09:40
Deep work
Two hours of concentration with nothing in it that asks for your attention. No vocals. No build, no drop. No track ending at the exact moment you finally got somewhere.
liner notes · what it's for
Put it on and forget it's there.
14:20
Reading
It sits underneath a page the way rain does, except it is harmony instead of hiss. Each chord holds for about sixteen seconds, so nothing ever pulls your eyes up off the line.
23:40
The day has finished with you
But your head has not noticed yet. Choose undertow: fifty beats a minute, a minor key, violet light. Pull tone down until it is all soft low end, and stop keeping track of it.
or nothing
Something to look at
Press f for full screen and h to hide the interface, and watch the colour move. Click anywhere and a ripple runs out through the field from the point you touched.
the alternative
Why not just put on a playlist?
A playlist is a finite list of recordings, and some part of you always knows it. Tracks end. The next one is mixed louder. A song you love comes on and you stop working to listen to it. And when the list runs out, something starts guessing on your behalf.
Zenthysis has no list. There is no next track, because there is no track. The music is being composed and synthesised in the moment you hear it — the engine steps through its timing grid about eight hundred times a second — and there is no recording anywhere inside it to return to.
So there is nothing to choose, nothing to skip, and nothing to set up before it sounds good. Open it, turn it down, and go back to work.
the one choice
Four moods. Pick one and leave it.
Each one is its own key, tempo and set of chords. Choose the band that matches the hour — the colour of the picture follows the harmony, so the screen changes with it.
Every chord in all four is an extended voicing — ninths, elevenths, sharp elevens. There is not a plain triad anywhere in the file, and that is most of why it sounds expensive rather than thin.
Here is the honest part: those four chords are fixed, they are written out in the file rather than generated, and they come round in order roughly once a minute. The surprise is never which chord arrives next. The surprise is everything played over the top of it, and none of that is written down anywhere.
the sound, in four controls
Four sliders, in plain English.
silent — this is the picture only
dusk
the renderer, lifted from the engine
click to send a ripple
the field is drawn by the engine's own renderer, lifted into this page and fed a slow pulse instead of the music this page makes no sound
densityhow much is going on
tone520 hz → 9,880 hz
feelambient
texturedesigned
feel and texture are live — drag them. on this page they move the picture above and nothing else, because this page makes no sound. once it is playing, the same two sliders change what you hear.
the part people repeat
There is a beat. There is no drum.
Leave Zenthysis running and you will start to feel a pulse. Go looking for the drum making it and you will not find one. There is no kick, no click and nothing in the file keeping time.
What you are feeling is the sustained layers stepping out of the way. It is the same breath a mix takes around a kick drum, except the kick was never there. The gap is the beat.
And because that breath lands on the metre's own accent pattern, the music can play in seven-eight and it grooves as 2 + 2 + 3 instead of limping through seven flat beats. There are thirteen time signatures in here, out as far as nineteen-sixteen, and every one of them carries its own accent pattern, written out rather than derived.
2
2
3
2
2
2
3
2
2
2
3
3
3
3
2
3
2
3
2
3
4
4
4
3
4
4
4
4
3
filled dots are where the music breathes. this diagram is silent.
if you're curious
It is not playing anything back.
Nothing goes into Zenthysis as audio. There is not one sound file inside it: every sound is built out of arithmetic while you listen — including the reverb, which is a seven-second room computed from scratch on every visit, while the start screen is still up (a smaller one on a device that says it has less to spare), and the hiss underneath everything, which is noise generated in a loop rather than a recording of any tape that ever existed. The files a take writes out run the other way. They are made from that live sound; nothing is ever read back in.
You never need to know any more than that. The rest of this is here because you might like knowing that somebody cared.
3.7 quintillion
rhythmic configurations
Thirteen layers, each free to take any of twenty-two note values — from a thirty-second triplet to four whole bars — in any of thirteen time signatures. Exactly 3,676,530,752,480,075,776 of them. That is a count of settings, not a promise of that many different songs — and it is before the four sliders touch anything.
the long version
Time is counted in 840 ticks per beat. That number is the smallest
one divisible by 8, 3, 5 and 7, which means thirty-second notes, triplets,
quintuplets and septuplets all land on exact whole numbers. Nothing slips out of
time, because nothing is ever rounded.
