Skip to content

deputyOS vs Umbrel

Umbrel is the most polished self-hosting appliance OS; deputyOS is built around AI assistants and managing several machines at once. An honest comparison of where each one wins.

Last reviewed

Short answer: Pick Umbrel if you have one box and want the smoothest possible experience. Pick deputyOS if AI assistants are the point, you run more than one machine, or you need to prove what's in the image.

Where Umbrel is the better choice

  • Far more polished UI, and a genuinely Apple-like setup experience
  • A large one-click app store covering media, storage, automation and Bitcoin
  • First-party hardware (Umbrel Home, Umbrel Pro) you can just buy
  • Much larger community, so more answers already exist when you get stuck
  • Mature encrypted backup with a well-regarded restore experience

Where deputyOS is the better choice

  • Built for several machines: one console shows every device and pushes commands to all of them
  • AI assistants are the product, not one app category among dozens
  • A genuine air-gapped tier — model weights baked in, egress denied at nftables
  • Signed, reproducible images with SLSA attestations you can verify yourself
  • A/B slot updates with watchdog rollback, so a bad update reverts itself

The honest framing

Umbrel is a better general-purpose home server OS than deputyOS, and it is not close. It has a bigger app catalogue, a more refined interface, real hardware you can buy, and years of accumulated community knowledge. If you want one box in the cupboard that runs Plex, Home Assistant, a Bitcoin node and an AI chat app, Umbrel is the more sensible choice and you should stop reading here.

deputyOS is narrower on purpose. It does one category — personal AI assistants — and it assumes you may end up running more than one machine.

Where the difference actually shows up

Running more than one box

This is the real split. Umbrel is designed around a single device. Running three means three web interfaces, three update decisions, and no combined view of what they are doing or costing. That is a reasonable design choice for the audience it serves, and Umbrel has said centralised multi-node administration is not what it is building.

deputyOS treats the fleet as the unit. Every device holds an outbound tunnel to the control plane, so one console lists them all with live status, pushes an update to a selection, runs an allow-listed command across the group, and aggregates spend per device. There is no port forwarding on any of them.

If you have one machine, this buys you very little. If you have five, it is the difference between a hobby and something you can actually operate.

What “AI appliance” means in each

Umbrel reaches AI through its app store — Ollama and Open WebUI are apps you install, and they behave like apps. That is flexible and it works.

deputyOS bakes the assistant into the image. The runtime, native modules, model weights and signature databases are all present before first boot, so nothing runs apt, npm, pip or git clone on your device. The upside is determinism: two devices flashed a month apart run identical software. The downside is that adding something not in the image means a rebuild rather than a click.

Three profiles ship: OpenClaw (the generalist, 23 channels), Hermes (memory-rich, Python, FTS5 recall), and Khoj (personal-knowledge indexing). You switch with deputyctl profile switch <id>.

Offline operation

Umbrel expects a network. Apps are pulled from a store, updates arrive over the internet, and remote access currently routes through Tailscale.

deputyOS has an air-gapped tier as a build flag. AIRGAP=1 bakes LFM2 via llama.cpp, points apt at a local mirror, and locks egress at nftables to RFC1918 and mDNS only. It runs with -net none. This matters for a narrow audience — compliance, classified environments, industrial control — and is irrelevant to everyone else.

Verifying what you installed

Umbrel ships signed releases, which proves who built them.

deputyOS additionally publishes reproducible builds with SLSA provenance and minisign signatures, so you can rebuild from source and compare hashes rather than trusting the publisher. Whether that matters depends entirely on your threat model; for most home users it does not.

Where Umbrel is simply ahead

Being specific rather than diplomatic:

  • Interface quality. Umbrel’s is better designed and better tested.
  • Breadth. Dozens of apps against three assistant profiles.
  • Hardware. You can buy an Umbrel Home; deputyOS expects you to supply a Pi, a mini-PC, or a VPS.
  • Community. Far more people have hit your problem before.
  • Maturity. Umbrel has been shipping for years. deputyOS is at v0.3.0.

That last one deserves weight. A public preview is a public preview.

Choosing

Umbrel if: one machine, you want breadth, you value polish, you would rather click than configure.

deputyOS if: AI assistants are the actual goal, you run or expect to run several machines, you need to work offline, or you need to verify provenance rather than trust it.

They are not really competing for the same job. Some people run both.

← All comparisons

Search

Search index isn't built yet. Run npm run build to generate it (Pagefind writes to dist/pagefind/).