Kontinuum stores music as code, not as a recording — and a deterministic engine performs that code live, in your pocket, forever. Every bar you hear was written moments before you heard it. It never stops, never plays the same session twice, and never needs the internet.
A recording stores decisions already made. Kontinuum stores the system that makes them: patterns, instruments, automation, arrangement and mixing as one structured document, carrying its own rules, probabilities and transformations. So the same composition performs differently every time and stays recognizably itself — and none of it is random. Same seed, same audio, sample for sample.
One performance, frozen. Copy bar 1 to bar 33 and bar 33 is a copy. Every listener, every play, gets the identical take.
bar 01 [kick.wav ][hat.wav ]
bar 02 [kick.wav ][hat.wav ]
bar 03 [kick.wav ][hat.wav ]
The behaviour. Every bar is derived, deterministic and different — and you edit the behaviour, or edit one bar and have that edit survive.
hat: euclid(5,16,2)
.every(4, rotate(3))
.sometimes(.25, degrade)
.humanize(6ms)
Generative is architectural here, not a button. It is a property of the document, which is why it keeps working with the AI switched off. The architecture →
Not a better version of any of these — a different thing. Streaming hands you a finished recording, prompt-to-track hands you a finished file, adaptive apps recombine finished stems, and a DAW hands you the tools and the years. Kontinuum hands you a composition that is still being written.
| Spotifystreaming | Sunoprompt to track | Endeladaptive audio | Abletona DAW | Kontinuummusic as code | |
|---|---|---|---|---|---|
| What you get | A catalogue of recordings | A finished file from a prompt | Adaptive soundscapes | The tools to make it yourself | Music being composed while you listen |
| Same on every play | Yes — it is one recording | Yes, once generated | Shifts with time and context | Yes | Never — and never random |
| Personal to you | The picks, not the music | Only what you typed | Your context, not your taste | It is your own work | The music itself BUILDING |
| Works with no internet | Downloads only | No — generation is a server | Yes | Yes | Yes, forever |
| Can you open what you hear | No | No — regenerate instead | No | Yes — your clips and edits | Yes, and it reads as code BUILDING |
| To get started | Press play | Type a prompt | Press play | Months to years | Press play — the depth is optional |
| Where the artist sits | Paid per stream of a recording | Absent, and contested | Composers hired to make stems | The artist is the user | Publishes a Creative Soul, paid per use |
A comparison of what each thing is, not of which sounds better — that claim we will make when we can show it, in a blind test anyone can take. Rows marked BUILDING are on the roadmap, not shipped. Trademarks belong to their owners; all four are good at what they do, which is a different job from this one.
A recording is one frozen performance, sold identically to everyone. On Kontinuum an artist publishes their musical mind instead — a Creative Soul, free or licensed, performed live inside your session: never the same twice, authored from their rules and choices, and every minute pays them. Not replacing artists — their work as it was always meant to be. The full artist story →
No feed, no autoplay bait, nothing optimizing for your screen time — one objective: the next bar feels right to you. Your taste model lives on your device and belongs to you. You are not pushed. You are heard.
There is no listener version and no musician version. Listen, DJ and Produce are three projections of one running composition — press play, perform it, or open the code that is writing it. Going deeper never means switching apps, because there is nothing to switch to: it is the same document all the way down.
Pick a style, press play. A waveform breathing with the music, nothing to learn.
A beat-locked performance row. Every move is an edit to the running document, quantized.
The DAW surface over the source. Drag a note or edit the line — same document, same undo.
Zero learning curve: pick a style, press play, live your day. This is the whole app for most people — and it quietly becomes yours. The full listener story →
Connect Spotify or SoundCloud — Kontinuum extracts your taste: groove, kick character, harmonic vocabulary, energy curves. No two people hear the same Kontinuum.
"I'm stressed." "More energy." "Sleep mode." Quick moods land within four bars; deeper asks recompose the arrangement. Every reaction quietly teaches it what you love.
Double-press: somewhere new. Triple: calmer. Or just ask Siri — steering never needs the screen on.
Kontinuum has the stems, so every instrument becomes a head-tracked object in space — kick ahead, percussion scattered, textures overhead. And AirPlay to any room, with the interface locked to what you actually hear.
Point it at a track in your own library. Kontinuum analyzes structure, harmony, groove and the character of every stem; the composer picks the instruments to play it — or designs new ones — and hands you a new, editable session derived from its musical characteristics. Analysis only: nothing is copied, sampled, or imitated note-for-note, and the audio never leaves your device.
Kontinuum is a musical runtime: songs, performances, artist identities, styles and instruments are all executable musical state inside one engine. The core ships tiny; everything else arrives as a pack you browse and download inside the app. No plugin can break the music; no missing plugin can silence it.
Classic machines with exact character, anyone's AUv3 plugins — or brand-new instruments the AI designs as validated patch graphs. Shareable data, played by the same engine.
Like an LLM loads a skill, Kontinuum loads a musical identity — a producer's sound, a scene, an era. Blend them with weights, switch eras mid-session, learn one from a handful of tracks, share it.
Sample kits are deterministic recipes the engine renders itself, mid-session, in under a second. Every genre is a pack; new styles are downloads, never app updates.
If music is code, the code that performs it should be readable too. The engine, the musical language, the app — MIT-licensed, built in the open at github.com/Kontinuum-ai/kontinuum. The code that meters an artist's royalties and the model that learns your taste are inspectable by anyone: trust here is not a promise, it's something you can read. Fork it, build instruments for it, publish styles for it — or just watch every commit land.
This page describes a target. Here is the honest split, so nothing above reads as a promise it should not. Everything marked shipped runs today; everything else is scoped in the open and you can read the plan.
In a blind A/B test, an ordinary listener should not reliably distinguish a Kontinuum session from a professionally produced and mastered track sold on Beatport.
The beta ships when the sound is ready — the roadmap is ordered by that, not by features. Leave an address and you'll hear the day it does.