Now make it open source

Eh, they’ve thought of that.

Openshift uses immutable os images and can install over PXE. You can even automate deployment using ipmi/idrac, though that was really buggy when I tried it in 2022.

I have run plenty of clusters on bare metal, both Openshift and vanilla. No VMs are needed.

Look at Openshift if you’re looking for immutable, production ready Linux infrastructure. Containers are quickly replacing VMs.

Should work well for that!

If you use cloudflare for dns only and turn cloudflare proxying off, none of your data or traffic goes to cloudflare’s servers. They just act as your dns server, telling your devices what IP to go to.

it might be better to skip the cloud server and use cloudflare for dynamic dns. The standardized way to restrict access to websites is with client certificates or a basic authentication (user/pass) proxy. That would help avoid issues with internet traffic passing through the VPN accidentally.

Nixos’ weakness is definitely it’s documentation. There’s often configuration snippets you can copy and paste, though. If you go with NixOS, make sure to come back with questions, the community is very helpful.

Unraid is bad at NAS and bad at docker. Go with a separate Nas and application server.

Is there a way for me to be “notified” if shell access of any form is gained by someone?

Falco is a very powerful tool for this.

If you’re not using something like synology, it isn’t really an issue to run applications and nas on the same machine. I would generally recommend separating them so you have more options in the future if you want to run muliple servers for HA or expansion, but it should be fine either way. It is worth noting that quad core N100 computers are like $150 on aliexpress if you want a cheap application server(s).

Generally it’s simpler if you have your NAS separate from your application server. Synology runs NAS really well, but a separate application server for docker/etc is a lot easier to use and easier to upgrade than running on Synology. Your application server can even have a GPU for media transcoding or AI processing. Trying to do everything on one box makes things more complicated and fragile.

I would recommend something like Debian or NixOS for the application server, and you should be able to manage it over SSH. You can then mount your NAS as an NFS share, and then run all your applications in Docker or NixOS, using the NAS to store all your state.

That’s fair. I’m just thinking I could never use something like this because I would be invading the privacy of others using my Jellyfin. I would live to see an anonymous view counter on every movie though tbh.

Seems pretty creepy to be collecting logs about what people watch. Why do people use this?

You need IP cameras and then you need a NVR server for recording, detection, and display. There are some good open source NVR programs out there with docker support. I’ve been wanting to try Viseron. There’s also ZoneMinder and Shinobi that seem to be good.

Unfortunately most consumer cameras are cloud only. This seems to be a list of cameras you can look into: https://wiki.zoneminder.com/Hardware_Compatibility_List

Your best bet is probably a chinese brand for cameras. Dahlua seems popular. There are also a bunch of PoE cameras on Aliexpress for $15-25, but I can’t attest to if they’re any good. Hikvision cameras seem to have been popular too, but they have been recently sanctioned by EU/US for human rights violations.

The man is good with a soldering iron, but damn, what a creation he made lol

STH measured 23w on theirs, but it can vary based on which one you buy. Tons of compute power with those 4 E cores.

https://www.servethehome.com/fanless-intel-n100-firewall-and-virtualization-appliance-review/4/

$150 fanless N100 pc with 4x2.5gbps from aliexpress and install OPNsense on it.

deleted by creator

I guess that makes sense, but I wonder if it would be hard to get clean data out of the per-token confidence values. The LLM could be hallucinating, or it could just be generating bad grammar. It seems like it’s hard enough already to get LLMs to distinguish between “killing processes” and murder, but maybe there could be some novel training and inference techniques that come up.

I thought confidence levels were for image recognition? How do confidence levels work for transformer LLMs?