GoldBean Client SDK

One client for 7 Chinese LLM providers.

npm install goldbean-client

Quick Start

const { createClient } = require('goldbean-client');

const gb = createClient('your-api-key');

// Simple ask
const response = await gb.ask('deepseek-v4-pro', 'Explain quantum computing in 3 sentences');
console.log(response.content);

// Full chat
const result = await gb.chat({
  model: 'kimi-k2.6',
  messages: [
    { role: 'system', content: 'You are a helpful assistant.' },
    { role: 'user', content: 'Write a Python function to reverse a linked list.' }
  ],
  maxTokens: 2000
});
console.log(result.content);
console.log(result.usage.total_tokens);

Supported Providers

| Provider | Models | Endpoint | |----------|--------|----------| | DeepSeek | V4-Pro, V4-Flash | /ai/ds/chat, /ai/ds/flash | | Kimi (Moonshot) | K2.6, K2.7-Code | /ai/kimi/chat, /ai/kimi/code | | Doubao (ByteDance) | Pro, Mini, Lite | /ai/doubao/chat, /ai/doubao/mini, /ai/doubao/lite | | Qwen (Alibaba) | Max, Turbo, Plus | /ai/qwen/chat, /ai/qwen/turbo, /ai/qwen/plus | | Hunyuan (Tencent) | Pro | /ai/hy3/chat | | ERNIE (Baidu) | 4.0 | /paid/baidu-llm-chat | | GLM (Zhipu) | 4 | /ai/glm/chat |

API Key

Get your API key at goldbean-api.xyz:

• Free: 2000 calls/day without signup
• Paid: From $5 for 100 calls (no expiry)

Methods

chat(params)

Send a chat completion request.

const response = await gb.chat({
  model: 'deepseek-v4-pro',    // Required
  messages: [                   // Required
    { role: 'user', content: 'Hello' }
  ],
  temperature: 0.7,            // Optional (ignored for Kimi, fixed at 1)
  maxTokens: 1000,             // Optional
});

ask(model, message, options)

Simple one-shot ask.

const response = await gb.ask('deepseek-v4-pro', 'What is 2+2?');

getModels(provider?)

List available models.

const models = await gb.getModels(); // All models
const deepseekModels = await gb.getModels('deepseek'); // Provider-specific

health(provider)

Check provider health.

const health = await gb.health('deepseek');

Response Format

{
  content: "The assistant's response text",
  provider: "deepseek",
  model: "deepseek-v4-pro",
  usage: {
    prompt_tokens: 12,
    completion_tokens: 56,
    total_tokens: 68
  },
  finishReason: "stop",
  raw: { /* full API response */ }
}

Error Handling

try {
  const response = await gb.chat({ model: 'unknown-model', messages: [] });
} catch (err) {
  console.error(err.message); // "Unknown model: unknown-model"
}

Retry Logic

The SDK automatically retries on:

• 429 Too Many Requests
• 5xx Server Errors
• Network errors

Default: 2 retries with exponential backoff.

License

MIT © GoldBean