AIbitat Framework

The web-browsing-plugin allows agents to search the internet via Google (using Serper.dev) or scrape and summarize specific website content.

To use this plugin, register it with your AIbitat instance using the experimental_webBrowsing function. Once registered, the agent can call a tool named web-browsing which accepts either a query (for searching) or a url (for scraping).

Required Environment Variables:

• SERPER_API_KEY: Required for Google searches via serper.dev.
• BROWSERLESS_TOKEN: Required for scraping website content via browserless.io.

Tool Parameters:

• query (string): A search query to perform on Google.
• url (string, URI format): A specific URL to scrape.

Note: If the scraped content exceeds 8,000 characters, the plugin automatically summarizes it using OpenAI's gpt-3.5-turbo-16k-0613 model.

import { experimental_webBrowsing } from 'aibitat/plugins/web-browsing';

const aibitat = new AIbitat(config);
aibitat.use(experimental_webBrowsing());

To start a conversation between multiple agents, instantiate AIbitat, define agents with specific roles, group them into a channel, and call .start().

In the example below, a client (human-like) asks a question, a mathematician solves it, and a reviewer confirms the result.

import {AIbitat} from 'aibitat'
import {cli} from 'aibitat/plugins'

const aibitat = new AIbitat()
  .use(cli())
  .agent('client', {
    interrupt: 'ALWAYS',
    role: 'You are a human assistant. Reply "TERMINATE" when there is a correct answer or there`s no answer to the question.',
  })
  .agent('mathematician', {
    role: `You are a Mathematician and only solve math problems from @client`,
  })
  .agent('reviewer', {
    role: `You are a Peer-Reviewer and you do not solve math problems. 
    Check the result from @mathematician and then confirm. Just confirm, no talk.`,
  })
  .channel('management', ['mathematician', 'reviewer', 'client'])

await aibitat.start({
  from: 'client',
  to: 'management',
  content: 'How much is 2 + 2?',
})

console.log(aibitat.chats)

Functions allow agents to execute code and return results back into the conversation.

To use a function:

1. Define the function using .function(config).
2. Add the function's name to the functions array in the target agent's configuration.

Function Configuration Object:

• name: Unique identifier for the function.
• description: A description used to help the LLM understand when to call the function.
• parameters: A JSON Schema object defining the input arguments.
• async handler(): The implementation of the function that returns the result to the conversation.

const aibitat = new AIbitat()
  .agent('coder', {functions: ['doSomething']})
  .function({
    name: 'doSomething',
    description: 'Let me do something for you.',
    parameters: {
      type: 'object',
      properties: {},
    },
    async handler() {
      return 'Result of execution'
    },
  })

const aibitat = new AIbitat()
  .agent('...', {functions: ['doSomething']})
  .function({
    name: 'doSomething',
    description: 'Let me do something for you.',
    parameters: {
      type: 'object',
      properties: {},
    },
    async handler() {
      return '...'
    },
  })

Plugins allow you to add new features or integrations to AIbitat. You can use built-in plugins or implement your own using the AIbitatPlugin interface.

Available Plugins:

• cli: Adds a CLI interaction mode to the chat.
• fileHistory: Saves chat history to a JSON file (defaults to the history folder).
• experimental_webBrowsing: Enables agents to search and navigate the internet (experimental).

Implementing a Custom Plugin:

import {AIbitatPlugin} from 'aibitat'

export function myPlugin(): AIbitatPlugin {
  return {
    name: 'my-plugin',
    setup(aibitat) {
      console.log(`setting up my plugin`)
      aibitat.onMessage(({from, to, content}) => {
        console.log(`${from}: ${content}`)
      })
    },
  }
}

// Usage
const aibitat = new AIbitat().use(myPlugin())

import {AIbitatPlugin} from 'aibitat'

export function myPlugin(): AIbitatPlugin {
  return {
    name: 'my-plugin',
    setup(aibitat) {
      console.log(`setting up my plugin`)

      aibitat.onMessage(({from, to, content}) => {
        console.log(`${from}: ${content}`)
      })
    },
  }
}

// To use:
// const aibitat = new AIbitat().use(myPlugin())