Maker Bogans' Grumpy Bot aka 'Baz'

Design goals

This is a Discord bot for the aussie Maker Bogans Discord server. If you don't know what any of that is, then you should go somewhere else.

Baz is intended support question answering over the discussion on the Maker Bogan's discord. For example:

• You: "!baz What's going to happen to my printer when I use cheap ass filament from Sidd?"
• Baz: Meatbag Daymo reckons your printer will explode. I'd listen to him, humans are squishy.

There's an almost endless set of fun and useful features we can implement but they all rely on the searchability of messages. Discord doesn't support search for bots, so we need to create our own database of posts.

Repo structure

• bazbot/ - The Python code lives in this directory (it's a python module)
• docker-compose.yaml - Docker services: Qdrant and Redis

Non repo files/paths:

.env is required to specify the TOKEN value (not in this repo) .venv is typically where you are going to install a virtual env for Python data is a directory where the persistent data from the services will be written.

Architecture

The basic bot is designed to be super simple and reliable. It's no good if we're working on Baz and we break it and posts are missed. For this reason, the main bot is separate from the 'worker'.

The worker listens for bot messages, executes those asynchronously, and returns data which will be printed out by the main bot. Since we will be invoking some kind of large-language model (probably GPT 3.5-turbo), the bot will take some time to respond. The bot shows 'typing' while it's waiting for a worker response.

bazbot/bot is the main bot. This imports src/vectors.py for the embeddings duties. This is invoked with a plain python -m bazbot.bot.

bazbot/worker is where the worker code goes.

bazbot/brain calls the large language model. We will use OpenAI until such time as we have a suitable open source LLM and chonky hardware to run it.

bazbot/attitude includes the class responsible for creating the prompts for a large language model. It's where we inject our bot's personality, and there we model emotional state based on response. Baz doesn't much like humans, but how much it doesn't like them depends on what you ask it to do.

harvest is a CLI script that has the bot join a server and harvest historical messages from every channel. It stores highwater dates in channels.pickel in case it's interrupted or there is an error during the harvest sequence.

searchcli is a CLI script for testing that allows you to do semantic search over the vector database.

launch.sh is a bash shell script to run the worker and the bot. arq will reload if there are any changes to the worker.

To-do:

Seems like we can basically put harvest into the regular bot code and just have it as a start-up action. The bot ought to function while a harvest is going on. The channels.pickle data structure would need to be updated under normal operation. The pickle should probably be an sqlite or something.

Code style

• Async, no blocking shit anywhere (vectors.py doesn't currently comply)
• Full type hints, in VS Code search for type checking mode in Pylance, and switch this to Basic.
• No clever shit, obvious code for non-Pythonistas.
• Doc strings and comments.
• 'Black' code formatting

Installing

• Install Python 3.11
• git clone this repo
• cd into the repo
• python3.11 -m venv .venv
• source .venv/bin/activate
• pip install -r requirements.txt
• Create a .env file with TOKEN= for Discord, and OPENAI_API_KEY= for OpenAI values, and GUILD_ID for the harvester tool.

Running the bot (manually)

• docker compose up -d to run the services
• source .venv/bin/activate to activate Python virtual environment
• arq bazbot.worker.WorkerSettings
• python -m bazbot.bot

Run python src/harvest.py to harvest historical posts.

How to contribute

• Create your own bot, maybe look at this tutorial
• Get up to speed on Discord.py
• Create your own discord server (it's called a guild in API-speak)
• Set up your '.env' file with TOKEN='dsflkjsdfljkf' etc in it.
• Clone and run the repo, make sure it works as you expect.

If you're not huge on Python or this docker stuff. Get WSL2 installed on your machine, if you use Windows. Install Docker for Windows and you should be fine.

PRs welcome but discuss them in the Bogans channel first, tbd.

License

Neckbeard Commons 1.0: Who Gives a Fuck.