There’s just something about self-hosting that hits different. I’m a fierce pragmatist, and I never came across anything sufficiently convincing enough for me to see the value in doing it. But then along came Open WebUI.

For those unfamiliar, Open WebUI is a universal client/frontend for LLM interactions. From what I see on Reddit and elsewhere, it seems most of its users actually run open source models locally on their own machines using a tool like Ollama. But since I don’t yet have a homelab/server setup in my garage and I’m only sporting 16gb of RAM on my MacBook Air M1, I’ve almost exclusively been using OpenRouter, and it’s been spectacular.

With just a single API integration, I have access to just about every available LLM, both proprietary and open source. I started by running Open WebUI on my local machine, but then realized that I couldn’t easily access it on mobile, which is when I decided that I needed to deploy it on a server.

I started with Railway, but the monthly platform costs were 3-4x of what I had heard it cost to have your own VPS that could host dozens of instances of Open WebUI, so I figured it was as good a time as any to give that a shot.

After looking into quite a few different providers, I settled on Hetzner, and everything’s been smooth for a few weeks now.

Docker seemed to make more sense this time around, which seems to be a real reoccurring theme with software development. I’ll bang my head against some tool or language or concept for a few weeks and then angrily push it to the side when I just can’t really internalize it. But then when I bump back into it a few weeks later, I realize that I knew more than I thought I did, and I’m also able to get more of it because I picked up some new tidbits in the interceding weeks. I suppose this is the value of experience!