But Tailscale is the real unlock in my opinion. Having a slot machine cosplaying as sysadmin is cool, but being able to access services securely from anywhere makes them legitimately usable for daily life. It means your services can be used by friends/family if they can get past an app install and login.
I also take minor issue with running Vaultwarden in this setup. Password managers are maximally sensitive and hosting that data is not as banal as hosting Plex. Personally, I would want Vaultwarden on something properly isolated and locked down.
That said, I'm not sure if Bitwarden is the answer either. There is certainly some value in obscurity, but I think they have a better infosec budget than I do.
Avoid stacking in too many hard drives since each one uses almost as much power as the desktop does at idle.
related "webdev is fun again": claude. https://ma.ttias.be/web-development-is-fun-again/
Also the "Why it matters" in the article. I thought it's a jab at AI-generated articles but it starts too look like the article was AI written as well
[1] https://martin.kleppmann.com/2025/12/08/ai-formal-verificati...
Waiting for the follow-on article “Claude Code reformatted my NAS and I lost my entire media collection.”
Is there a replica implementation that works in the direction I want?
From to time, test the restore process.
Took a couple hours with some things I ran across, but the model had me go through the setup for debian, how to go through the setup gui, what to check to make it server only, then it took me through commands to run so it wouldn't stop when I closed the laptop, helped with tailscale, getting the ssh keys all setup. Heck it even suggested doing daily dumps of the database and saving to minio and then removing after that. Also knows about the limitations of 8 gigs of ram and how to make sure docker settings for the difference self services I want to build don't cause issues.
Give me a month and true strong intention and ability to google and read posts and find the answer on my own and I still don't think I would have gotten to this point with the amount of trust I have in the setup.
I very much agree with this topic about self hosting coming alive because these models can walk you through everything. Self building and self hosting can really come alive. And in the future when open models are that much better and hardware costs come down (maybe, just guessing of course) we'll be able to also host our own agents on these machines we have setup already. All being able to do it ourselves.
I’ll bite. You can save a lot of money by buying used hardware. I recommend looking for old Dell OptiPlex towers on Facebook Marketplace or from local used computer stores. Lenovo ThinkCentres (e.g., m700 tiny) are also a great option if you prefer something with a smaller form factor.
I’d recommend disregarding advice from non-technical folks recommending brand new, expensive hardware, because it’s usually overkill.
And then you can only use distros which have a raspberry pi specific build. Generic ARM ones won't work.
I build out my server in Docker and I’ve been surprised that every image I’ve ever wanted to download has an ARM image.
Then someday we self-host the AI itself, and it all comes together.
My self hosted things all run as docker containers inside Alpine VMs running on top of Proxmox. Services are defined with Docker Compose. One of those things is a Forgejo git server along with a runner in a separate VM. I have a single command that will deploy everything along with a Forgejo action that invokes that command on a push to main.
I then have Renovate running periodically set to auto-merge patch-level updates and tag updates.
Thus, Renovate keeps me up to date and git keeps everyone honest.
(In)famous last words?
It's not tariffs (I'm in Switzerland). It's 100% the buildout of data centers for AI.
We've gone a step further, and made this even easier with https://zo.computer
You get a server, and a lot of useful built-in functionality (like the ability to text with your server)
I agree you could use LLMs to learn how it works, but given that they explain and do the actions, I suspect the vast majority aren't learning anything. I've helped students who are learning to code, and very often they just copy/paste back and forth and ignore the actual content.
And I find the stuff that the average self hoster needs is so surface level that LLMs flawlessly provide solutions.
I am writing a personal application to simplify home server administration if anybody is interested: https://github.com/prettydiff/aphorio
p0wnland. this will have script kiddies rubbing their hands
But I want to host an LLM.
I have a 1U (or more), sitting in a rack in a local datacenter. I have an IP block to myself.
Those servers are now publicly exposed and only a few ports are exposed for mail, HTTP traffic and SSH (for Git).
I guess my use case also changes in that I don’t use things just for me to consume, select others can consume services I host.
My definition here of self-hosting isn’t that I and I only can access my services; that’s be me having a server at home which has some non critical things on it.
Thanks
I suspect it may have been related to the Network File System (NFS)? Like whenever I read a file on the host machine, it goes across the data-center network and charges me? Is this correct?
Anyway, I just decided to take control of those costs. Took me 2 weeks of part-time work to migrate all my stuff to a self-hosted machine. I put everything behind Cloudflare with a load balancer. Was a bit tricky to configure as I'm hosting multiple domains from the same machine. It's a small form factor PC tower with 20 CPU cores; easily runs all my stuff though. In 2 months, I already recouped the full cost of the machine through savings in my AWS bill. Now I pay like $10 a month to Cloudflare and even that's basically an optional cost. I strongly recommend.
