Self-Hosted Software: Apps Worth Running at Home
The software layer of a home lab: which platform to build on, what each one assumes about your machine, and why the migration path out matters more than the feature list.
Hardware is the part people research and software is the part they live with. This pillar is about the second one, and it is the newest and thinnest section of the site, which I would rather say plainly than pad.
The choice that comes first
Before any individual application, there is a platform decision, and it is worth understanding as three different assumptions about what a home server is for rather than as a feature comparison:
Proxmox is a hypervisor that can store things. Its unit of work is a virtual machine or a container, and storage exists so those have somewhere to live.
TrueNAS is a storage system that can run things. Its unit of work is a dataset, and apps exist so the storage has something to serve.
Unraid is a media server built around a storage model nobody else uses. Its defining feature is that the drives do not have to match.
Proxmox vs TrueNAS vs Unraid works through what follows from each, including the structural difference in how they store things, which is not a preference. That article is labelled as research rather than testing in its first sentence, because I have not run all three in production and it matters which kind of comparison you are reading.
The thing to check before you commit
The migration path out, which almost nobody researches until it is too late, and which is asymmetric.
Moving between systems that both use ZFS is straightforward, since a pool can be imported. Moving from Unraid to ZFS means rebuilding the array and copying everything, because the storage models are not compatible. That asymmetry costs nothing to weigh at the start and a great deal to discover later.
The cheap test is to run whichever you are leaning toward in a virtual machine first, with fake disks, and click through the thing you will do most often.
What belongs in this pillar
- Platform guides: what to build the lab on, and what each choice assumes
- Application walkthroughs: putting a web interface in front of a local model, remote access without opening ports, container stacks worth copying
- Operational habits: updates, backups of configuration rather than just data, and what to do when something stops working at an inconvenient hour
Why this pillar is short
Most of what belongs here is a how-to whose entire value is that the commands work. I write from a container without the hardware or the running services to verify against, so a walkthrough I published today would be a walkthrough I had not tested.
I would rather this section grow slowly and be right than arrive complete and be approximately right. The queue for it exists; the machines to check it against are the thing that unblocks it, which is the same bottleneck as the benchmark database.
In the meantime, the one page here is the platform comparison, which is genuinely useful without a machine because it is about design assumptions rather than commands. And the local AI runtime that most self-hosted app guides sit on top of has a tested starting point in running your first local LLM with Ollama, including the API that a web interface would point at.
FAQ
What should I self-host first?
Something whose failure you would find annoying rather than painful, so you learn the operational habits before anything important depends on them. A local model runtime is a good first choice because nothing outside your machine breaks when you get it wrong, and it gives you an immediate reason to care about backups and updates.
Do I need Docker, or virtual machines?
Containers for applications, virtual machines for whole systems that need to be isolated or that run a different operating system. Most home labs end up with both, and most single-purpose applications you want are distributed as containers, so that is the format to be comfortable with first.
Can I run everything on one machine?
You can, and all three platforms will let you try. The compromise is real though, and the honest question is which half you will be annoyed to compromise on. Storage integrity and virtual machine density pull in different directions, and a machine doing both is doing neither as well as a machine doing one.
Should I virtualise TrueNAS under Proxmox?
It is a common setup and it works, with one condition that is not optional: pass the storage controller through to the virtual machine rather than handing it virtual disks, so ZFS is talking to real hardware. Done casually it puts a layer between ZFS and the disks it is checksumming, which undermines the reason to run ZFS at all.
Which platform uses the least electricity?
Unraid has a structural advantage on a large, mostly-idle array, because a read spins one drive rather than all of them. How much that is worth depends on your drive count and how idle the array really is, so it is arithmetic rather than a general claim. Put your own numbers into the power cost calculator; on a small array that is busy most of the time the difference largely disappears.
Everything in this pillar
One article, newest first.