> ## Documentation Index
> Fetch the complete documentation index at: https://docs.crazyrouter.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Crazyrouter SDK Complete Guide

> Use the official OpenAI SDK with Crazyrouter and run an end-to-end test for chat, streaming, model listing, embeddings, and native async routes

> Last updated: 2026-06-23

## When To Use This Guide

Crazyrouter supports the official OpenAI SDK for OpenAI-compatible routes. Use it for chat, Responses API, images, audio, embeddings, and model listing. For video, Kling, Luma, Suno, Midjourney, and other async or provider-native routes, use the same API key with `fetch` against Crazyrouter native paths.

Default international API entry:

```text theme={null}
https://api.crazyrouter.com
```

OpenAI SDK base URL:

```text theme={null}
https://api.crazyrouter.com/v1
```

<Warning>
  Do not set the OpenAI SDK `baseURL` to `https://api.crazyrouter.com`, because the SDK expects a `/v1` base. Do not set it to a full endpoint such as `https://api.crazyrouter.com/v1/chat/completions`, because the SDK appends endpoint paths itself.
</Warning>

## Install The SDK

```bash theme={null}
npm install openai
```

Use Node.js 18 or later. Node.js 18+ includes `fetch`, `FormData`, and `Blob`.

## Configure Environment Variables

macOS / Linux:

```bash theme={null}
export CRAZYROUTER_API_KEY="sk-your-api-key"
export CRAZYROUTER_BASE_URL="https://api.crazyrouter.com"
```

Windows PowerShell:

```powershell theme={null}
$env:CRAZYROUTER_API_KEY = "sk-your-api-key"
$env:CRAZYROUTER_BASE_URL = "https://api.crazyrouter.com"
```

## Create A Client

```ts theme={null}
import OpenAI from 'openai';

const apiKey = process.env.CRAZYROUTER_API_KEY;

if (!apiKey) {
  throw new Error('Missing CRAZYROUTER_API_KEY');
}

const baseURL = process.env.CRAZYROUTER_BASE_URL ?? 'https://api.crazyrouter.com';

export const client = new OpenAI({
  apiKey,
  baseURL: `${baseURL}/v1`,
  timeout: 180_000,
});
```

## Chat Completions

```ts theme={null}
const response = await client.chat.completions.create({
  model: 'gpt-5.5',
  messages: [
    { role: 'system', content: 'You are concise.' },
    { role: 'user', content: 'Say sdk-ok and nothing else.' },
  ],
  max_tokens: 16,
});

console.log(response.choices[0]?.message?.content);
```

## Streaming

```ts theme={null}
const stream = await client.chat.completions.create({
  model: 'gpt-5.5',
  messages: [{ role: 'user', content: 'Count from 1 to 3.' }],
  max_tokens: 32,
  stream: true,
});

for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? '');
}
```

## Model Listing

```ts theme={null}
const models = await client.models.list();

for (const model of models.data.slice(0, 10)) {
  console.log(model.id);
}
```

## Embeddings

```ts theme={null}
const embedding = await client.embeddings.create({
  model: 'text-embedding-3-small',
  input: 'Crazyrouter SDK test',
});

console.log(embedding.data[0].embedding.length);
```

## Responses API

Use the Responses API for GPT models that support `/v1/responses`.

```ts theme={null}
const response = await client.responses.create({
  model: 'gpt-5.5',
  input: 'Return only: responses-ok',
  max_output_tokens: 16,
});

console.log(response.output_text);
```

<Note>
  Do not test Claude models through the Responses API. Use `/v1/messages` or OpenAI-compatible `/v1/chat/completions` for Claude.
</Note>

## Images, Audio, And Transcription

OpenAI-compatible image and audio routes use the same SDK client.

```ts theme={null}
const image = await client.images.generate({
  model: 'gpt-image-2',
  prompt: 'A simple white mug on a clean table',
  size: '1024x1024',
  output_format: 'png',
});

console.log(image.data[0]?.url);
```

```ts theme={null}
const speech = await client.audio.speech.create({
  model: 'tts-1',
  voice: 'alloy',
  input: 'Hello from Crazyrouter.',
});

const audio = Buffer.from(await speech.arrayBuffer());
console.log(audio.length);
```

## Native Async Task Routes

Video, Kling, Luma, Suno, Midjourney, and similar async tasks are usually not OpenAI SDK methods. Call Crazyrouter native HTTP paths with the same API key.

```ts theme={null}
async function crazyrouterFetch<T>(
  path: string,
  options: RequestInit & { json?: Record<string, unknown> } = {},
): Promise<T> {
  const response = await fetch(`${baseURL}${path}`, {
    ...options,
    headers: {
      Authorization: `Bearer ${apiKey}`,
      ...(options.json ? { 'Content-Type': 'application/json' } : {}),
      ...options.headers,
    },
    body: options.json ? JSON.stringify(options.json) : options.body,
  });

  const body = await response.json().catch(() => ({}));
  if (!response.ok) {
    throw new Error(`Crazyrouter API error ${response.status}: ${JSON.stringify(body)}`);
  }

  return body as T;
}
```

Create a unified video task:

```ts theme={null}
const task = await crazyrouterFetch<{
  id?: string;
  task_id?: string;
  data?: { task_id?: string };
}>('/v1/video/create', {
  method: 'POST',
  json: {
    model: 'veo-3.1-fast',
    prompt: 'A quiet city street after rain, cinematic lighting',
    aspect_ratio: '16:9',
    size: '1080P',
  },
});

const taskId = task.data?.task_id ?? task.task_id ?? task.id;
console.log(taskId);
```

