gzl.dev/blog
field notes, mostly
18 posts
August 27, 2026
7 min
Nine ways to say yes
I taught a Light Phone to speak an Oura ring's protocol over BLE. Then I spent a week failing to tap a button that doesn't exist.
Light PhoneBrightOuraBluetoothAndroid
August 21, 2026
6 min
Field notes on LightOS
What the Light Phone III's OS actually is underneath: Android with the Google removed, hardware that talks in key events, and conventions nobody voted on.
Light PhoneLightOSAndroid
August 21, 2026
1 min
A tape recorder, not a voice recorder
BrightRecorder: clips named for where and when, tapes that are just folders of WAVs, and the wheel finally doing the job it was born for.
Light PhoneBrightRecorder
August 18, 2026
2 min
Why I put an app store on a phone designed to have none
Yes, I see the irony. The argument for building it, why there's no server behind it, and the afternoon I accidentally dropped everyone's apps from the index.
Light PhoneBrightMarket
August 14, 2026
2 min
A good fork says thank you
A lot of what I ship is other people's work with my name nearer the top than it deserves. Some rules I try to fork by.
Light Phoneopen source
August 11, 2026
2 min
My bug reports file themselves
Shake the phone three times and the app files its own GitHub issue. Then an AI triages the queue every hour — under strict instructions that guessing is worse than waiting.
Light PhoneAItooling
August 8, 2026
2 min
Fifteen copies of the same file
The least interesting code I maintain is the code I'd save first in a fire. On finally building the shared library.
Light PhoneKotlin
August 8, 2026
2 min
BrightMarket is out
The app store ships: browse on a desktop, install by QR, track updates on the phone, and get your own app into the catalog.
Light PhoneBrightMarket
August 7, 2026
2 min
One column of scores
Twenty-two leagues from four keyless public APIs, in one grayscale column. How BrightSports gets scores with no account and no push.
Light PhoneBrightSports
August 6, 2026
3 min
The MetroCard inside my phone
I dissolved a transit card in acetone and moved its nervous system into the battery compartment. Tap-to-pay is back, sort of.
Light PhonehardwareNFC
August 6, 2026
2 min
Every button does what you tell it to
The follow-up to the keylayout post: BrightControl itself, why the defaults took longer than the code, and the trust problem with accessibility services.
Light PhoneBrightControldesign
August 5, 2026
1 min
Roll is out
The camera app ships: a filter wheel on real hardware, purikura with four-shot strips, date backs, and a privacy lesson from launch day.
Light PhoneRoll
August 4, 2026
2 min
The keylayout file Light left behind
The scroll wheel is just key events. Everything I learned reading Generic.kl off the phone, and the accessibility service that became BrightControl.
Light PhoneAndroid
August 2, 2026
3 min
Rewriting the camera
Roll's whole design is one gesture, stolen from a contact sheet. Plus the filters that got away from me — purikura on a monochrome phone, and datamoshing experiments that mostly produce mud.
Light Phonedesign
July 29, 2026
1 min
A QR scanner, and its desktop twin
The modern world hands you addresses as QR codes and the Light Phone couldn't read them. A scanner for the phone, a generator page for every other screen.
Light PhoneLightQR
March 22, 2026
1 min
BasilNet, a year later
What broke, what I'd trim, and how the media server quietly became infrastructure for everything else I build.
Dockerhomelabself-hosting
March 8, 2026
1 min
I built a bad DSLR on purpose
It took worse pictures than my phone. It's one of the most useful things I've ever built.
hardwarephotography
January 10, 2025
1 min
Building BasilNet: My Home Server Stack
How I built a 15-service media server with Docker Compose, Gluetun VPN, and Home Assistant integrations.
Dockerhomelabself-hosting
← all posts
March 22, 2026 · 1 min

BasilNet, a year later

Dockerhomelabself-hosting

A year ago I wrote up BasilNet, my Docker home server stack, with the confidence of someone whose setup had been stable for almost three weeks. Time to report back on what survived.

What broke, and what it taught me

The best bug of the year: qBittorrent and the arr apps run inside Gluetun’s network namespace, so their traffic can’t leak around the VPN. Great for privacy. It also means they don’t live on my LAN — and one day I needed one of them to talk to Plex, which runs in host network mode, and nothing worked. The containers couldn’t reach the server’s own LAN address because, from inside the VPN namespace, that address effectively doesn’t exist.

The answer took an evening I won’t get back: go through the Docker bridge gateway. From inside the namespace, 172.18.0.1 is the way home. It’s obvious in retrospect, which is true of every networking problem I have ever had. I wrote it down in the server’s own docs so future me doesn’t get to enjoy discovering it twice.

Also in the spirit of honest reporting: I switched VPN providers along the way for boring price reasons, and the migration was one env block in the Gluetun config. That part of the design earned its keep.

What I’d do differently

Fewer services. The stack at its peak ran more than fifteen containers, and the honest usage graph says maybe eight earn their memory. There’s a homelab failure mode where you start running software because you can, and the maintenance quietly becomes a second job with no salary. I’ve been slowly trimming toward the set that gets used weekly: Plex, the arr apps behind the VPN, the request manager the family actually touches, Home Assistant.

Overseerr remains the surprise winner of the whole project. Nobody in my life wants to talk about Docker, but everybody can ask for a movie.

The job it grew into

The thing I didn’t predict: BasilNet became infrastructure for my other projects. Every Light Phone app I ship can back itself up, and those backups land here — one agent, one container, one file per app. The server that started as “I want a media server” is now the quiet destination for a phone’s worth of notes and photos.

That’s the actual lesson of the year. A home server’s value isn’t the stack you install in the first excited week. It’s having a machine of your own, always on, that every future project gets to assume exists.