fetch. Nothing throws, so requests silently skip a proxy, a Cloudflare binding, a data-residency egress, or a logging fetch.

@google/genai has httpOptions.fetch since 2.23.0. In 1.1.0 the adapter sends requests through it. In 1.1.1 and 1.2.0 it does not.

Steps:

  1. Run npm i @tanstack/ai@1.2.0 @tanstack/ai-gemini@1.2.0 zod in an empty ES module project.
  2. Save the script below as repro.mjs.
  3. Run node repro.mjs (Node 24).
import { chat } from '@tanstack/ai'
import { createGeminiChat } from '@tanstack/ai-gemini'

// Fail loudly if a request skips the custom fetch.
globalThis.fetch = async (input) => {
  throw new Error(`global fetch called for ${input}`)
}

let customFetchCalls = 0
const myFetch = async () => {
  customFetchCalls++
  const chunk = {
    candidates: [
      { content: { role: 'model', parts: [{ text: 'hi' }] }, finishReason: 'STOP' },
    ],
  }
  return new Response(`data: ${JSON.stringify(chunk)}\n\n`, {
    headers: { 'content-type': 'text/event-stream' },
  })
}

const adapter = createGeminiChat('gemini-2.5-flash', 'fake-key', {
  httpOptions: { fetch: myFetch },
})

async function main() {
  let text = ''
  let error
  for await (const chunk of chat({
    threadId: 'repro',
    adapter,
    messages: [{ role: 'user', content: 'Hi' }],
  })) {
    if (chunk.type === 'TEXT_MESSAGE_CONTENT') text += chunk.delta
    if (chunk.type === 'RUN_ERROR') error = chunk.message
  }
  console.log(JSON.stringify({ text, customFetchCalls, error }))
}

main()

Expected (this is the 1.1.0 output):

{"text":"hi","customFetchCalls":1}

Actual on 1.1.1 and 1.2.0:

{"text":"","customFetchCalls":0,"error":"global fetch called for https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:streamGenerateContent?alt=sse"}

Where it comes from: #1973 (released in 1.1.1) replaced the SDK with a fetch client when the config has an apiKey. On main:

  • apiCall in packages/ai-gemini/src/client/api.ts calls the global fetch. It never reads httpOptions.fetch.
  • createInteractions in src/client/interactions.ts does not pass fetch to createHttpClient. So image generation (createGeminiImage) skips the custom fetch too.

Vertex AI and enterprise still go through the SDK and are not affected.

https://stackblitz.com/github/kurochenko/tanstack-ai-gemini-fetch-repro

Run npm start. It is the script above. It needs no API key and makes no network call. To see 1.1.0 work, set @tanstack/ai-gemini to 1.1.0 in package.json and run npm install.

Do you intend to try to help solve this bug with your own PR?

Yes, I am also opening a PR that solves the problem along side this issue

Terms & Code of Conduct

  • I agree to follow this project's Code of Conduct
  • I understand that if my bug cannot be reliable reproduced in a debuggable environment, it will probably not be fixed and this issue may even be closed.
  1. tanstack-source added labelsbug

  2. github-actions[bot](GitHub profile, opens in a new tab) added labelswaiting-on: maintainer

  3. AlemTuzlak closed this

Sign in to join this conversation.