LLMDiscordBridge

A simple app that exposes an OpenAI-like LLM API to be used to drive a Discord bot.

This is primarily designed to interface with llama-server from the ggml-org/llama.cpp project.

Usage

Run python discord_bridge.py to run the bridge. An optional -s <file> parameter can be used to specify a settings file to load besides the default settings.json.

The bot can be configured to listen for messages on specific channels of a server and reply to them.

Calling $register @Bot (where @Bot is the bot's Discord name) will tell it what channel to listen and reply to.

Bots will ignore commands for other bots.

Bot Commands

• $help @Bot Used to list out these commands
• $ping @Bot Used to get the bot to reply "Pong!" to test that it can read and write on a channel
• $register @Bot Used to make the bot listen and reply to messages on a channel
• $remove @Bot Remove the bot from a channel so it doesn't reply to any messages
• $reset @Bot Clear the bot's memory of that channel's conversation
• $speak @Bot Used to make the bot say something to initiate conversations
• $whisper @Bot Used to make the bot say something without remembering it

Setup

Discord Bot

• Set up a Discord bot if you don't have one already
• Make sure the Message Content Intent checkbox is set to true
• Add it to the server of your choice, make sure it has message read and write permissions

Llama Server

Llama Server needs to be started with a command similar to the following:

llama-server.exe -ngl 9999 --host 0.0.0.0 -dev CUDA0 --jinja -c 42000 -m Qwen3-30B-A3B-Thinking-2507-Q4_K_S.gguf

First Time

• Activate your Python environment if needed
• Run python -m pip install -U discord.py to install the Discord library
• Run python discord_bridge.py to run the bridge
• You will be asked to fill in some info
  • Token: the token of the bot, used to authenticate with Discord
  • URL: the URL to the LLM's OpenAI-like API, default http://localhost:8080 (no slash at the end)
  • Prompt: the system prompt to configure how your bot will act
  • LLM API Token: the token of the LLM's API, used to authenticate with OpenRouter or other API providers (optional)
• These settings will be saved to settings.json if you need to change it later

Manual settings.json

The settings.json can be manually updated for advanced config.

• token: Discord bot token
• url: LLM API endpoint URL (default: http://localhost:8080, gets /v1/chat/completions added to it internally)
• prompt: System prompt for the LLM, may include the contents of other files using the syntax ${filename.ext}
  • e.g. You are the character as defined below:\n${character_prompt.txt}
• channels: List of registered channel IDs (automatically managed)
• apiParams: Dictionary of LLM API parameters to override defaults (optional)
  • e.g. "temperature": 0.6
  • e.g. "top_p": 0.9
  • e.g. "model": "deepseek/deepseek-chat-v3-0324" (used for OpenRouter.ai)
• llmToken: LLM's API authentication token (optional)
• multiIO: Bool to enable the multi I/O system (default: false, optional)
• multimodal: Bool to enable multimodal image input support (default: false, optional)
• tools: tool configs that the bot is allowed to use (not implmented)

TODO

• Add chat history memory