Skip to content

wladpaiva/aibitat

Repository files navigation

This project is a fork from the original autogen but done in TypeScript.

AIbitat - Multi-Agent Conversation Framework

AIbitat is a stateless & extensible framework designed to enable interaction between multiple agents while allowing human participation.

Think of it as Slack but for AI agents. Create agents, add them to channels and let them chat with each other. You can also add intractable agents to channels and let them participate in the conversation.

Usage

For a more complete example, check out the examples folder.

You can install the package:

# to install bun go to https://bun.sh and follow the instructions
# then run to start a new project:
mkdir my-project
cd my-project
bun init

# then install aibitat
bun install aibitat

Create an .env file and add your OPENAI_API_KEY:

OPENAI_API_KEY=...

Then create a file called index.ts and add the following:

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'])

// aibitat.onMessage(console.log)

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

console.log(aibitat.chats)

Then run:

bun run index.ts

Roadmap

Feature What it does Status
Direct messages Agents talk directly with each other.
Feedback Keep chat alive until some agent interrupts the chat.
Channels Agents talk with many others, like in a Slack channel.
Error handling Manage rate limits smoothly without crashing.
Function execution Agents can execute tasks and understand the results.
Web browsing Navigate on the internet. 🚧
Cache Save chat history for faster and fewer API calls. 🕝
File interaction Interact with local files by read/write/execute 🕝
Code execution Agents can run code and share the results. 🕝
Cost limit Limit the number of interactions by cost. 🕝

Providers

Provider Status
OpenAI
Anthropic
Cohere 🕝
Fireworks.ai 🕝
Hugging Face 🕝
Replicate 🕝

Documentation

Some terms used in this documentation:

  • interrupt: NEVER, ALWAYS. When NEVER, the agent will never interrupt the conversation. When ALWAYS, the agent will always interrupt the conversation. (Note: any of them can interrupt the conversation if they reply "INTERRUPT" or terminate the chat when replying "TERMINATE".)
  • role: The role of the agent. It is used to describe the role the agent will perform in the chat.
  • maxRounds: The maximum number of chats an agent or a group will reply to the conversation. It is used to prevent loops.

new AIbitat(config)

Creates a new AIbitat instance. The config object can be used to configure the instance. By default, aibitat uses OpenAI and GPT-3.5-TURBO as the provider for the conversation and GPT-4 for predicting the next agent to speak. You can change it by passing provider and model to the AIbitat constructor or by setting them on the specific agent/channel config. Default config:

{
  providers: 'openai',
  model: 'gpt-3.5-turbo',
  interrupt: 'NEVER',
  maxRounds: 100,
  chats: [
    // {
    //   from: '🧑',
    //   to: '🤖',
    //   content: `Talk about something`,
    //   state: 'success',
    // },
  ],
}

.agent(name, config)

Creates a new agent. The config object can be used to configure the agent. By default, agents use the interrupt from the AIbitat config.

.channel(name, agents, config)

Creates a new channel. The config object can be used to configure the channel. By default, maxRounds is set to 100.

.function(config)

Functions are used to execute code and return the result to the conversation. To use it, call the function method passing the definitions for the function and add its name to the functions property to the node config:

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

The results will then be sent to the provider and returned to the conversation.

.use(plugin)

Plugins are used to extend the functionality of AIbitat. They can be used to add new features or to integrate with other services. You can create your own plugin by implementing the AIbitatPlugin interface.

To use a plugin, call the use method:

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

const aibitat = new AIbitat().use(cli())

You can also create your own plugin by implementing the AIbitatPlugin interface:

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}`)
      })
    },
  }
}

List of available plugins:

  • cli: Adds a CLI interaction to the chat.
  • fileHistory: Saves the chat history to a JSON file. Defaults to history folder
  • experimental_webBrowsing: Adds a web-browsing function to the chat that enable agents to search and navigate on the internet. NOTE: this plugin is experimental and may not work as expected.

Life cycle

AIbitat supports Life Cycle events, which trigger at specific moments. The following diagram shows the life cycle of a conversation:

Life cycle

The life cycle events in AIbitat consists of:

  • start: Called when the chat starts.
  • error: Called when there's a known error (see src/error.ts). To retry, call .retry().
  • message: Called when a message is added to the chat history.
  • terminate: Called when the conversation is terminated. Generally means there is nothing else to do and a new conversation should be started.
  • interrupt: Called when the conversation is interrupted by an agent. Generally means the agent has a question or needs help. The conversation can be resumed by calling .continue(feedback).

Contributing

Copy the .env.example file to .env and add your keys/tokens in there.

To install dependencies:

bun install

To run:

bun run examples/beginner-chat.ts

Check out the examples folder for more examples.

This project was created using bun init in bun v1.0.3. Bun is a fast all-in-one JavaScript runtime.

Sponsors

License

MIT