9
days of building
YT LIVE STREAMERYT LIVE STREAMER
YT Live Streamer is one tool in a much bigger story — the build-out of an atmospheric water harvesting dome, and the channel documenting every step.
AtmosFarm follows the journey of designing, building and testing an ambitious atmospheric water harvesting dome — a large-scale engineering project that extracts water from the air using Venturi principles, airflow management, cooling systems and renewable energy.
The dome is more than a single invention; it's the foundation the whole channel is built around. Viewers see every stage — rough ideas, prototypes, real-world testing, failures, redesigns, and eventual deployment.
A project this size needs many skills, so the channel naturally expands into the areas that feed it.
Designing components, building prototypes, and solving the technical problems the dome throws up.
Using the property as a testing ground while documenting daily life, breeding and expansion.
Building the tools and software that support both the channel and the future projects — including this app.
Vehicles and portable systems for filming, livestreaming and working on the project anywhere.
Letting viewers follow the projects in real time — the reason YT Live Streamer exists.
Skateboarding and personal interests — showing the person behind the projects, keeping it real.
YT Live Streamer — the desktop app, all 19 plugins, the Android remote, the self-hosted player, the WordPress plugin and this entire website — was designed and written in conversation with Claude AI. Every feature, bug fix and design decision came out of a back-and-forth, from a single-file prototype to a 14,000-line plugin-based application.
It wasn't a straight line. Before this, I spent three nights testing remote desktop setups to get a workable development environment, and had two failed attempts building this with ChatGPT before the approach with Claude stuck. The whole point was to test AI-assisted development as a workflow — a trial run for the bigger AtmosFarm build ahead.
Around June 16–17 the way this got built changed. Up to here, everything was made in conversation with Claude in the browser chat — which was brilliant for designing features and writing the code. But once it came to debugging the app live on the machine — reading the real files, running commands, watching the actual errors, restoring backups — Claude Code turned out to be far more effective, because it works directly on the local project instead of relying on copy-pasting terminal output back and forth. That switch is what unblocked the final stretch: the corruption repair, the playlist fix and the 3.0.0 release.
The real development arc, reconstructed from the build record.
Setting up and testing remote desktop environments to get a reliable place to build from.
Earlier tries at building the app with a different AI didn't come together. The lessons carried into the next attempt.
Starting from a v1.0.3 base — a single-file ffmpeg streamer — through the early download and playlist features. The browser-tab and add-to-playlist experiments were tried and reverted here.
Webcam and capture-card input, picture-in-picture, the big settings overhaul, reliability and automation, watermarks and scheduling — up to the v2.0.0 feature-complete release.
YouTube Search, the My Channel OAuth integration, the Remote Control API, and the companion Android remote app. Mobile screen-share was built (v2.0.4) and later set aside.
The turning point: every feature rebuilt as a standalone plugin. The app became a small core plus drop-in .py files — the structure everything since has used.
Overlays, recording, chapters, trimmer, multi-RTMP, the AI playlist builder, stream preview. Plus a long fight with tkinter segfaults — solved with lazy loading and stripping unicode from the UI.
The SFTP manager, the self-hosted player website v2, and the start of this promotional site — built page by page with the same terminal aesthetic.
Broadcast control and the analytics dashboard land. Help and the debug checklist move to the website, a full help center and the complete version history go up, and the Windows Phase 1 build appears. A file-corruption issue that was stopping tabs and plugins from loading was tracked down and repaired, and plugin loading was made reliable again across restarts.
Local recording arrives via a MediaMTX relay — the app can now capture an exact copy of the stream while it broadcasts to YouTube, with no risk to the live feed. It started life as a proven manual setup, then got packaged up for everyday use.
A long-standing bug where the playlist looped the first video instead of advancing through the list was tracked down and fixed — videos now play in order and the playlist loops as a whole. The relay recording moved from a manual setup into a proper in-app plugin, and the dated built-in file pickers were replaced with the native system file chooser across the whole app, with a safe fallback. It all rolled up into v3.0.0 — a major release, with the README, changelog and website brought up to date.
The streaming software is the first tool in the kit. The dome is the real project — and the whole thing is documented as it happens.