Query the task:

```ts theme={null}
const result = await crazyrouterFetch(
  `/v1/video/query?task_id=${encodeURIComponent(taskId!)}`,
);

console.log(result);
```

## Upload A Local Image

For image-to-video, vision, and image-editing inputs, upload local reference images first.

```ts theme={null}
import { readFile } from 'node:fs/promises';
import { basename } from 'node:path';

async function uploadImage(filePath: string) {
  const bytes = await readFile(filePath);
  const form = new FormData();

  form.append(
    'file',
    new Blob([bytes], { type: 'image/png' }),
    basename(filePath),
  );
  form.append('purpose', 'model_input');

  const response = await fetch(`${baseURL}/v1/files/uploads`, {
    method: 'POST',
    headers: {
      Authorization: `Bearer ${apiKey}`,
    },
    body: form,
  });

  const body = await response.json();
  if (!response.ok) {
    throw new Error(`Upload failed: ${JSON.stringify(body)}`);
  }

  return body as {
    id: string;
    url: string;
    mime_type: string;
    size: number;
    expires_at: string;
  };
}

const uploaded = await uploadImage('./reference.png');
console.log(uploaded.url);
```

<Warning>
  Temporary upload URLs are for model inputs, not permanent asset hosting.
</Warning>

## Complete SDK Smoke Test

Save this as `crazyrouter-sdk-smoke.mjs`:

```js theme={null}
import OpenAI from 'openai';

const apiKey = process.env.CRAZYROUTER_API_KEY;
const rootBaseURL = process.env.CRAZYROUTER_BASE_URL ?? 'https://api.crazyrouter.com';

if (!apiKey) {
  throw new Error('Set CRAZYROUTER_API_KEY before running this test.');
}

const client = new OpenAI({
  apiKey,
  baseURL: `${rootBaseURL}/v1`,
  timeout: 180_000,
});

const checks = [];

async function check(name, fn) {
  const started = Date.now();
  try {
    const detail = await fn();
    checks.push({ name, ok: true, ms: Date.now() - started, detail });
    console.log(`ok ${name}: ${detail}`);
  } catch (error) {
    checks.push({ name, ok: false, ms: Date.now() - started, detail: error.message });
    console.error(`fail ${name}: ${error.message}`);
  }
}

await check('models.list', async () => {
  const models = await client.models.list();
  return `${models.data.length} models`;
});

await check('chat.completions.create', async () => {
  const response = await client.chat.completions.create({
    model: 'gpt-5.5',
    messages: [{ role: 'user', content: 'Return exactly sdk-ok' }],
    max_tokens: 16,
  });
  return response.choices[0]?.message?.content?.trim() ?? '(empty)';
});

await check('chat.completions.stream', async () => {
  const stream = await client.chat.completions.create({
    model: 'gpt-5.5',
    messages: [{ role: 'user', content: 'Return exactly stream-ok' }],
    max_tokens: 16,
    stream: true,
  });

  let text = '';
  for await (const chunk of stream) {
    text += chunk.choices[0]?.delta?.content ?? '';
  }
  return text.trim() || '(empty stream)';
});

await check('embeddings.create', async () => {
  const response = await client.embeddings.create({
    model: 'text-embedding-3-small',
    input: 'Crazyrouter SDK smoke test',
  });
  return `${response.data[0].embedding.length} dimensions`;
});

await check('raw /v1/video/query route shape', async () => {
  const response = await fetch(`${rootBaseURL}/v1/video/query?task_id=doc-smoke-test`, {
    method: 'GET',
    headers: {
      Authorization: `Bearer ${apiKey}`,
    },
  });

  const text = await response.text();
  if (response.status === 404 && text.toLowerCase().includes('not found')) {
    throw new Error('route returned a path-level 404');
  }

  return `route reachable, status ${response.status}`;
});

const failed = checks.filter((item) => !item.ok);
console.log(JSON.stringify({ total: checks.length, failed: failed.length, checks }, null, 2));

if (failed.length > 0) {
  process.exit(1);
}
```

Run it:

```bash theme={null}
npm install openai
node crazyrouter-sdk-smoke.mjs
```

A successful test should verify:

* `models.list` returns the available model count
* `chat.completions.create` returns text
* `chat.completions.stream` receives streamed text
* `embeddings.create` returns an embedding vector
* `/v1/video/query` returns a task-not-found or other business status, which confirms the native async query route exists

## Troubleshooting

| Symptom                            | Fix                                                                                                                 |
| ---------------------------------- | ------------------------------------------------------------------------------------------------------------------- |
| `401 Unauthorized`                 | Check `CRAZYROUTER_API_KEY` and the `Authorization: Bearer sk-...` header                                           |
| `404` or `/v1/v1/...`              | OpenAI SDK uses `https://api.crazyrouter.com/v1`; native HTTP uses `https://api.crazyrouter.com` plus the full path |
| Chat works but Responses API fails | Check whether the model supports `/v1/responses`; do not use Claude on Responses API                                |
| Image or video request times out   | Increase SDK timeout; prefer async task routes for video                                                            |
| `403 Forbidden`                    | The token is not allowed to call the selected model or route                                                        |