Pangolin would work, it's still wireguard underneath but with a webUI to set up things and simplify it. It's fairly easy to install on your server with their docker compose examples.
Yeah from a performance and safety standpoint I don't think there's much difference. There is some argument that repair parts availability might be worse with Chinese vehicles, but the usual domestic manufacturers are also actively hostile to repairs now anyway so I don't know how different that will really be.
Plus if the general requirement is 'high quality code' then that automatically excludes the majority of LLM usage where people just vibe code it and rely heavily on the LLM to do the work. VS someone writing their own code while using an LLM to help find problems.
I think “the submitter is entirely responsible for their code” is a very reasonable position.
Especially since I just don't see how it's possible to fully prevent people from using AI if they want to use it. Yes a lot of code might be more obvious and could be rejected with an "absolutely no AI policy" but some people will submit perfectly good code where AI was involved in some way.
Performance on opnsense is fine for anything you'll be doing at home, so I wouldn't factor that into your choice.
Features on opnsense are probably better, just due to all the packages you can install.
Do you prefer being 100% in a CLI all the time? If not I'd say opnsense, the GUI certainly makes things a lot easier when you're not fully familiar with something and spending every day interacting with it.
You don't technically need to back up the PBS server as there's not much there other than a basic config. As far as the backup data itself, you can mirror to another PBS server or I think it supports S3 for offsite storage.
The immich public proxy app is built for this, you run it on your isolated public VLAN, and it displays a shared album to people while loading the images from immich.
If a politician is so out of the loop that they have no idea what their own voters want, they should not be in office. Nice to see that work for once since it feels like it rarely does.
The SFF or MT form factors are a lot better, I'd say MT is the best as it has full height PCIe slots. Keep in mind the dell/hp/lenovo models all use proprietary motherboard form factors and power supplies, but not that big of a deal I think since there are so many parts available if something does break.
I highly recommend 7th gen intel or newer, as you get the much better quicksync support and quality.
If you get a desktop class CPU (i5-7500 for example), the whole computer will typically draw around 15W at idle with an SSD, which is pretty decent.
If you need less idle power draw then you'll want a mobile/notebook class CPU (like an i3-7100u) as the idle usage should be less than 5W. But those typically only come in the micro/mini form factors.
Also good to remember that every 3.5" HDD draws around 7W when idle and spun-up (typically difficult to spin down on servers since there's always some process accessing files).
Or the 'modern workflow' version of tell the AI you want to convert the PDF, so it burns a bunch of tokens installing pandoc and converting it for you!
Maybe run the public VPN at home with NAT enabled, and use it as the default gateway in the private VPN. Never done it but I think I've seen some guides on that concept.
Pangolin would work, it's still wireguard underneath but with a webUI to set up things and simplify it. It's fairly easy to install on your server with their docker compose examples.