Wisty.js is a JavaScript natural language processing library to build contextual chatbots or virtual assistants on the web.

• Powered by machine learning thanks to TensorFlow.js.

• Supports training and inference in Node.js and most browsers : you can integrate a bot into your front-end without having to rely on some infrastructure or a cloud service.

• It reduces the amount of hardcoded logic you have to write : new features are just a few training dialogs away.

Status

The project is in alpha : the API is unstable as new feature keeps being added.

Installation

$ npm install wisty

import * as wisty from 'wisty';

wisty.tools.parseStories(...);

If you intend to run Wisty on Node.js, then the TensorFlow C Backend can speeds up your bot :

$ npm install @tensorflow/tfjs-node

const tf = require('@tensorflow/tfjs-node');
const wisty = require('wisty');

You can also use CDNs :

<script src="https://cdn.jsdelivr.net/npm/@tensorflow/tfjs"></script>
<!--
    If you intend to use Universal Sentence Encoder model :
    <script src="https://cdn.jsdelivr.net/npm/@tensorflow-models/universal-sentence-encoder"></script>

    If you attend to parse WistyML Training files :
    <script src="commonmark.min.js"></script>
    (see https://github.com/commonmark/commonmark.js)
-->

<script src="https://cdn.jsdelivr.net/npm/[email protected]/dist/index.umd.min.js"></script>

As the API is currently unstable, it is recommended to use a fixed version.

Demos

You can find open source demos of chatbots running with Wisty.js in the demos folder.

• Passwords : An assistant trying to help you identify yourself on a web site if you have troubles logging in.

  • The implementation is quick and dirty JavaScript in the HTML and a customized style for BotUI.

• Restaurant : A chatbot taking orders for a restaurant.

  • Implemented from scratch with Vue.js 3 and Vite.

Documentation

For the moment, only the API reference is available on GitHub Pages.

License

Wisty.js is licensed under the MIT license.