discord-chatgpt-v2

A Discord bot allowing users to interact with large-language models from OpenAI, DeepSeek, or Google.

This is a complete rewrite of my original discord-chatgpt bot, updated with better code quality, privacy, and features like split message responses, multiple model providers, and better conversational context.

Running the bot

1. Download and install Node.js if you don't have it
2. Clone (or download and unzip) the repository and cd into it with your terminal
3. Run npm install
4. Rename config-.json to config.json
   • This prevents your config from being overwritten should you update your bot.
5. Create a new Discord application
   1. Set its name, description (about me), and picture as you see fit
   2. Copy the Application ID and paste it in the credentials.discord_application_id config field
   3. Go to the "Bot" tab and create a new bot if it's not created already
   4. Copy the bot token and paste it in the credentials.discord_bot_token config field
   5. Scroll down and make sure "Message content intent" is enabled
6. Set your Discord user ID in the bot.owner_id config field. Get this by turning on developer mode in settings and right-clicking on your profile.
7. Configure an API key for your AI provider of choice:
   • OpenAI: Get your API key and paste it into the credentials.openai_secret config field
   • DeepSeek: Get your API key and paste it into the credentials.deepseek_secret config field
   • Google: Get your API key and paste it into the credentials.google_secret config field
   • Note: Using these APIs isn't free. See their respective pricing resources for details.
8. Configure your AI provider and chosen model by updating the gpt.provider and gpt.model config fields:
   • OpenAI: Set provider to openai and model according to their models reference
     Recommended model(s): gpt-4o-mini, gpt-4o
   • DeepSeek: Set provider to deepseek and model according to their models reference
     Recommended model(s): deepseek-chat
   • Google: Set provider to google and model according to their models reference
     Recommended model(s): gemini-2.0-flash, gemini-2.0-flash-lite
9. Make any other optional changes to the config file, then save it.
10. Register the bot's slash commands and Apps menu items by running node registerCommands.js
11. Start the bot with node bot.js
    • Pro tip: Install PM2 and run the bot with pm2 start.
12. Once the bot logs in, an invite URL will be logged. Open it and follow the instructions to add the bot to your server.
13. Try it out by DMing or pinging the bot!

Configuration

The bot can be configured by editing the config.json file, as you did during setup. All config options are as follows:

