tyster.io

installation

how to get your own moemoe up and running

this page is for those interested in running their own installation of moemoe.

prerequisites

before diving in, ensure you have the following:

  • node.js 18+ and npm installed
  • git for cloning the repository
  • Discord bot token from the Discord developer portal
  • AI backend with an API

AI backend

moemoe requires some sort of AI backend to function and you've plenty of options available:

running a local LLM gives you full control and no API costs, but does require a somewhat beefy PC depending on the model's size. i would recommend koboldcpp for how awesome it is.

a mid-tier gaming PC may be able to run 7-13B quantizized models. larger models will require more resources.

as of writing, moemoe is running on Mistral-Small-24b-Instruct-2501. bear in mind, smaller models will struggle with remembering rules and function syntax.

quick KoboldCPP setup:

  1. download KoboldCPP
  2. grab a model that will fit on your PC
  3. run KoboldCPP and take note of the URL
    • by default this will be localhost:5001

cloud providers

openai and other provider support is coming soon. for now, local models are the way to go.

bot installation

clone the repository

git clone https://github.com/conjureme/moemoe-discord.git
cd moemoe-discord

install dependencies

npm install

this will install all required packages including discord.js

configure environment

copy the example environment file:

cp .env.example .env

open .env and add your variables:

DISCORD_TOKEN=your_bot_token_here
CLIENT_ID=client_id_here
API_PROVIDER=koboldcpp

configure AI settings

edit config/ai.json and adjust sampler settings to your liking:

{
  "maxTokens": 500,
  "temperature": 1.0
}

currently a very unpleasant method of adjusting settings—a UI is in the works.

customize moemoe's personality

edit config/bot.json to define moemoe's personality or make your own persona:

{
  "name": "moemoe",
  "description": "a friendly and helpful discord bot",
  "data": {
    "system_prompt": "you are moemoe, a helpful discord bot...",
    "first_mes": "hiya! i'm moemoe!",
    "mes_example": "example dialogue here"
  }
}

this entire file is structured as a v2 Tavern Character Card. you can copy and paste characters if you prefer.

deploy commands

before first run, deploy the slash commands:

npm run deploy

start the bot

npm run dev

inviting moemoe

once running, you'll need to invite moemoe to your server:

  1. go to the discord developer portal
  2. select your application
  3. navigate to OAuth2 > URL Generator
  4. select these scopes:
    • bot
    • applications.commands
  5. select perms you want or ensure bot has proper roles
  6. copy the generated URL and open it to invite moemoe

bot in your server, try these:

  • try /ping to verify bot and commands are working
  • test AI interaction with @moemoe hello!
  • test AI responses to ensure it's acting correctly

troubleshooting

moemoe won't respond

  • check that your AI backend is running and accessible
  • verify the APIs in .env are correct
  • ensure moemoe has permission to read and send messages
  • enable debug mode and check terminal output for errors

commands aren't showing

  • run npm run deploy or npm run deploy:global again
  • refresh discord with CTRL + R and retry
  • try deploying to your specific guild instead of globally

AI responses are weird

  • ensure config/ai.json is using the right instruction format and sampler settings
  • try lowering your temperature

voice support

this is another WIP. voice functionality is currently hardcoded, but you can implement it by setting up and following the following steps.

for voice features, you'll also need:

whisper transcription

koboldcpp supports running a whisper model in tandem, you can get one from here:

next steps

now that moemoe is running:


always remember to keep your bot token and API keys secret! never commit them to version control or hand them out.