This bot allows users to query GPT-3 completions inside a Twitch chat. Simply type !ai <text-to-complete> and the bot will respond with a completion generated by GPT-3.
You can either run pre-built latest executable for windows or run from source code or build your own executable.
- A Twitch account
- An OpenAI API key (register at their website and go to account -> api keys, read disclaimer down below for more info)
Clone this repository to your local machine:
git clone https://github.com/vodkabydlo/GPT-Twitch-Chatbot.git
Then get to the directory the bot is in and install all dependencies:
npm install
To run the app from terminal
npm run start
to build a standalone .exe
npx electron-builder build --win portable
One token is approximately 4 characters in english language. All other languages are more expensive and therefore with a set amount of tokens replies in other languages will be shorter.
Set amount of tokens in the app to set maximum amount of tokens to be spent on each completion. I found it perfect between 50 and 200 tokens. Less tokens you spend the dumber and shorter responses will be, so for clever responses use 150-200.
Twitch messages are limited in characters (approx. 500). The bot will split messages if OpenAI's reply is too long and split
To query a GPT-3 completion, simply type !ai <prompt> in the Twitch chat. The bot will respond with a completion generated by GPT-3.
For example, if you type !ai The quick brown fox jumps over the lazy dog, the bot might respond with
The quick brown fox jumps over the lazy dog, but then it turns around and jumps back again just for fun.
- This bot uses the OpenAI GPT-3 API, which may generate offensive or inappropriate completions. Use at your own risk.
- Also keep in mind that the bot is using commercial OpenAI API. Initially as of December 2022 OpenAI give you $18 of free credit and several months to test it. To use it further you will need to pay for GPT.
- Since bot is using tmi.js, the response of GPT will be from your name.