How it works
This page covers two things. The first is a guided tour of the controls so you know what each button does. The second explains how the engine composes a track from scratch.
Pick a mood
Each card on the splash represents a separate compositional world with its own scales, chord progressions, synth voices, and rhythmic feel. The tag under each name (Deep Ambient, Baroque, Lo-fi Chill, and others) hints at the territory. When you want a sample of the whole catalog, press Radio and the engine will shuffle moods at random, switching to a fresh one each time the current track ends.
Press play
The play button generates a brand-new composition in the selected mood. Each press of New Composition produces a fresh track with new chords, a new melody, and a new seed. The track title shows the key, scale, and BPM so you always know what you are hearing.
Three listening modes
Single mode plays one composition and then stops. Flow mode chains a fresh track in the same mood as soon as the current one finishes, which is useful when you have found a vibe you want to stay in. Drift mode behaves as an endless radio, jumping to a different random mood each time a track ends, and is the right choice for long focus sessions.
Save what catches you
Three buttons let you keep what you make. MIDI downloads the score itself, which you can open in any DAW to inspect every chord, voice line, and rhythmic choice. WAV and MP3 download the audio recording at studio quality. Share copies a link that reproduces the exact track because the seed, key, scale, and BPM are encoded directly in the URL.
Tracks panel
The last 25 tracks you generate are saved across page reloads in the Tracks panel. You can replay any of them, redownload the audio, regenerate the MIDI, or share the link again. Everything stays in your browser and never leaves your device.
Mixer
Each track is a layered mix of up to six voices: pad, melody, bass, arp, counter-melody, and drums. The mixer lets you solo, mute, or adjust each voice in real time. When you want only the pad and drums for deep work, pull every other voice down.
A few small things
If you want to switch moods entirely, use the mood switcher in the player rather than returning to the splash. Drift mode is built for long sessions, so it runs comfortably in the background while you work.
How it composes
Every sound and every note is built from scratch. Two ideas, working together, shape each track from the first beat to the last.
- SPEAC, an acronym for Statement, Preparation, Extension, Antecedent, and Consequent, is David Cope's grammar for describing how classical pieces unfold. It gives each track an arc of tension and release rather than letting it churn forward until time runs out.
- Markov chains, trained on an extensive library of real compositions ranging from Bach to modern pop, decide which note is most likely to come next based on what has just been played. They give each mood its characteristic vocabulary.
This project owes a great deal to Brian Eno, who has built generative music systems since the 1990s and coined the term ambient music. He spent decades arguing that music can be a system rather than a recording, and OSCILLØN FAE is a small contribution to that conversation.
Music has always worked algorithmically. The I–vi–IV–V progression took hold with one 1950s hit and went on to power thousands of doo-wop songs. The twelve-bar blues form is shared structural DNA across an entire genre. Pachelbel's Canon appears inside Let It Be, No Woman No Cry, and a thousand other pop ballads. Borrowing recognizable structures is how genres form in the first place.
OSCILLØN FAE makes this process explicit. It learns from the choices of skilled composers and recombines those choices under fresh structural constraints. Every sound is crafted from scratch using oscillators, filters, and envelopes.
Why not just use AI?
Large language models are excellent creative tools, and I use them daily as a pair programmer. The point of this project is narrower. Music has five centuries of codified theory behind it, and you can get remarkably far using math, statistical tables, structural grammar, and a set of opinionated rules. OSCILLØN FAE runs entirely in your browser on your CPU with no inference required. It is cheap to operate, deterministic in its output, transparent in its decisions, and works completely offline.
The composition is open
Modern AI music tools hand you an MP3, sometimes accompanied by stems or MIDI. Those auxiliary files are extracted from the finished audio by a separate model, which means the generator itself cannot explain why it chose a particular chord or where a melodic line came from. The transcript arrives after the fact, reverse-engineered from a black box.
OSCILLØN FAE works in the opposite direction. The MIDI is the source of the composition, and the audio is rendered from it. Every chord, counter-melody, bass line, and rhythmic choice is a deliberate decision made by the engine before a single sample is generated, guided by Markov tables trained on real compositions and by SPEAC tension arcs that shape the piece. The engine is deterministic and inspectable, built from documented rules rather than opaque weights inside a neural network.
Download the WAV or MP3 when you want the recording. Download the MIDI when you want the score. Open the score in any DAW and you can read the entire composition: chords, counter-melodies, bass lines, and timing. The music is composed rather than synthesized, and the score is yours to read, remix, and build on.
Every track is generated from a seed, so a shared link reproduces the exact composition on anyone's machine.
Open the player, pick a mood, and let the engine show you what it can do.