Vercel AI SDK with APIBox: Use Claude, GPT, Gemini, and DeepSeek Through One Provider

The shortest way to use APIBox with the Vercel AI SDK is to configure an OpenAI-compatible provider with baseURL: "https://api.apibox.cc/v1". After that, generateText, streamText, tool calling, and model switching can use the same AI SDK patterns while APIBox routes requests to Claude, GPT, Gemini, DeepSeek, and other supported models.

1. The exact setup values

Use this setup when you need to:

• connect Vercel AI SDK to an OpenAI-compatible provider
• set one baseURL for multiple model families
• call Claude, GPT, Gemini, or DeepSeek from the same AI SDK code
• keep model switching inside configuration instead of rewriting routes
• use one provider pattern across Next.js and Node.js services

2. Install the required packages

npm install ai @ai-sdk/openai-compatible

Create an environment variable:

APIBOX_API_KEY=your_apibox_key

Do not expose this key in frontend code. Keep model calls in server routes, server actions, jobs, or backend services.

3. Create a shared APIBox provider

Create a small provider module so the rest of your app does not repeat baseURL and key handling.

// lib/apibox.ts
import { createOpenAICompatible } from '@ai-sdk/openai-compatible';

export const apibox = createOpenAICompatible({
  name: 'apibox',
  apiKey: process.env.APIBOX_API_KEY,
  baseURL: 'https://api.apibox.cc/v1',
});

Now every model call can import the same provider.

4. Generate text with APIBox

// app/api/summarize/route.ts
import { generateText } from 'ai';
import { apibox } from '@/lib/apibox';

export async function POST(req: Request) {
  const { text } = await req.json();

  const result = await generateText({
    model: apibox('gpt-5'),
    system: 'You summarize technical content clearly and concisely.',
    prompt: `Summarize this text:\n\n${text}`,
  });

  return Response.json({ summary: result.text });
}

The provider stays the same when you switch models:

model: apibox('claude-sonnet-4-6')
model: apibox('gemini-2.5-pro')
model: apibox('deepseek-r1')

5. Stream text for chat interfaces

Streaming is usually the better default for chat UI because users see output sooner.

// app/api/chat/route.ts
import { streamText } from 'ai';
import { apibox } from '@/lib/apibox';

export async function POST(req: Request) {
  const { messages } = await req.json();

  const result = streamText({
    model: apibox('claude-sonnet-4-6'),
    system: 'You are a precise assistant for software developers.',
    messages,
  });

  return result.toTextStreamResponse({
    headers: {
      'Content-Type': 'text/event-stream',
    },
  });
}

If you are building a production chat app, add request validation, authentication, rate limits, and logging before opening the route publicly.

6. How to choose models in a Vercel AI SDK app

The important point is not that one model is always best. It is that APIBox lets your AI SDK application keep a stable integration while you test and change models.

Related reading:

• LLM API Cost Estimation Guide
• Multi-Model Routing Guide
• Coding Model Comparison

7. Troubleshooting common errors

401 Unauthorized

Check:

• APIBOX_API_KEY exists in the runtime environment
• the key is not wrapped in quotes with extra spaces
• the server was restarted after editing .env.local
• the key is not being sent from browser code

404 model not found

Check the model name in the APIBox pricing or model list page. Do not assume official provider model IDs always match gateway model IDs.

Useful link:

• APIBox model pricing

Request hangs or streams slowly

Check:

• whether the selected model is a reasoning model
• output length
• your hosting timeout
• whether the route is running in the expected runtime

For streaming routes, keep the response as a stream instead of buffering the full answer first.

Tool calling behaves differently across models

OpenAI-compatible does not mean identical behavior. A tool schema that works with one model should still be tested with the exact models you plan to offer.

For tool design, read:

• What Is Function Calling?

8. Production checklist

Before shipping, verify:

• API key is server-side only
• model names are configured through environment variables or a small allowlist
• request size is limited
• user-level rate limits exist
• prompts are versioned
• errors are logged with model name and latency
• cost is tracked per route or feature
• fallback model behavior is explicit

9. Recommended setup

To use APIBox with the Vercel AI SDK, install ai and @ai-sdk/openai-compatible, create a provider with baseURL: "https://api.apibox.cc/v1", pass your APIBox key, and call models with generateText or streamText. This gives a Next.js or Node.js app one OpenAI-compatible integration path for Claude, GPT, Gemini, DeepSeek, and other models.

Register for APIBox to create an API key, then use the provider setup above in your Vercel AI SDK project.