Is “my users” a valid answer? xD

While you can, there has been research that showed that the LLMs answers can be influenced by the order in which the possible options are presented.

I haven’t looked into how these “System One” “Noul” models (absolutely stupid nomenclature imo) work but the hope would be that theyre less susceptible to such problems.

If you don’t mind me asking, what made you go with VoidAuth vs Authelia or something else entirely?

I’m in a similar boat as OP and while VoidAuth looks very promising I’m put off a little by the young age and size of the project.

Harper is great, but it only does English. Really wish their “extensible core” was extended to other languages sometime.

Got it working thanks to your troubleshooting tips now. Also found a very neat way to handle secrets from another comment.

I tend to run a DB instance per service as that makes backup restoration much easier for me. An idle postgres sits at around 50MB which is a cost I’m willing to pay.

Thank you again for your help :)

Managed to get it working by passing in env vars from a secret now.

ArgoCD has a really handy web UI that allows you to quickly see what kind of resources get deployed.
Especially for learning k8s I found that much easier to visualize than raw kubectl outputs.

Passing in the secrets once via the global: section is very neat. Got it working now with a few of the other tips and stole your trick for my secret handling. Thank you :)

Did you also have to set all these env vars by hand?
I am wondering if it might have something to do with rendering Helm Charts under ArgoCD.

I’ll give it another try with your recommendations.

And should I get it working finally, I will obviously switch back to using Secrets.
I only removed them to reduce possible points of failure.

As for blueprints, that’s a task for future me xD

Same here. Literally just set it up and now this.

I hope the author will roll this back or someone else makes a fork. I don’t want to immediately switch technology to XMPP/Matrix/… and have to do it all over again.

Also the normal and rpi versions are two completely independent implementations of the same software. So now the LLMs have twice the maintenance load.

I didn’t diff the two files but even the startup and control code appears to be custom for each version.

Since projects of the same language often use the same tooling this makes it easier to clean up the whole directory by running something like this:

for d in ./*/ ; do (cd "$d" && somecommand); done

somecommand could be cargo clean if you’re in the Rust directory for example.

Depends on the exact system but there will be a method to switch to a newer release channel without reinstalling. Rinse and repeat every x years.

Interesting, GPG has been working just fine for me so far.

My main issue with it remains that barely anyone else uses GPG.

Not just on the web. I’ve previously used it to embed a short clip in a presentation.

The nice thing is that it doesn’t do a massive screencap but only captures the text. This way the replay will be freshly rendered at native resolution.

Hoping to get a piece of that fish

How well NixOS fits your purpose really depends on what you want to do with the OS. If you’re just going run a bunch of docker containers, you could manage them via Nix but its a little cumbersome.

Where NixOS really shines for small servers are the so called NixOS Options. They allow you to install tons of services on bare metal but manage all the configuration for you. E.g. open the correct firewalls ports, run a dedicated DB or cache, etc. and all those simply require you to enable them with an ... = true;.

Smaller projects might not have a NixOS Option available and some options are more and/or easier configurable than others, but if you’re running just a few common services you could feasibly manage your whole server with just one native config file and no docker shenanigans.

I’d recommend checking what’s available under the link above. If you wanna go the container route instead, you have the option of just using docker non-declaratively as on every other distro (but then you lose some of the benefits NixOS gives you), or you can declaratively have NixOS manage all the docker containers. There are a few ways to do and manage this so some further research will be required.

Then helmfile might be worth checking out

You dont need to manually handle the WG config files. This isn’t really an issue when it’s just you and your two devices, but once you start supporting more people, like non-technical family members, this gets really annoying really quickly.

Tailscale (and headscale) just require you to log in, which even those family members can manage and then does the rest for you. They also support SSO in which case you wouldn’t even have to create new accounts.

There are some experimental models made specifically for use with Home Assistant, for example home-llm.

Even though they are tiny 1-3B I’ve found them to work much better than even 14B general purpose models. Obviously they suck for general purpose questions just by their size alone.

That being said they’re still LLMs. I like to keep the “prefer handling commands locally” option turned on and only use the LLM as a fallback.

I’ve had this exact same gripe and can thankfully report that running EarlyOOM has fixed this for me.