LLMCord - An LLM client as a discord bot

This bot allows you to use LLMs such as ChatGPT, LLaMA3, and Gemini 1.5 from Discord.

You are also able to use models such as LLama 3.2 Vision however image generation is not supported (yet!)

Installation

Using pre-created executables

Download the executable for the latest release from Releases

NOTE: Currently MacOS and Linux do not have pre built binaries

Manual

1. Clone this repo or download the source code from one of the Releases

$ git clone https://github.com/grqphical/LLMCord.git

2. Create a virtual environment

$ python -m venv venv

3. Activate the virtual environment

$ ./venv/Scripts/activate

NOTE On linux the path will be ./venv/bin/activate

4. Install dependencies

$ pip install -r requirements.txt

Viewing detailed logs

If you want to view more detailed logs in order to debug issues, in your .ENV file add the line APP_ENV=dev

Converting to binary

If you want to convert the script into a binary run pyinstaller llmcord.spec

Usage

1. Create a new Discord Application at the Discord Developer Portal and configure it to be a bot. (Guide Here)
2. Create a .env file and add your bot's token as DISCORD_TOKEN
3. Create a file called llmcord.toml and add this to it:

system_prompt = "Use markdown if neccessary."
default_model = "llama3-70b"

[models.llama3-70b]
base_url = "https://api.groq.com/openai/"
model = "llama3-70b-8192"
token = "API_TOKEN_HERE"
client = "OpenAIClient"

NOTE I am using Groq AI here as an example so I have selected the OpenAIClient due to it's API being compatible with OpenAI however LLMCord also includes a Google Gemini 1.5 client that you can use as GeminiClient

Explanation of config

system_prompt - Allows you to control the system prompt of the AI

default_model - Specifies the default model REQUIRED

For each model you wish to make create it as a seperate table with a display name. In this case I chose llama3-70b. base_url - Base URL for the API REQUIRED

model - Which model to use. Only letters, numbers, underscores and dashses are valid characters here REQUIRED

token - API token for API, REQUIRED IF NEEDED BY API

client - Which Client to use to send the request. The two included clients are OpenAIClient and GeminiClient. REQUIRED

Then finally run

$ python -m llmcord

Now you should be able to use LLMCord in any servers you have invited it to.

Contributing

See CONTRIBUTING.md

Plugins/Custom Clients

You are able to make your own API clients by creating a python script in the plugins/ folder. Simply create a class that inherits from plugins.BaseClient and implement the get_response method. Make sure it is a coroutine. In order to access the API use the aiohttp library as it is already installed and used by LLMCord.

Look in the plugins folder for examples of custom clients

Changelog

1.4.0

• Added file upload support for models with vision capabilities

• Moved away from cogs API, will rework later on

• Updated discord.py

• Added the ability to see more detailed log if you set APP_ENV equal to 'dev'

1.3.1

• Changed license from MIT to Mozilla Public License 2.0

• Added license to about section of bot

• Bump aiohttp from 3.9.5 to 3.10.2 by @dependabot in #3

• Added PyInstaller spec

1.3.0

• Added compare command that allows you to send a response to two LLM's and compare their responses

• Added about command

• Improved Documentation

• Added build scripts

1.2.3

• Refactored code to work with Pyinstaller

• Pre built executables are now able to be distrubuted

1.2.2

• Moved commands into their own Cogs

• Added unit tests for Config and Context

• Fixed linting errors

• Added CI for unit tests

1.2.1

• Fixed error handling with certain edge cases when loading the configuration

• Added error handling when selecting an invalid/undefined model. Before it would just get stuck in an infinite loop

• Improved documentation

1.2.0

• Added custom logger

1.1.0

• Added support for custom clients

• Included OpenAIClient and GeminiClient by default

• Changed /list to now show every model on a different line

Running the test suite

Make sure to install requirements.txt then run pytest to run the tests.

Roadmap

I may or may not add these features later on.

• Custom API Handlers (ability to define your own plugins to work with non-OpenAI compatible APIs)

• File uploads

• Integration with non LLM's (eg. DALL-E for image generation)

License

LLMCord is licensed under the Mozilla Public License 2.0