About Golden Karaoke

Golden Karaoke is a social music platform shaped like the chat apps you already use: servers, channels, roles, voice rooms, direct messages. On top of that structure it adds the parts that normally live in four other applications — song discovery through YouTube and Spotify, a karaoke screen with lyrics that follow the song, an effects chain for a real microphone or a real guitar, virtual and MIDI instruments, and a recording vault that starts private and stays that way unless you say otherwise.

The point is not to publish. Most music-making is practice, rehearsal and messing about with friends, and almost none of it deserves an audience. The product is built around that loop — talk, listen, sing, play, record, come back tomorrow — with sharing as something you reach for rather than something you are pushed into.

Two groups, and the same software has to serve both. Someone with an audio interface, a condenser microphone, a MIDI keyboard and opinions about buffer sizes. And someone with a laptop, the headphones that came with their phone, and four friends who want to sing badly on a Thursday. The first group needs routing, latency measurement and preset scenes; the second needs a screen that works on the first try. Neither is the afterthought.

The product is in preview and much of it is still being written. As of the date above:

• Accounts, sign-in and profiles work. You can create an account with an email address and sign back in.
• The Audio Control Center lists your input and output devices, shows live level meters, plays a test tone, measures round-trip latency and monitors connected MIDI hardware.
• Communities, chat, discovery, the karaoke screen, the recording vault and live rooms are in build. Their pages exist and say so rather than pretending.

Nothing on this site is described as finished when it is not. If a page claims a feature works, it has been exercised — not merely written.

The web app is Next.js and React with TypeScript; data, authentication and storage run on Supabase; audio is processed in the browser with Web Audio and AudioWorklets, and hardware controllers talk to it over Web MIDI. Heavier analysis — stem separation, key and tempo detection — is designed to run in separate workers so the main application never becomes a single large audio server.

The codebase is written by two AI agents working in parallel lanes under a human owner, one on the platform and one on the music engine, with every change reviewed before it lands. That is an unusual way to build software and it is worth saying out loud rather than discovering in the commit log. The repository is private while that work is in progress.

• Audio stays on your device. Realtime effects and monitoring are computed in your browser. Raw audio is not streamed to a server to be processed.
• External music keeps its boundary. YouTube and Spotify are for finding music and playing it in their own players. Their audio is never downloaded, captured or fed into processing. See the note on the home page.
• Private by default. Recordings begin visible only to you. Every widening of that is a deliberate act.
• Runs on free infrastructure. Which is a constraint, not a boast: it shapes what gets cached, how much gets stored, and why some features arrive as adapters waiting for a budget.

There is no support address yet. While the project is in preview, questions and problems go through the issue tracker on the repository, which requires access to the private repo. A public route will exist before the product leaves preview.