One of the most frustrating things I deal with is remembering relevant information. Some people thrive on having a great knowledge storing system like Zettlekasten - a pen-and-paper system where you write individual ideas or pieces of information on separate slips or index cards and then link them through a unique numbering or tagging system. In Zettlekasten, each note becomes part of a network of ideas, making it easier to retrieve and connect information later. It sounds great on paper (pun intended) but like anything that requires repetition, I often stop doing it after day 4. So, I’m introducing my own Germanesque system for note-taking - Klugscheisser (German for smartass. Literally: knowledge shitter).
What is Klugscheisser?
Klugscheisser is a Slack bot that learns knowledge when you tell it to and retrieves it when you ask. Think of it as the collective memory of your Slack workspace that doesn’t disappear when you close a tab or get distracted by searching for the perfect reactji.
What makes Klugscheisser special is that it doesn’t just do exact keyword matching. It uses vector embeddings and semantic search to understand the meaning behind your questions.
Features That Actually Work
- Knowledge Acquisition: Simply mention
@klug-bot --learnfollowed by whatever information you want it to remember. - Document Learning: Attach files (.md, .json, .csv, .txt, .pdf) and Klugscheisser will digest them.
- Semantic Retrieval: Ask questions in natural language and get relevant answers.
- Spring Cleaning: Remove outdated information with the
--deletecommand. - Offline Uploading: Bulk import documents when you’re feeling particularly educational.
The bot logs all learning and deletion actions to a separate slack channel, so you can keep tabs on who’s teaching your bot what. And, let’s be honest, because we all have that one colleague who would absolutely try to teach the bot something completely inappropriate, you can also specify who is allowed to teach the bot new things.
Under the Hood
For those who care about such things, Klugscheisser is built on:
- Python 3.10+
- PostgreSQL for stable storage
- Chroma for vector embeddings
- Google Gemini for natural language understanding
The whole thing runs in Docker, so deployment is as painless as technology allows these days (which is to say, mostly painless with occasional bouts of existential questioning).
Getting Started
Want to try it out? The setup process is pretty straightforward, especially if you’re familiar with Docker. Here’s the TL;DR:
- Clone the repository
- Set up environment variables
- Run docker
compose up --build - Configure your Slack app
- Start teaching your bot all the things
For those who want the detailed instructions with all the fancy command-line snippets, check out the Klugscheisser GitHub repository.
Real-World Examples
Here’s how Klugscheisser operates in the wild:
Learning:
@klug-bot --learn John Smith leads Backend/SRE.
Querying:
klug-bot Who's in charge of the backend?
Deleting (because we all say things we regret):
klug-bot --delete source:slack date:2025-02-22
Why I Built This
If you’ve worked in any organization larger than three people, you know the pain of knowledge management. Information gets siloed, buried in Slack threads, lost in Google Docs, or worst of all, stored exclusively in that one engineer’s brain who’s currently backpacking in the mountains with no cell service. After the 157th time of asking “Wait, who do we contact for AWS issues again?” I decided enough was enough. The solution couldn’t be yet another wiki that no one updates. It had to be integrated where we already communicate. Also, I really wanted to call something “Klugscheisser” in a professional setting and get away with it.
What’s Next?
There are a few exciting features on the roadmap:
- Summarizing conversations in threads into knowledge units.
- Handling images for learning.
- Maybe, just maybe, a feature that automatically identifies important information without being explicitly told to learn it.
Try It Out
If you’re tired of being your team’s human search engine, give Klugscheisser a try. It’s open source, MIT licensed, and ready for you to deploy. And if you find yourself in a Slack conversation soon saying, “I remember reading something about that but I can’t find it,” well, that’s precisely the problem this smartass bot is designed to solve.
Disclaimer: The views expressed in this article are my own and do not necessarily represent the views of my employer.