I am a Software Engineer - and I am afraid of magic.

At least since that magic is also quite real, fast and hungry for data and adoption.

With that unwell feeling of missing something really important and maybe hitting an iceberg quite soon, I went on the journey to explore, demystify and tame(?) AI.

Luckily I have a bit of free time on my hand. With that and some curiosity - I began to work on “Relagent” about 1 month ago as my primary project.

Starting point

At that time I had not much experience with AI. My expertise could be summarized with knowledge queries and AI assisted coding. To answer daily questions I use Duck.ai because it requires no account and promises a high level of privacy. I’d been using them for simple questions, in order to get quick introduction or as a starting point of a longer research. As an software engineer I had used the Cursor IDE for about 1 month with varying results.

The challenge

Can I build a self-hosted AI assistant that is affordable, actually useful and not leaking my data around the world?

I had been looking at projects like AnythingLLM, LocalGPT , GPT4All, Project Goose or LMStudio. Which I find very interesting and inspiring. But my goal was different. I wanted the convenience of just using my mobile device to get help with daily tasks.

Initial plan

In 2025 I started consuming information on Artificial Intelligence from a Software Engineering perspective. Based on all the theoretic knowledge I drew up this initial architecture: Pasted image 20251205005918.png