This is a JavaScript implementation of the MegaHAL algorithm, more specifically a direct, nearly one-to-one port of kranzky/megahal. Major credits to kranzky, without whom this wouldn't exist.

It is not currently usable as a CLI application, but you can embed it into your own Node.js code.

TypeScript types are provided within the package.

Training a personality is rather slow and may take a few seconds to complete. Anyone is more than welcome to suggest any performance improvements, I am sure many can be made.

A Sooth stochastic predictive model is also implemented almost one-to-one and may later be extracted into a separate package.

Both are licensed under the Unlicense license, in the spirit of the original packages. So basically, do whatever you want with it, but no warranty!

Installation

Use your favorite package manager:

pnpm install megahal.js

yarn add megahal.js

npm install megahal.js

Usage

To start you must import the main class from the module.

import MegaHAL from 'megahal.js'

Using MegaHAL

Create a new instance of MegaHAL, and provide the personality you would like to use, using the name.

If you omit the personality name, default will be used.

Use .reply(line) to get a response for a a given line of text, and also learn from that line in the process.

You can use .train(filename) or .train(lines) to manually train the bot on the go.

import { MegaHAL } from 'megahal.js'

const bot = new MegaHAL('default')

const input = 'hello!'
const message = bot.reply(input)

Loading personalities

MegaHAL comes with several personalities pre-loaded. You can list them using MegaHAL.list().

To add a custom personality, use the addPersonality method.

To switch personalities on an existing instance, use the become method.

// list all personalities

console.log(MegaHAL.list())

// load a custom personality
MegaHAL.addPersonality('name', [
  'line 1',
  'line 2',
  // ...
])

// switch personality on-the-fly
import { MegaHAL } from 'megahal.js'
const bot = new MegaHAL('default')
bot.become('sherlock')

Continuous learning

You can save & load your continuous learning to a file, at your discretion so do it as often as you feel appropriate.

import { MegaHAL } from 'megahal.js'

const bot = new MegaHAL('default')

// save bot:
await bot.save('/path/to/file.zip')

// load bot at a later time:
await bot.load('/path/to/file.zip')