Anyway it's impressive how AWS costs had been creeping slowly and imperceptibly over time. With my own machine, I now have way more compute than I need. I did a calculation and figured out that to get the same CPU capacity (no throttling, no bandwidth limitations) on AWS, I would have to pay like $1400 per month... But amortized over 4 years my machine's cost is like $20 per month plus $5 per month to get a static IP address. I didn't need to change my internet plan other than that. So AWS EC2 represented a 56x cost factor. It's mind-boggling.
I think it's one of these costs that I kind of brushed under the carpet as "It's an investment." But eventually, this cost became a topic of conversation with my wife and she started making jokes about our contribution to Jeff Bezos' wife's diamond ring. Then it came to our attention that his megayacht is so large that it comes with a second yacht beside it. Then I understood where he got it all from. Though to be fair to him, he is a truly great businessman; he didn't get it from institutional money or complex hidden political scheme; he got it fair and square through a very clever business plan.
Over 5 years or so that I've been using AWS, the costs had been flat. Meanwhile the costs of the underlying hardware had dropped to like 1/56th... and I didn't even notice. Is anything more profitable than apathy and neglect?
On self-hosting: be aware that it is a warzone out there. Your IP address will be probed constantly for vulnerabilities, and even those will need to dealt with as most automated probes don't throttle and can impact your server. That's probably my biggest issue along with email deliverability.
Haproxy with SNI routing was simple and worked well for many years for me.
Istio installed on a single node Talos VM currently works very well for me.
Both have sophisticated circuit breaking and ddos protection.
For users I put admin interfaces behind wireguard and block TCP by source ip at the 443 listener.
I expose one or two things to the public behind an oauth2-proxy for authnz.
Edit: This has been set and forget since the start of the pandemic on a fiber IPv4 address.
Historically, managed platforms like Fly.io, Render, and DigitalOcean App Platform existed to solve three pain points: 1. Fear of misconfiguring Linux 2. Fear of Docker / Compose complexity 3. Fear of “what if it breaks at 2am?”
CLI agents (Claude Code, etc.) dramatically reduce (1) and (2), and partially reduce (3).
So the tradeoff has changed from:
“Pay $50–150/month to avoid yak-shaving” → “Pay $5–12/month and let an agent do the yak-shaving”
I just wish this post wasn’t written by an LLM! I miss the days where you can feel the nerdy joy through words across the internet.
Having others run a service for you is a good thing! I'd love to pay a subscription for a service, but ran as a cooperative, where I'm not actually just paying a subscription fee, instead I'm a member and I get to decide what gets done as well.
This model works so well for housing, where the renters are also the owners of the building. Incentives are aligned perfectly, rents are kept low, the building is kept intact, no unnecessary expensive stuff added. And most importantly, no worries of the building ever getting sold and things going south. That's what I would like for my cloud storage, e-mail etc.
CC lets you hack together internal tools quickly, and tailscale means you can safely deploy them without worrying about hardening the app and server from the outside world. And tailscale ACLs lets you fully control who can access what services.
It also means you can literally host the tools on a server in your office, if you really want to.
Putting CC on the server makes this set up even better. It’s extremely good at system admin.
simonw•2h ago
I actually think Tailscale may be an even bigger deal here than sysadmin help from Claude Code at al.
The biggest reason I had not to run a home server was security: I'm worried that I might fall behind on updates and end up compromised.
Tailscale dramatically reduces this risk, because I can so easily configure it so my own devices can talk to my home server from anywhere in the world without the risk of exposing any ports on it directly to the internet.
Being able to hit my home server directly from my iPhone via a tailnet no matter where in the world my iPhone might be is really cool.
philips•2h ago
Now I have tailscale on an old Kindle downloading epubs from a server running Copyparty. Its great!
ryandrake•2h ago
Skunkleton•2h ago
mfcl•2h ago
Forgeties79•52m ago
I could send a one page bullet point list of instructions to people with very modest computer literacy and they would be up and running in under an hour on all of their devices with Plex in and outside of their network. From that point forward it’s basically like having your own Netflix.
Jtsummers•2h ago
Why did people use Dropbox instead of setting up their own FTP servers? Because it was easier.
atmosx•2h ago
All these are manageable through other tools, but it’s more complicated stack to keep up.
Frotag•2h ago
Cyph0n•1h ago
1. 1-command (or step) to have a new device join your network. Wireguard configs and interfaces managed on your behalf.
2. ACLs that allow you to have fine grained control over connectivity. For example, server A should never be able to talk to server B.
3. NAT is handled completely transparently.
4. SSO and other niceties.
For me, (1) and (2) in particular make it a huge value add over managing Wireguard setup, configs, and firewall rules manually.
drnick1•1h ago
Speaking of that, I have always preferred a plain Unbound instance and a Samba server over fancier alternatives. I guess I like my setups extremely barebone.
ryandrake•1h ago
simonw•1h ago
Tailscale gives me an app I can install on my iPhone and my Mac and a service I can install on pretty much any Linux device imaginable. I sign into each of those apps once and I'm done.
