OpenAI-compatible providers

Any service with an OpenAI chat-completions API works with the same chatbot code: OpenRouter, Groq, DeepSeek, xAI, Together, a local Ollama or LM Studio, or your own endpoint.

Presets

Provider	Base URL	Default model
openrouter	https://openrouter.ai/api/v1	openai/gpt-5.5
deepseek	https://api.deepseek.com	deepseek-chat
groq	https://api.groq.com/openai/v1	set model
xai	https://api.x.ai/v1	set model
together	https://api.together.xyz/v1	set model
ollama	http://localhost:11434/v1	set model, no key needed
lmstudio	http://localhost:1234/v1	set model, no key needed
openai_compatible	your baseUrl	set model

Chat

const { Chatbot, OpenAICompatibleInput } = require('intellinode');

// hundreds of models through one key
const bot = new Chatbot(OPENROUTER_API_KEY, 'openrouter');
const input = new OpenAICompatibleInput('You are a helpful assistant.', { model: 'anthropic/claude-sonnet-5' });
input.addUserMessage('Who painted the Mona Lisa?');

const responses = await bot.chat(input);

The default model of a preset can be set once on the chatbot instead of every input:

const groq = new Chatbot(GROQ_API_KEY, 'groq', null, { model: 'llama-3.3-70b-versatile' });

Extra headers, for example the OpenRouter attribution headers:

const bot = new Chatbot(OPENROUTER_API_KEY, 'openrouter', null, {
  headers: { 'HTTP-Referer': 'https://my-app.example', 'X-OpenRouter-Title': 'My app' },
});

Local models: Ollama and LM Studio

No key is needed; pick any model you pulled:

const local = new Chatbot(null, 'ollama', null, { model: 'qwen3' });

console.log(await local.listModels());   // the models the server has

const input = new OpenAICompatibleInput('You are a helpful assistant.');
input.addUserMessage('Give me three names for a coffee shop.');

for await (const chunk of local.stream(input)) process.stdout.write(chunk);

listModels() is available for every compatible provider.

Your own endpoint

const custom = new Chatbot(MY_API_KEY, 'openai_compatible', null, {
  baseUrl: 'https://my-host/v1',
  model: 'my-model',
});

Everything else works the same

Streaming, the tool loop, structured output, the retries & timeouts and every Gen function accept these providers:

const { Gen } = require('intellinode');

const code = await Gen.generate_component('a pricing card', null, 'ollama', { model: 'qwen3' });
const data = await Gen.generate_json('Where is Rome?', schema, DEEPSEEK_API_KEY, 'deepseek');

A ChatGPTInput sent to a compatible provider is converted to a chat-completions request, so existing code can switch providers by changing the chatbot only.

Embeddings

The embedding controller accepts the compatible providers that serve an embeddings endpoint (openrouter, together, ollama, lmstudio, or openai_compatible with a baseUrl):

const { RemoteEmbedModel, EmbedInput } = require('intellinode');

const embed = new RemoteEmbedModel(null, 'ollama');
const vectors = await embed.getEmbeddings(new EmbedInput({ texts: ['hello world'], model: 'nomic-embed-text' }));