TanStack Start dev server fails with NitroViteError: Vite environment "ssr" is unavailable after running @hey-api/openapi-ts

[JosefStanek]

Hi,

I’m trying to integrate Hey API into a fresh TanStack Start project.

I’m not sure whether this is a bug in Hey API, TanStack Start, Nitro, or if I’m missing some configuration.

Environment

• Windows 11
• Node.js v22.17.0
• @tanstack/react-start 1.168.26
• vite 8.1.0
• nitro-nightly 3.0.1

Steps to reproduce

1. Create a new TanStack Start project.
2. Verify that npm run dev works.
3. Create an openapi-ts.config.ts:

export default {

input: "http://localhost:5239/openapi/v1.json",

output: "../../client/src/api/generated",

plugins: ["@hey-api/client-fetch"],

};

4. Run npx @hey-api/openapi-ts
5. Start the TanStack Start application again using: npm run dev

Expected behavior

The application should start normally.

Actual behavior

The development server immediately fails with: NitroViteError: Vite environment "ssr" is unavailable

Stack trace:

Vite environment "ssr" is unavailable
at httpError (.../nitro/dist/runtime/internal/vite/dev-worker.mjs:208:17)
at ViteEnvRunner.fetch (.../dev-worker.mjs:77:13)
at Module.fetch (.../dev-worker.mjs:154:12)

What I already verified

• package.json does not change.
• package-lock.json does not change.
• npm ls vite reports the same version before and after running the generator.
• Only one version of Vite is installed.
• The generated files are not imported anywhere.
• Deleting the generated files does not fix the issue.
• Generating outside src does not fix the issue.
• Clearing .vite, .nitro, .output, reinstalling dependencies and node_modules does not resolve the problem.

At this point it seems that simply running the generator is enough to leave the TanStack Start dev server in a broken state, even though the generated files are never imported.

Is this a known issue, or is there a recommended way to integrate Hey API with TanStack Start?

[VectorPeak]

@hey-api/openapi-ts shouldn’t touch Vite/Nitro state. It reads the config, fetches the OpenAPI spec, and writes files to output; if package.json / lockfile don’t change and the generated files aren’t imported, there isn’t an expected path where it would make the TanStack Start dev server fail.

One cache worth checking for TanStack Start is .vinxi in addition to .vite, .nitro, and .output:

rm -rf .vinxi .vite .nitro .output node_modules
npm install
npm run dev

The recommended integration is just to keep generation as an explicit script, ideally writing inside the app package:

{
 "scripts": {
 "api:generate": "openapi-ts"
 },
 "devDependencies": {
 "@hey-api/openapi-ts": "latest"
 }
}

not sure about the Nitro/Vite "ssr" is unavailable failure, worth checking whether TanStack/Vinxi is reacting to some file created outside the app root or a stale .vinxi cache. If you can share a minimal repo where npm run dev works, then npx @hey-api/openapi-ts, then npm run dev fails with no imports, you can verify whether there’s any generator side effect we’re missing.

[JosefStanek]

I reproduced the issue without @hey-api/openapi-ts. A fresh TanStack Start project works with deployment adapter None, but fails with Nitro adapter. Clearing .vinxi, .vite, .nitro, .output and node_modules does not help. The error is "Vite environment ssr is unavailable" from nitro/dist/runtime/internal/vite/dev-worker.mjs.
Thanks for help

[mrlubos]

@VectorPeak Appreciate your help!