The first time I set it up that took less than five minutes from idea to now-my-devices-are-securely-networked.
zeroxfe•1h ago
right, like browsers are just sugar on top of curl
edoceo•52m ago
SchemaLoad•42m ago
dangoodmanUT•2h ago
dangoodmanUT•2h ago
drnick1•1h ago
I am not sure why people are so afraid of exposing ports. I have dozens of ports open on my server including SMTP, IMAP(S), HTTP(S), various game servers and don't see a problem with that. I can't rule out a vulnerability somewhere but services are containerized and/or run as separate UNIX users. It's the way the Internet is meant to work.
CSSer•1h ago
In every case where a third party is involved, someone is either providing a service, plugging a knowledge gap, or both.
sauercrowd•1h ago
In many cases they want something that works, not something that requires a complex setup that needs to be well researched and understood.
buildfocus•1h ago
You can also buy quite a few routers now that have it built in, so you literally just tick a checkbox, then scan a QR code/copy a file to each client device, done.
Topgamer7•1h ago
heavyset_go•1h ago
This is what I do. You can do Tailscale like access using things like Pangolin[0].
You can also use a bastion host, or block all ports and set up Tor or i2p, and then anyone that even wants to talk to your server will need to know cryptographic keys to route traffic to it at all, on top of your SSH/WG/etc keys.
> I am not sure why people are so afraid of exposing ports. I have dozens of ports open on my server including SMTP, IMAP(S), HTTP(S), various game servers and don't see a problem with that.
This is what I don't do. Anything that needs real internet access like mail, raw web access, etc gets its own VPS where an attack will stay isolated, which is important as more self-hosted services are implemented using things like React and Next[1].
[0] https://github.com/fosrl/pangolin
[1] https://news.ycombinator.com/item?id=46136026
edoceo•55m ago
heavyset_go•32m ago
I personally wouldn't trust a machine if a container was exploited on it, you don't know if there were any successful container escapes, kernel exploits, etc. Even if they escaped with user permissions, that can fill your box with boobytraps if they have container-granted capabilities.
I'd just prefer to nuke the VPS entirely and start over than worry if the server and the rest of my services are okay.
esseph•1h ago
With tailscale / zerotier / etc the connection is initiated from inside to facilitate NAT hole punching and work over CGNAT.
With wireguard that removes a lot of attack surfaces but wouldn't work if behind CGNAT without a relay box.
SchemaLoad•46m ago
Behind a VPN your only attack surface is the VPN which is generally very well secured.
sva_•38m ago
Edit: This is the kind of service that you should only expose to your intranet, i.e. a network that is protected through wireguard. NEVER expose this publicly, even if you don't have admin:admin credtials.
SchemaLoad•26m ago
I now know better, but there are still a million other pitfalls to fall in to if you are not a full time system admin. So I prefer to just put it all behind a VPN and know that it's safe.
buran77•36m ago
Ideal if you have the resources (time, money, expertise). There are different levels of qualifications, convenience, and trust that shape what people can and will deploy. This defines where you draw the line - at owning every binary of every service you use, at compiling the binaries yourself, at checking the code that you compile.
> I am not sure why people are so afraid of exposing ports
It's simple, you increase your attack surface, and the effort and expertise needed to mitigate that.
> It's the way the Internet is meant to work.
Along with no passwords or security. There's no prescribed way for how to use the internet. If you're serving one person or household rather than the whole internet, then why expose more than you need out of some misguided principle about the internet? Principle of least privilege, it's how security is meant to work.
zamadatix•32m ago
There was a popular post less than a month ago about this recently https://news.ycombinator.com/item?id=46305585
I agree maintaining wireguard is a good compromise. It may not be "the way the internet was intended to work" but it lets you keep something which feels very close without relying on a 3rd party or exposing everything directly. On top of that, it's really not any more work than Tailscale to maintain.
SoftTalker•26m ago
drnick1•8m ago
This incident precisely shows that containerization worked as intended and protected the host.
Etheryte•28m ago
drnick1•15m ago
Are you sure that it isn't just port scanners? I get perhaps hundreds of connections to my STMP server every day, but they are just innocuous connections (hello, then disconnect). I wouldn't worry about that unless you see repeated login attempts, in which case you may want to deploy Fail2Ban.
Frotag•18m ago
But some peers are sometimes on the same LAN (eg phone is sometimes on same LAN as pc). Is there a way to avoid forwarding traffic through the server peer in this case?
comrade1234•1h ago
PaulKeeble•1h ago
shadowgovt•59m ago
ssl-3•22m ago
But Tailscale is just a VPN (and by VPN, I mean: Something more like "Connect to the office networ" than I do "NordVPN"). It provides a private network on top of the public network, so that member devices of that VPN can interact together privately.
Which is pretty great: It's a simple and free/cheap way for me to use my pocket supercomputer to access my stuff at home from anywhere, with reasonable security.
But because it happens at the network level, you (generally) need to own the machines that it is configured on. That tends to exclude using it in meaningful ways with things like library kiosks.
SchemaLoad•50m ago
LLMs are also a huge upgrade here since they are actually quite competent at helping you set up servers.
Melatonic•4m ago