Overview
A self-hosted AI assistant running on a server I built at home. The AI uses RAG (Retrieval Augmented Generation) files as a knowledge base — I gave it the full Proxmox documentation (685 pages) so it can answer infrastructure questions directly from the source material.
The AI is connected to the internet and I integrated it with Telegram — so I can chat with my AI assistant from anywhere just by sending a message in Telegram, making it a proper personal assistant for planning and reminders.
Technology Stack
Supermicro Server + GPU
An older server I set up at home with a graphics card — the GPU allows AI models to run much faster than on a normal CPU
Ollama
Tool that downloads and runs AI models locally on my own hardware
AnythingLLM
The chat interface — I can talk to the AI through the browser and give it documents to read
Ubuntu Server 22.04
Linux operating system running on the server
Docker
Used to run AnythingLLM in an isolated container
Telegram
Messaging app integration — chat with the AI from anywhere via a Telegram conversation
Key Features
RAG Knowledge Base
RAG (Retrieval Augmented Generation) files — I gave the AI the full Proxmox documentation so it can answer questions from the actual source material
Internet Connected
The AI can search the web for up-to-date information, not limited to the knowledge base
Telegram Integration
Linked to Telegram so I can send messages to my AI from my phone — available from anywhere without opening a browser
Multilingual
Works in German and English — useful for both school and personal use
What I Learned
- Why a graphics card makes such a big difference for AI — the GPU handles the parallel computations that AI models require
- That different AI models are made for different tasks — a code-focused model is not just "stronger", it's simply a different tool
- How much hardware resources AI actually requires — the RTX 2060 with only 6 GB VRAM turned out to be a real bottleneck for larger models