Skip to main content

​
Vercel AI SDK

Use Vercel’s ai package together with Incredible’s OpenAI-compatible endpoint by creating a client with @ai-sdk/openai-compatible. The example below mirrors the basic_stream.ts sample bundled with the Incredible SDK tests and shows the minimal pieces you need to stream completions.

​
Requirements

  • Node.js 18 or newer
  • An Incredible API key (the API currently works without one, but plan for it)
  • Packages: ai, @ai-sdk/openai-compatible
npm install ai @ai-sdk/openai-compatible
# or: pnpm add ai @ai-sdk/openai-compatible
Set your API key (if required) before running the script: 
export INCREDIBLE_API_KEY="your incredible api key"

​
Configure the Incredible client

At the top of your script, import both the OpenAI-compatible helper and Vercel’s streamText runner. basic_stream.ts uses these lines to plug Incredible into any Vercel AI SDK project: 
import { createOpenAICompatible } from '@ai-sdk/openai-compatible';
import { streamText } from 'ai';

const incredible = createOpenAICompatible({
  baseURL: 'https://api.incredible.one/v1',
  apiKey: process.env.INCREDIBLE_API_KEY ?? 'your incredible api key',
  name: 'incredible',
  headers: {
    'User-Agent': 'Mozilla/5.0',
  },
});
  • createOpenAICompatible gives you a drop-in client for any OpenAI-style model ID (use small-2 for public access).
  • The optional User-Agent header mirrors the sample script; keep it if your environment standardises outbound requests.

​
Stream a completion

The full sample streams tokens to stdout while keeping the rest of the script light: 
async function main() {
  const prompt = process.argv[2] ?? 'Say hello in one sentence.';
  const modelId = 'small-2';

  const result = await streamText({
    model: incredible(modelId),
    messages: [{ role: 'user', content: prompt }],
    maxRetries: 1,
  });

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

  process.stdout.write('\n');
}

main().catch((error) => {
  console.error('Vercel AI SDK request failed', error);
  process.exit(1);
});
  • Pass any prompt as a CLI argument: node basic_stream.mjs "Generate a haiku".
  • streamText yields an async iterator, so you can render the stream in a UI instead of stdout.

​
Next steps

  • Swap in other Incredible model IDs (check the Models page for current options).
  • Handle retries and errors with your own policy by wrapping the streamText call.
  • Pair this setup with Vercel AI SDK’s React hooks for server actions or edge streaming.