Llaminate is a simple but powerful library designed to abstract-away differences between chat completion APIs across LLM providers.

It's robust at managing prompts, message histories, token usage and integrating tools. Ideal for quickly building applications to interact with LLM services, that may need to switch between them.

npm i llaminate

import { Llaminate } from "llaminate";

Basic usage

const llaminate = new Llaminate({
  endpoint: Llaminate.MISTRAL,
  key: "12345-abcde-67890-fghij-klm",
  model: "mistral-small-latest",
  rpm: 720 // requests per minute (rater limiter)
});

const completion = await llaminate.complete("Hello, AI!");
console.log(completion.message);

Streaming is made easy:

const stream = await llaminate.stream(
  "How much wood would a woodchuck chuck if a woodchuck could chuck wood?"
);

for await (const sum of stream) {
  console.log(sum);
}

Llaminate is current an alpha release, but it is tested against:

• Llaminate.ANTHROPIC
• Llaminate.DEEPSEEK
• Llaminate.GOOGLE
• Llaminate.MISTRAL
• Llaminate.OPENAI

(Or you can use the URL to your provider.)

More features

Llaminate can handle tools, chat history, structured output, images and documents, tracking your usage, and more:

const system = ["You are a sarcastic assistant who answers very briefly and bluntly."];

const attachments = [{
  type: Llaminate.JPEG,
  url: "https://live.staticflickr.com/7194/6964010157_9af8648f53.jpg"
}];

const schema = {
  type: "object",
  properties: {
    reply: {
      type: "string",
      description: "Your response to the user's query."
    },
    thoughts: {
      type: "string",
      description: "Your internal thoughts about the user's query."
    }
  },
  required: ["reply", "thoughts"]
};

const snarky = await mistral.complete("What do you see in this image?", {
  system, attachments, schema
});

console.log(snarky.message.reply);
console.log(snarky.message.thoughts);
console.log(snarky.tokens.total);

Documentation

Full documentation is available on the GitHub pages website for the probject.

This project is licensed under the MIT License. See the LICENSE file for details.