Twenty-two note values sit on that grid, from a thirty-second triplet up to four whole bars. Every one of the thirteen layers can take any of them, independently. The pad is allowed to chatter and the ceramic layer is allowed to drone; what stops either becoming a wall is not a rule about what you may choose, but the fact that a layer set to a very fine grid speaks proportionally less often on it, and that the engine stops adding notes once enough are already ringing.
The melodic line is not a random walk. The generator writes a short cell and then almost never plays it the same way twice: it bends a degree, jumps an octave, softens a note, or drops one entirely to leave a hole in the phrase. Rests are what stop music sounding typed.
None of this is a model. There is no training data and no inference anywhere in the file. It is arithmetic, running in your browser — and it has no idea what you like, because it has no way to find out.
- pad
- The sustained bed. Wide, slow, the chord itself.
- bass
- The root, an octave down. Walks a little once the feel slider is up.
- pluck
- The melody, such as it is. This is the layer that carries the phrase.
- bell
- High, sparse, struck. The thing you notice without meaning to.
- air
- Shimmer above everything. Mostly felt rather than heard.
- ceram
- Bit-crushed FM ticks. Texture, never timekeeping.
- grain
- Tiny grains of tone, scattered two or three octaves up.
- reso
- A filter pinged hard enough to nearly sing.
- morph
- The vowel voice. Quiet at low texture, forward at high.
- throat
- A seat with three voices that take turns — formant, talkbox, shift. The one that seems to be saying something.
- metal
- Bowl, spring, gong, struck in turn. Answers throat from the other side.
- warp
- Dive, riser, stutter. Movement rather than notes.
- grit
- Fold, crush, rasp. The dirt, a little at a time.
The last four are seats rather than instruments. Each holds three voices and plays one at a time, changing over after two or three firings, and only a few seats are open in any chord — which ones is redrawn at every chord turn.
You cannot make it again.
There is a record button. Sixty seconds, and what comes out is the mix, every layer as its own file, and the room as another — built on your machine and handed straight to you.
What you cannot do is make that minute a second time. The noise and the reverb are drawn fresh every session and are not tied to the seed, so a take is the only copy of itself that will ever exist — ours included. The seed re-aims the composer. It does not reproduce the audio.
So if something extraordinary happens at 11:42 tonight and you were recording, you have it. If you were not, it happened, and then it was over. Either way you cannot repeat it, and neither can we.
Making the same take twiceimpossible
Recording longer than 60 seconds at a timenone
Per-layer volume, solo, or a mixernone
MIDI or DAW integrationnone
Stems that add back up to the mix exactlynone
Switching mood — lands on the next chord turnup to ~19s
Installable app, offline cache, service workernone
Screen reader support inside the machinepartial
what it takes from you
It doesn't know you're here.
No account. No email address. No sign-up wall three minutes in.
Once Zenthysis has loaded it makes no network requests at all — no analytics, no tracking pixel, no font from someone else's server, no library, nothing phoning home. The machine is one HTML file with everything already inside it; the photograph on its start screen and the short clip it opens on arrive with the page rather than when you press begin, so pressing tells nobody anything. There are no cookies and no accounts, and nothing about you is sent anywhere, because there is nowhere for it to be sent to.
It does remember where you left the sliders. That is written to your own browser's local storage, on your own machine, and it never travels — the same file that has no way to send anything also has no way to send that. Clearing your browser data clears it, and a browser that refuses to store simply gets the defaults.
What it does not remember is you. There is no profile, no history of what you played and no opinion about what you like. It knows where you last left the controls. That is the whole of it.
It is a room, not a service.
specification
The whole machine, in numbers.
Everything here is generated live. There are no audio files inside it.
Composition
moods4
chords per mood4, fixed order
chord hold~15–19 s
layers13
note values22
metres13, to 19/16
rhythmic configs3.68 × 1018
Synthesis
ticks per quarter840
= LCM of8, 3, 5, 7
tone filter520 → 9,880 Hz
reverb7 s IR, computed
mono below300 Hz
master fade-up5 s
audio files insidenone
What it takes from you
network, once loadednone
accountsnone
cookiesnone
analyticsnone
third-party codenone
kept on your machineyour settings
enforced by CSPdefault-src 'none'
questions
Fair questions.
Is it free?