• object credentials: Contains authentication settings
  • string openai_secret: Your OpenAI API key
  • string deepseek_secret: Your Deepseek API key
  • string google_secret: Your Google AI Studio API key
  • string discord_bot_token: Your Discord bot's token
  • string discord_application_id: Your Discord application/client ID
• object gpt: Contains language model settings
  • string provider: One of openai, deepseek, or google
  • string model: A model from your provider. See the guide above for reference.
  • number temperature: The model's temperature value, ranging from 0 to 2.
  • boolean should_stream: Whether or not the response should be streamed. If true, responses will be faster as they will be processed and sent by the bot while still being generated. This may not be supported with all models.
  • object[] messages: An array of message objects to be inserted at the beginning of every API request. Note that a system message containing the date and time, bot name, and other basic instructions is added automatically, placed after this set of messages.
    • string role: Set to system, assistant, or user. system messages can be used to influence the model's behavior and give it information, assistant messages are those sent by the model, and user messages are those sent by the user.
    • string content: The message's text content
  • number context_msg_count_max: The maximum number of messages above the prompt message to use as context.
  • number context_msg_count_min: The minimum number of messages above the prompt message to use as context, regardless of token usage.
  • number context_tokens_max: The minimum number of input tokens that context should use.
  • object text_files: Contains settings related to user-sent text files
    • boolean enabled: Whether or not user-sent text files should be read and processed by the model.
    • string[] extensions: An array of attachment file extensions to treat as text files.
    • number max_bytes: The max size that text files can be in order to be processed.
  • object vision: Contains settings for Vision - supported by most openai and google models
    • boolean enabled: Set to true to allow supported models to process images.
    • string[] extensions: An array of image file extensions to allow the model to view.
    • number max_bytes: The max size that image files can be in order to be processed.
    • boolean low_resolution: Set to true to use low detail mode.
    • number tokens_base: The number of tokens used by every image regardless of resolution. Update this value to match your model's specifications.
    • number tokens_per_tile: The number of tokens used by each tile_sizex tile of a high resolution image (after resizing). This doesn't apply when low_resolution is enabled. Update this value to match your model's specifications.
    • object resize: Contains short and long side dimensions to calculate final high-res image dimensions. Resizing happens remotely.
      • number short_side: The short side length
      • number long_side: The long side length
    • number tile_size: The length of one side of a high-res image tile.
  • object audio: Contains settings for Audio understanding - supported only by google models
    • boolean enabled: Set to true to allow supported models to process audio files.
    • string[] extensions: An array of audio file extensions to allow the model to view.
    • number max_bytes: The max size that audio files can be in order to be processed.
  • object replacements: Contains key-value pairs for strings to replace with other strings in model responses, where the key is the target string and the value is the replacement string. By default, this is used to replace common LaTeX control strings with their respective Unicode characters.
  • boolean ignore_bots: Whether or not bot messages (not from this bot) should be excluded from context provided to the model.
• object bot: Contains settings for the Discord bot
  • string owner_id: The Discord user ID of the bot maintainer (you, most likely). Only this user can use admin commands like /users.
  • object status: Contains settings for the bot's activity status
    • string type: Set to Playing, Watching, or Listening, determines the part in bold at the beginning of the status. Set to Custom to remove the prefix and use text to set the entire status.
    • string text: The text following the activity type. {messages_month} is replaced with the number of messages sent to the bot this month, and {messages_total} is replaced with the number of messages sent to the bot in total.
  • boolean split_responses: Determines whether or not model responses are split and sent by paragraph. When this is false, the model's response will be sent as a single message instead of several smaller messages. Responses will still be split if they exceed Discord's character limit.
  • number response_part_min_delay: The minimum number of milliseconds of delay that should exist between sending message parts. This will not impact the speed at which the response is generated, only how fast it's sent. Low numbers for this option might lead to the bot hitting rate limits, causing uneven and extended delays.
  • string|number embed_color: The accent color to use on slash command embeds. Should be in hexadecimal format, including the leading #, or a decimal number.
  • number interaction_max_age_hours: The maximum age (in hours) of saved interaction JSON files before automatic deletion
• object messages: Contains settings for every user-facing message sent by the bot. These aren't be listed here. Use each key's name and existing value to guestimate its purpose.

User access control

As the owner, you're always allowed to use the bot, but other users can be allowed or blocked by you by using the /users commands or by editing the access.json file.

access.json contains the following properties:

• boolean public_usage: When set to true, anyone except explicitly blocked users can use the bot. When set to false, only users explicitly allowed can use the bot.
• object users: Contains a property for each user ID that has used the bot so far. When a user invokes the bot, they are automatically added to this list with a value of 0.
  • number <user_id>: This value can have one of three states: 1 explicitly allows the user, 0 means the user follows the public_usage setting, and -1 explicitly blocks the user.

Edits made to access.json are applied immediately. Ensure that the file is properly JSON-formatted before saving to avoid errors and crashes.

The /users commands are as follows:

• /users allow <user>: Grants explicit usage access to a user.
• /users block <user>: Explicitly blocks a user.
• /users unlist <user>: "Unlists" a user so their access is determined by access.public_usage.
• /users list: Lists users sorted into allowed, blocked, and unlisted categories.

Context management

Sometimes language models can produce unexpected or unwanted results. If this happens during a conversation, use the /contextbarrier command to ignore all previous messages. Replying to a message before the barrier will still include it as context, but otherwise context barriers give you a fresh start.

Saving/removing responses

Interactions are saved as JSON files in the interactions folder for the duration of time set in config.bot.interaction_max_age_hours.

If the model's output isn't formatted correctly in Discord and you want to remove it or view it as raw text, or to get the raw text output for any other reason, right click on any sub-message produced by the bot and enter the Apps menu for download and removal options.

• Delete response message: Deletes the immediate response sub-message
• Get response text: Replies with a text file containing the complete response as raw text/markdown

The latter option only works if the interaction JSON file is still available (hasn't been deleted).