Thank you! These days it’s good not to master English enough to be mistaken for an ai bot

My mother language is Italian, not English. But I spent some time living in the US so maybe there is that at my advantage.

LazyNVR was started without any AI assistance… Both use of ai and the rule came afterward.

I am not native English speaker, thanks for the correction.

Smartpants, you wouldn’t have read it either way…

And I wrote in the first sentences that no ai has been used to write it’s it’s all my personal salt

Should I edit the original post? I can delete the text and leave the link

My data is a raid5 (Linux MD raid) on top of which I have btrfs partition. They are 4 dad’s.

I have another raid5 on USB mechanical disks again MD raid5 + btrfs, which is my 1st level backup.

Os, Gentoo Linux, is on an nvme ext4, going to migrate to btrfs when I have time.

Server is a Ryzen 9 with 64gb ram that I also use for AI inference thanks to two Nvidia RTX A4000. Currently one on PCIe 16x and one on pcie 4x. I am waiting for a pci-e bifurcation card to move both cards on 8x 8x pcie bus… Splitting my CPU bound 16x lanes in two…

Edit: I host lost of stuff… Mostly on bare metal or podman rootless…

Check out my wiki at https://wiki.gardiol.org/ where I keep track of what i install and how.

I use btrfs on main data disks and backup disks. Recently it caught a data corruption due to a wrong write (neutrino? Luck? Halloween approaching?).

Then it occurred to me that similar corruption would have been undiscovered for years until I needed that file on ext4…

So, that’s a good reason to migrate…

I need to replace my server root filesystem from ext4 to btrfs…

And get rid of a leftover MD/raid1 as well …

I have no idea what mas is, and I never did element calls. You should try it or check online.

Authelia here

All text config files, so either you love it or hate It

Stable and works fine.

All i Need and use works, if you have anything specific on mind…

Continuwuity Is a single binary that runs with very minimal memory and literally zero CPU footprint, and basically minimal setup, where Synapse on the same hardware was a mess to setup hard to configure, and was noticeable both on CPU and RAM usage.

Maybe for many users synapse scale better, i have no idea, but for a single user or few users instance, no way.

I have been on Continuwuity since Conduwuit was dropped!

No issue, fast, very lean on resources… Never missed fucking Synapse once!

Great work guys!

Yeah problem is when upgrades break stuff…

Unattended upgrade?

Why? Specially on critical systems?

If you do such upgrades, schedule for a time you don’t use the system at very least …

Reverse proxy my friend…

Using one static IP per service feels so much like … 1990s…

… As well as owing a /24, I admit.

Not really, they are testing your passwords against have I been pawnd. I agree that transparency should apply here, and also an upfront question to skip.

But it’s not “unsafe” or a privacy breach per se. No password has actually been “sent” anywhere only a partial hash that is useless for accessing your home assistant.

Yes i do.

I esperiment a lot, but my stable setup is Qwen 3.8 27B Q6_K on my home server. Hardware is Ryzen 9 5950x with 64gb ddr4 and two old but effective NVIDIA RTX A4000 for a total of 32gb VRAM.

Qwen 3.8 is the best model I can run in terms of output quality. It can perform all tasks pretty well and it’s reasonably fast at 500ts prefill and 30/40ts processing. There are much faster models but honestly not as good.

I can also run gemma4 all on CPU, with no GPU at all.

The key point is having both GPU on pci-e CPU, not on chip set bridge, so I am using a PCI riser 8x8x to share the 16x slot between the two GPUs. Inference speed is unaffected, but prefill benefits immensely.

Software side, use llama.cpp on Linux, you get the best performance possible at the expense of a bit more complexity compared to other engines.

From a power consumption perspective, I tined down my CPU governor and maximum frequency, and also the GPUs maximum watts. At ide the system is around 80w. At peak, it’s about 280w with the limiters, and inference speed doesn’t suffer too much. Without limiters I can easily reach 650w.

Edit: I run llama-swap in front of llama-cpp, and openwebui as agent manager/ web GUI. I also have ComfyUI installed because why not but seldomly use it.

You can read more of my setup on my wiki https://wiki.gardiol.org/8-llm-and-ai (no ads, no monetization whatsoever because fuck a hobby should not be a side gig)

I bought it new, but of course you can repurpose old hardware!

I bought a firewall appliance based on an Intel CPU, with 4x 2.5gbps Ethernet ports, 8gb ram. Installed opnSense on it and been super happy even since!

A bit pricey the hardware, but it’s been totally rock solid for almost two years now.

I have configured unbound with ad blocking on opnSense, and it has every single option I would ever want on it. Amazing stuff.