There is no paywall on this page and nothing to buy. No card, no email address, no trial. You press the button and it plays.
Do I need an account?
No. There is no sign-up, no login and no profile. It knows the settings you last chose, because your own browser kept them for you, and nothing else about you at all.
Am I hearing a recording?
No, and not partly. There are no audio files inside Zenthysis — no samples, no loops, no stems, nothing pre-rendered. Every sound is generated while you listen, including the reverb and the hiss behind everything. (Recording writes stems out; there are still none going in.)
Does it work offline?
It needs a connection to load the page and the short clip it opens on, once. After that it makes no network requests at all, so a dropped connection changes nothing you can hear.
It is not an installable app, though. There is nothing to install, and you still need to be able to open the page.
Will it work on my phone?
It plays, and the panel is built for it: a strip along the bottom edge with four faces — listen, shape, engine, piece — one tab or a swipe apart. The first is just what you need to listen: play, mood, volume. The other three stay out of the way until you ask.
Nothing is keyboard-only any more. It is still at its best on a laptop with headphones, because a bigger screen shows more of the picture at once.
Do you collect my data?
Nothing at all. No analytics, no cookies, no telemetry, no error reporting, no account. Once Zenthysis has loaded it makes zero network requests, so there is no channel for anything to leave your machine even if we wanted one. The clip it opens on is fetched with the page, not when you press begin, so the moment you press is not something the server gets to learn.
It does save your slider positions, your layer settings, and the mood and metre you last chose to your own browser's local storage, so the volume you chose last night is the volume you get tonight. That never leaves your machine, and clearing your browser data clears it.
Can I save one I like?
Yes — as audio, and as the settings that made it.
The record button takes sixty seconds and writes them out as files: the mix, every layer that spoke on its own, the room as a separate file, and for a few layers the reverb belonging to that layer alone. A notes.txt travels with them, carrying the licence, what each file peaked at, the seed and a short settings code. They are built on your machine and handed straight to you. Nothing is uploaded, because there is nowhere to upload to.
Put that code after the # on a Zenthysis link and the same seed and settings load back. That part of a URL is never sent to a server, so you can hand somebody a session without anything about it leaving either machine.
The honest limit: a seed re-aims the composer, but it does not reproduce the audio. The noise and the reverb are drawn fresh every session on purpose, so even from a clean start the sound is a near miss rather than the same thing. If you want the exact minute you heard, record it — that file is the only copy there will be.
Can I use what I export?
Yes — in your own work, anywhere, including work you are paid for. You do not need to ask, and you do not need to credit Zenthysis. The one thing it does not cover is selling the files on as sound or sample packs.
What a take writes out is a one-off: the noise and the reverb are drawn fresh every time and are not tied to the seed, so that file cannot be made again by anyone, us included.
The honest limit: nobody can promise an automated copyright system will never flag a track. Those systems match against whatever has been registered with them, and Zenthysis has no say in what anyone registers — including someone who releases a take of their own.
How long can I leave it running?
As long as you like. It has no length and no end. Each layer stops adding notes once enough are already ringing, so the note count cannot run away, and there is a live count of sounding voices on screen if you are curious.
Does it work with a screen reader?
Better than it did. All six sliders are real range controls now — they take keyboard focus and respond to arrow keys; five announce their value in words rather than numbers, so feel reads as "phrasing" instead of "55", and volume announces decibels. Every layer control has a name and an on or off state instead of thirteen buttons that all said the same thing. Zenthysis honours your system's reduce-motion setting, and the start button takes focus.
The honest remainder: the panel is dense, the interface text is small, and the spectrum display is decorative and marked as such. We would rather tell you here than waste your time.
What is Zenthysis?
The name of this machine and the site it lives on. The honest part: the name was the easy half. The machine is the half that had to work.
The voices the texture slider brings in were aimed at measurements taken from the records of the person behind Zenthysis, who chose by ear from there.
Those records are the target. The machine has never heard them — none of that audio is in here, and none of anyone's is.
You can hear them, though: the records are here, released as Omnitemporal.
Nothing is playing yet.
Press it and something begins that has never existed and will never exist again. It comes up out of silence over about five seconds, it asks you for nothing, and it runs for as long as you leave the tab open.
headphones recommended · it never plays the same thing twice
silent