There's a running theme in developer hardware right now: drop serious money on an AI workstation, or get out of the way. Beefy GPU towers, mini PC clusters, subscription stacks to match.
Meanwhile, my daily driver is a Raspberry Pi built into a mechanical keyboard. And honestly? I've never been more productive on a machine.
Am I torturing myself, or am I onto something? Let's find out.
From a printer farm to a daily driver
I've been messing with Raspberry Pis since the 3B days. At one point I had a 3D printer farm running with about 16 Pi 3Bs driving all of my printers. I loved those little turds. I've upgraded with every generation since, mostly to keep answering the same question: how much can I actually squeeze out of these things?
For years, my Pi 400 ran my self-hosted Gitea server — and quite a few other services. It's *still* running them, and my son uses it as his daily machine now. Yes, it had some lag. It's a keyboard-sized computer. But it was a blast to tinker with, and it ran full-time services without complaint.
Enter the Pi 500+
When the Pi 5 came out, I figured a Pi 500 couldn't be far behind. Then Raspberry Pi went further and announced the Pi 500+ — and it's a genuinely different beast:
- 16GB of RAM
- 256GB NVMe SSD, stock
- A mechanical keyboard with replaceable, per-key RGB backlit keycaps
I pre-ordered it the day it was announced (at the original $200 price — memory costs have pushed the price up several times since, so I got in at the right moment) and it has been my main device ever since. This thing is a powerhouse. For a potato.
An Omarchy-inspired desktop, because there wasn't one
I've slowly built my own Omarchy-inspired desktop install for it. When I started, Omarchy didn't have an official Pi variant — and to be clear, it still doesn't; the official installer targets x86 only. So this is my own take on the same idea: lean, keyboard-driven, fast.
I'd come back to the project in waves and got it to a stable daily-driver point, but with pain points. Then Omarchy 4.0 ("Quattro") landed, with its rewritten desktop shell and the whole "this is what a lean desktop should feel like" energy — and it lit the fire again. I knew this machine could be quick for a true potato. I just had to put in the work.
Replacing my default apps with my own
I had used Zed on the Pi from time to time. Zed's Linux docs warn about Vulkan driver issues on some machines — and my old MacBook Pro absolutely shows them. But on the Pi's newer hardware? Nothing. Zed runs clean.
With Zed working nicely, and already using Rust with Raven (my agent harness — the whole story is here), I set out on a mission: replace all of my default apps with my own Rust variants built on gpui-kit, worthy of a Raspberry Pi.
None of these need to be "power user" applications, either. My son uses my old Pi 400 and loves it, so everything gets tested on both devices. If a 4GB Pi 400 runs it happily, it ships. Then as my son or I find bugs or issues we file and my agents and I implement later.
Here's what's built so far:
- Piwrite — simple markdown writing
- Picalc — system calculator
- Pifile — basic file manager (more planned updates, but slow going with everything else going on)
- Pitype — typing practice and learning
- Pimon — Pi-specific hardware monitor
- Hearth — agent management platform that hosts my Raven agent and Grok Build
And the idea backlog:
- Pibook — bookshelf application
- Picam — snapshots and video recording
- Pikids — a kid-safe desktop suite, re-imagined for the Pi (yes, that means Minesweeper and friends)
- Whatever else a barebones OS needs
For anything that already exists, I'll fork it if I can bend it to match my app structure and theming. If not, I have my own template ready — theme, base app, and repo scaffolding — so every app in the suite builds the same way.
Why am I doing this?
Because I wanna. First and foremost. And with the current state of AI-assisted development, there's no excuse not to manage it.
Raspberry Pi OS is great — genuinely. But there's always room for improvement, and a machine with 16GB of RAM, an NVMe drive, and a mechanical keyboard is too nice to sit there running somebody else's defaults.
The honest downsides
The web. That's it. That's the list.
But seriously — Chromium has been the absolute slowest thing on this device, and it's often not even Chromium's fault; some websites are just absurdly CPU/GPU-intensive. X (the site formerly known as Twitter) drags the whole machine down, even as the only tab. GitHub isn't as bad. My portfolio and blog run great — though they're due for a refresh, but that's neither here nor there. Apple's website? Don't even think about it on a Pi.
Meanwhile my shader-heavy new-tab extension, Aurora Flow, is dialed in and runs great. The web is a choice, and apparently most of it chooses violence.
AI on-device is a no — obviously. I've run small offline models on a Pi before (I wrote up the whole setup), and I still keep local models on my home server for when I need them. But for day-to-day agent work I use Ollama Cloud for the heavy lifting, and Ollama Cloud + Raven on this machine works beautifully.
Build times — solved, mostly. With the mold linker, I can rebuild and run tests on the Pi without it turning into a 30-minute ordeal. Rust on a Pi is genuinely pleasant now.
And if gpui ever becomes a problem, Omarchy's ecosystem is full of very efficient C++ apps I can fall back on. So far, though, the tests have been wildly promising on system resources, responsiveness, and overall usage.
So, am I torturing myself?
No. I'm daily-driving the most interesting computer I own.
The Pi 500+ cost a fraction of an AI workstation, sips power, sits quietly on my desk, runs my self-hosted stack, hosts my agent platform, and now runs an entire app suite I built myself. My kid learned on the previous one. The whole thing is fast *because* I paid attention to it — not because I threw hardware at the problem.
That's not torture. That's the point.
FAQ
Isn't a Pi too slow for real work?
The Pi 500+ with 16GB and an NVMe SSD handles my actual workload — writing, file management, agent tooling, even Rust builds with mold. Heavy browser sites are the weak spot, and that's a web problem, not a Pi problem.
Can I run Omarchy on a Raspberry Pi?
Not officially. The official Omarchy installer is x86-only, and there's no aarch64 build yet. My setup is Omarchy-inspired — same philosophy, my own implementation.
Why Rust and gpui-kit instead of just using terminal apps?
Because I want GUI apps that feel native and stay light. gpui-kit gives me a consistent component library to theme once and reuse across the whole suite. Terminal-only would be the easy road, but the Pi 400 proved my kid will use a real desktop if it's a good one.
Can you run AI on a Raspberry Pi?
Small models, yes — see my offline LLM on a Pi 5 guide. But for serious agent work, I lean on local models on my home server or Ollama Cloud.
What's next for the app suite?
Pibook, Picam, and Pikids are on the backlog, plus filling out Pifile. Everything ships through the same template so the suite stays consistent.
Where's the code?
Everything lives on GitHub and mirrored to my gitea in both directions so I can use my personal when GitHub inevitably goes down for the day.