My current setup

ai
homelab
Author

Werner Schiller

Published

September 17, 2026

A few words about how i use AI โ€ฆ

A short overview over AI in my homelab

Besides experimenting, I have 3 major use-cases for AI in my homelab:

  • conversation agent for HomeAssistant
  • vscode / zoo code / agentic coding
  • OpenWebUI for whatever tasks i imagine

๐Ÿ  HomeAssistant conversation agent

This was my first productive implementation. I experimented a lot with local AI, but it just was not good enough, so one day i just bought some credits from DeepSeek and this was an eye-opener.

There is still a lot of experimenting going on, but currently i am using glm-5.3-flash from z.ai. Their coding plan Lite is very nice and the best bang-for-the-buck i currently know of.

Also of course STT and TTS are noteworthy, but after some early experiments i decided to use my time differently.

Naturally, this deserves a separate post, which will be done some time in the future.

๐Ÿฆ“ VScode / zoo code / agentic coding

I use several models for different tasks in vscode for my agentic loop coding. Detailled description of those is for another day, but i have a good readme (wellโ€ฆ written by AI, but still good) at my AI-snippets project.
I drop mention bifrost AI gateway here because this is the basis of my setup. And the reason why i do not have to sell my firstborn to be able to afford my workflow and by current coding rate.

There will be another post about this, where i go into more detail about the different modes and what the core idea behind that is.

๐Ÿ•ธ๏ธ OpenWebUI

Openwebui is what i use if the workload does not fit in any other category. There is a lot of configuration, specialized modes, automations, skills and it was the reason why i built ssh-mcp and caldav-mcp in the first place.
Again - this deserves a separate post, which i will post soon.

๐Ÿ› ๏ธ Others / Hardware

I have a server with 32GB RAM and a Nvidia 5060 Ti (because, again, i do not want to sell my firstborn), which i use for hosting my services (for example bifrost, as i mentioned before, but also an ollama instance with qwen3-embedding:4b for - well - embedding, a qdrant server and many things more). I also am using it for experiments with STT, TTS, local ImageGen and so on.
Well - i dropped local ImageGen in favor of more productive use-cases.

๐Ÿ—ฟ Outro

This is my first real content posting. I still have to get the knack for it, but expect some more postings. I will try to make a weekly schedule after i got the first few posts online.

Thanks for reading!