@yavqo/aisdk

Official Yavqo AI SDK for Node.js — powered by Sunlight AI.

Installable via:

npm install @yavqo/aisdk
# or
pnpm add @yavqo/aisdk
# or
yarn add @yavqo/aisdk

Provides first-class support for:

• Sunlight 2 — fast, efficient coding model
• Sunlight 2 Pro — frontier reasoning model

Quick Start

import { createYavqo, sunlight2, sunlight2Pro } from '@yavqo/aisdk';

// Zero-config — works out of the box
const yavqo = createYavqo();

// Simple prompt
const { text } = await yavqo.generateText({
  model: sunlight2,
  prompt: 'Write a hello world in TypeScript',
});

console.log(text);

// With system + messages
const { text: text2 } = await yavqo.generateText({
  model: sunlight2Pro,
  system: 'You are a helpful coding assistant.',
  messages: [{ role: 'user', content: 'Explain closures in JS' }],
  temperature: 0.7,
  maxTokens: 1024,
});

Streaming

import { createYavqo, sunlight2Pro } from '@yavqo/aisdk';

const yavqo = createYavqo();

const { textStream, final } = await yavqo.streamText({
  model: sunlight2Pro,
  prompt: 'Write a short story about a robot learning to paint',
});

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

const result = await final;
console.log('\n---\nUsage:', result.usage);

Chat completions

import { Yavqo, sunlight2 } from '@yavqo/aisdk';

const client = new Yavqo();

const completion = await client.chat.completions.create({
  model: sunlight2,
  messages: [
    { role: 'system', content: 'You are Sunlight, a helpful AI.' },
    { role: 'user', content: 'Hello!' },
  ],
  temperature: 0.7,
});

console.log(completion.choices[0].message.content);

Structured output (JSON) — with Zod validation

import { createYavqo, sunlight2 } from '@yavqo/aisdk';
import { z } from 'zod';

const yavqo = createYavqo();

// Option A: raw JSON Schema
const { object } = await yavqo.generateObject<{ name: string; age: number }>({
  model: sunlight2,
  prompt: 'Generate a person with name and age',
  schema: {
    type: 'object',
    properties: {
      name: { type: 'string' },
      age: { type: 'number' },
    },
    required: ['name', 'age'],
    additionalProperties: false,
  },
  schemaName: 'person',
});
console.log(object); // { name: "Alice", age: 30 }

// Option B: Zod schema — validated locally
const Person = z.object({ name: z.string(), age: z.number() });
const { object: person } = await yavqo.generateObject({
  model: sunlight2,
  prompt: 'Generate a person',
  schema: Person,
});
console.log(person);

Tools / Function calling

const { text, toolCalls } = await yavqo.generateText({
  model: sunlight2Pro,
  messages: [{ role: 'user', content: 'What is the weather in Tokyo?' }],
  tools: [
    {
      type: 'function',
      function: {
        name: 'get_weather',
        description: 'Get weather for a city',
        parameters: {
          type: 'object',
          properties: { city: { type: 'string' } },
          required: ['city'],
        },
      },
    },
  ],
});

API Reference

createYavqo(config?) / new Yavqo(config?)

interface YavqoConfig {
  apiKey?: string;      // optional — defaults to built-in key
  baseURL?: string;     // optional — defaults to Sunlight AI gateway
  headers?: Record<string, string>;
  fetch?: typeof fetch;
  timeoutMs?: number;   // default 60_000 (0 = no timeout)
  maxRetries?: number;  // default 2
  retry?: {
    maxRetries?: number;
    initialDelayMs?: number;
    maxDelayMs?: number;
    backoffFactor?: number;
    jitter?: boolean;
    retryStatusCodes?: number[];
  };
}

Models

import { sunlight2, sunlight2Pro } from '@yavqo/aisdk';

sunlight2        // Sunlight 2
sunlight2Pro     // Sunlight 2 Pro

Methods

| Method | Description | |--------|-------------| | yavqo.generateText(opts) | Non-streaming text generation | | yavqo.streamText(opts) | Streaming generation | | yavqo.generateObject(opts) | JSON object generation (with optional schema) | | yavqo.chat.completions.create(opts) | Low-level chat completions | | yavqo.chat.completions.stream(opts) | Low-level streaming | | yavqo.listModels() | List available models |

Common options

{
  model: string;            // required
  prompt?: string;
  system?: string;
  messages?: ChatMessage[];
  temperature?: number;
  top_p?: number;
  maxTokens?: number;
  tools?: Tool[];
  tool_choice?: ToolChoice;
  stop?: string | string[];
  responseFormat?: { type: 'text' | 'json_object' | 'json_schema' };
  headers?: Record<string, string>;
  extraBody?: Record<string, unknown>;
  signal?: AbortSignal;
  maxRetries?: number;
  retry?: RetryConfig;
}

Retry & backoff

Automatic retry for transient failures (429, 500, 502, 503, 504, 529) with exponential backoff + jitter and Retry-After support.

const yavqo = createYavqo({
  maxRetries: 3,
  retry: { initialDelayMs: 300, maxDelayMs: 5000, backoffFactor: 2 },
});

CJS + ESM

// ESM
import { createYavqo } from '@yavqo/aisdk';
// CJS
const { createYavqo } = require('@yavqo/aisdk');

Testing

npm test
npm run lint

TypeScript

Fully typed. Requires Node >= 18.

License

MIT © Yavqo