JeliaWhat it is
JellyDazzle is an audio-reactive music visualiser for macOS on Apple Silicon. Put on a record, a film or a livestream, open the app, and it turns the sound into moving colour and shape. On recent versions of macOS it listens to the system output, so it follows whatever app is making the sound.
It is a homage to DAZZLE.EXE, a much-loved visualiser from an earlier era of personal computing. The site carries a page to its original creator. The source is on GitHub under the MIT licence, and the app is available to download from the site.
Who it is for
Anyone with an Apple Silicon Mac who likes music and something to look at while it plays. There is nothing to configure. Two keys drive the show: one changes the colour, one changes the shapes, and both fade rather than cut, because nothing here strobes.
How it works, in plain words
Each frame is built from up to four layers: a ground that fills the screen, a field over most of it, a figure that drops distinct shapes on top, and a spark that adds glints. They blend by brightness, so the dark parts of one layer let the layer below show through.
Each layer runs for a turn of its own length, then a new routine is drawn at random and fades in. At startup the engine measures every routine and decides which layer it suits by how much of the screen it actually covers. Timing is random by design. There is no fixed loop.
What I built, and how
- AI-guided, not AI-generated. I set the direction, judged every result and decided what shipped. An AI assistant did much of the typing, in a language most people would not attempt by hand.
- Measured, not guessed. Problems are reproduced on a test harness and measured before they are fixed. One release fixed a layer that, measured, was appearing in none of the frames it should have.
- Honest release notes. The site says plainly what went wrong in earlier downloads: it would not open on older Macs, it looked frozen on first launch, patterns repeated. Each is explained and marked as fixed.
- Shipped properly. The download is code-signed and notarised by Apple, so it opens without warnings, and it runs on Macs older than the one that built it.
What it shows I can do
- Use AI coding tools at the hard end: low-level code where a mistake does not throw a friendly error.
- Take a product from idea to a signed, notarised release that installs cleanly on someone else's machine.
- Debug by measurement, and write down what went wrong in language a user can follow.
- Keep a long-running project organised: a public changelog, open source, and a clear note of what is finished and what is not.
The same discipline applies to any software a client depends on. Know what it does, measure it when it misbehaves, and tell people the truth about it.
Questions people ask
Which Macs does it run on?
Macs with Apple Silicon. Listening to the system's own audio output needs a recent version of macOS; older systems can still run it.
Do I need to play music first?
Yes, ideally. In silence it keeps running on its own internal clocks, but you see perhaps half of what it does.
Why assembly?
Partly as a test of how far AI-guided coding can go at the lowest level, and partly as a homage to the original. The patterns themselves are a mix: the engine modes are assembly, the large library of plug-ins is C.
Can I see the code?
Yes. It is on GitHub under the MIT licence, linked from the site.