js-ai-core

Core HTTP/TLS client for AI providers with zero dependencies. Today this package ships:

• HttpClient: a minimal TLS + raw HTTP/1.1 client
• OpenAIProvider: wrapper for the OpenAI Responses API

Install

npm install @jeandev_2030/js-ai-core

Quick start (OpenAI)

const { OpenAIProvider } = require('@jeandev_2030/js-ai-core');

const client = new OpenAIProvider({
  apiKey: process.env.OPENAI_API_KEY
});

async function run() {
  const response = await client.createResponse({
    model: 'gpt-4.1-mini',
    input: 'Ola!'
  });

  console.log('status:', response.status);
  console.log('id:', response.id);
  console.log('text:', client.extractText(response));
}

run().catch(console.error);

API

HttpClient

const { HttpClient } = require('@jeandev_2030/js-ai-core');
const http = new HttpClient({ timeout: 30_000 });

Methods:

• request({ method, url, headers, body })
• get(url, options)
• post(url, body, options)

Return value:

{
  status: Number,
  headers: Object, // lowercased header names
  data: Object | String | null
}

Notes:

• Uses TLS sockets and raw HTTP/1.1 requests
• No redirects, retries, or streaming
• Default headers: Content-Type: application/json, User-Agent: js-ai-core/1.0, Connection: close

OpenAIProvider

const { OpenAIProvider } = require('@jeandev_2030/js-ai-core');
const openai = new OpenAIProvider({
  apiKey: process.env.OPENAI_API_KEY,
  baseUrl: 'https://api.openai.com/v1' // optional
});

Methods:

• createResponse({ model, input, headers, options })
  • headers are merged into the request headers
  • options is merged into the JSON payload sent to /responses
• extractText(response) returns concatenated text from response.raw.output

Errors:

If the API returns status >= 400, the provider throws:

{
  provider: 'openai',
  status: Number,
  message: String,
  code: String | null,
  type: String | null,
  raw: Object
}

Node support

• Node.js >= 18

License

MIT