GladdBot

Curated Twitch chatbot for Gladd using Gemini. Feel free to fork this repo and use it for your own channel. Minimal setup instructions are below.

Setup

Node.js >=v20.6.0 is required to run this. Rename the .env.example file to .env. Next, you'll need 2-3 things:

A Google AI API Key
A service Twitch account
A database (optional)

API Key

Go to aistudio.google.com, click Get API Key on the left, and follow the instructions from there. Store the API key in the .env file.

Twitch Account

Warning

I would recommended you use a separate Twitch account to run this; however, if you do want to user yours, then the steps are the same.

Go to the Twitch Developer Console
1. If you have an existing application:
  1. Add https://twitchtokengenerator.com as a redirect url
2. If you don't have an application, create a new one:
  1. Use whatever name
  2. Use https://twitchtokengenerator.com for the redirect url
  3. Select Chat Bot for the category
  4. Select Confidential for the client type
3. Paste the client ID and client secret into the .env file
Create a new Twitch account or use an existing one
Go to twitchtokengenerator.com
1. Select Bot Chat Token
2. Fill in the Client Secret and Client ID fields.
3. Scroll down and click Generate Token. The Client ID under the Generated Tokens section should match the one you provided.
4. Store the access token and refresh token somewhere temporary.
Use this site to get the user ID of the account you just created. Store this in the .env file.

Database

There are two ways to store token data: locally in a JSON file or a database. If you choose a JSON file, the file needs to look like this:

{
  "accessToken": "",
  "refreshToken": "",
  "scopes": ["chat:edit", "chat:read"],
  "expiresIn": 0,
  "obtainmentTimestamp": 0
}

You'll need to read from this file and supply the data to auth.addUser and write to it in the auth.onRefresh event.

Replace the existing code in the auth region with this:

const tokenData = await fs.readFile("path_to_json_file", "utf-8");

const auth = new RefreshingAuthProvider({
    clientId: process.env.TWITCH_CLIENT_ID!,
    clientSecret: process.env.TWITCH_CLIENT_SECRET!,
});

auth.onRefresh(async (_, data) => {
  await fs.writeFile("path_to_json_file", JSON.stringify(data))
})

auth.addUser(process.env.TWITCH_USER_ID!, JSON.parse(tokenData), ["chat"]);

If you want a more secure option, you can use a database, which is what this project uses. Your table should look similar to this (example uses PostgreSQL):

CREATE TABLE tokens (
  id SERIAL CONSTRAINT PRIMARY KEY,
  access_token VARCHAR(100),
  refresh_token VARCHAR(100),
  expires_in INT,
  obtainment_timestamp BIGINT
);

Once you have either of these options set up, fill in the access token and refresh token that you saved earlier.

Customizing

System instructions

In order to customize the bot's personality and responses, edit the instructions.txt file to your liking. It's better to provide more information in order to get more accurate responses. You can use the current file as a reference. Further reading.

Safety settings

Gemini's safety settings are for the AI's responses as opposed to what the user provides. They default to BLOCK_MEDIUM_AND_ABOVE. Obviously, set them appropriately to fit your channel's atmosphere. In most cases, the default settings will suffice, though some perfectly normal messages might get blocked. Play around with it until you find a happy medium. Further reading.

Character limit

The character limit for a Twitch message is 500, so you should set MAX_OUTPUT_LENGTH to something under that. It is possible to set a higher limit, and the bot will send parts of the full response in separate messages, but I don't recommend doing this as it can flood chat quickly.

kablekompany/gladdbot