Connect Vercel AI SDK to XiuRouter

Use an OpenAI-compatible server-side provider and verify text generation, streaming, and usage.

Use @ai-sdk/openai-compatible to connect Vercel AI SDK to XiuRouter Chat Completions. Keep both the provider and API key on the server, outside browser bundles.

Setup options

First setup: install the provider, save a server environment variable, and create a XiuRouter instance.
Existing setup: the product page offers only the first-setup path. Retain the old provider instance when switching so you can roll back.
Chat-assisted setup: unavailable; the deployment environment manages secrets and production variables.

Before you start

Use a JavaScript or TypeScript project with server-side execution, a project-specific XiuRouter key, and an exact model ID from the current catalog.

Install dependencies

npm install ai @ai-sdk/openai-compatible

Use @ai-sdk/openai-compatible, not @ai-sdk/openai.

Save the server key

Use .env.local in Next.js or an uncommitted .env in a Node project:

XIUROUTER_API_KEY=YOUR_XIUROUTER_API_KEY

Ignore the file in Git. Do not prefix the variable with NEXT_PUBLIC_, which exposes it to the browser.

Create the provider

Create a server-only file such as lib/xiurouter.ts:

import { createOpenAICompatible } from "@ai-sdk/openai-compatible";

export const xiurouter = createOpenAICompatible({
  name: "xiurouter",
  apiKey: process.env.XIUROUTER_API_KEY,
  baseURL: "https://router-api.xiu.ai/v1",
});

Generate text

import { generateText } from "ai";
import { xiurouter } from "@/lib/xiurouter";

const { text, usage } = await generateText({
  model: xiurouter("YOUR_MODEL_ID"),
  prompt: "Explain why the sky is blue in two sentences.",
});

console.log(text);
console.log(usage);

The script should print a complete response and token usage.

Stream output

import { streamText } from "ai";
import { xiurouter } from "@/lib/xiurouter";

const result = streamText({
  model: xiurouter("YOUR_MODEL_ID"),
  prompt: "Write a four-line poem.",
});

for await (const textPart of result.textStream) {
  process.stdout.write(textPart);
}

Verify

Run the generateText example on the server.
Confirm both text and usage are returned.
Inspect the key, model, /v1/chat/completions, and successful result in XiuRouter.
Run streamText and verify that the stream finishes completely.
Test tools and structured output separately for the selected model.

Troubleshoot

Symptom	Next step
`apiKey` is `undefined`	Restart Next.js; for Node use `node --env-file=.env` or set the startup environment
Browser CORS errors or exposed key	Move the provider from client code to a Server Component, Route Handler, or backend
Model not found	Copy the exact ID and check the key’s model scope
Wrong endpoint	Check the import and ensure Base URL ends with `/v1`

Roll back

Restore the old provider instance, or remove xiurouter and its environment variable. Confirm production traffic uses the old route before revoking the project key.

Sources and review date

Reviewed on 2026-08-31. This translation preserves that date. Basic setup requires a full server response and matching usage record; verify tools and structured output with the target model.