Skip to content
Guide · concepts

How Septet makes a tune

A Septet tune is grown, not recorded. Seven players read one shared score in a fixed order, a single seed decides every choice they make, and the same seed always grows the same song. Two ideas unlock the whole rack — the band model and the colour language. Learn those two and the rest of the machine is learnable the same way.

The band model

The players are a graph that listens, not a stack of independent tracks. Every player generates its own part, and any player may follow any other and let what it hears shape what it plays — exactly like musicians in a room. There is no hierarchy and no cascade: the only fixed thing is the order the engine resolves them in, motif → structure → chords → drums → bass → arp → melody.

Each player carries one musical concern, and its signature colour is that concern. Here is the band, in roster order — the lit dot is the player's signature hue, and the last line says who it listens to:

Motif — One short cell — a contour, a rhythm — every other player can quote. Listens to no one — it's the seed.

Structure — The song's form and arc: the sections, the build, the cadence that lands. Listens to the motif.

Chords — Key, scale and the chord changes every downstream player reads. Listens to the motif.

Drums — The kit and the drive — the energy and pocket the band rides. Listens to every other player.

Bass — Walks the changes, trading call and answer over the chords. Listens to every other player.

Arp — The busy voice — figuration threading the harmony. Listens to every other player.

Melody — The lead line: its shape, its question and its answer four bars later. Listens to every other player.

Because everyone can listen to everyone, the graph is full of loops — bass hears arp while arp hears bass. The engine settles them with a fixed, cheap run-through: a cold first pass where each player leans on its own idea, then a second pass where the mutual listening closes and the parts lock to each other. No dice, no waiting on it to converge — just the same two passes every time.

That determinism is the point. You don't draw notes; you turn a knob to change what a player is trying to do, and the engine resolves the actual notes in tune and in time. The control below sets the key on the Chords player; the screen beside it reports what the band resolved.

Harmony
KeyD MAJOR
ChordA7 · V
NextDmaj7 · I
.fc-knob[data-bus="harmony"] + .fc-display — the Chords player: one control sets intent, the screen reports what the band resolved.

Listening itself is a control too. Every player except the motif carries a strip of source chips saying who it follows — a lit chip means "I'm listening to this voice", and the chip's LED burns the followed player's signature colour. Here is the bass, following the whole band:

Bass listens to
.fc-source — a followed chip is raised and lit; its LED burns the target's hue, so "follow Chords" lights red and "follow Arp" lights lime. Turn one down to take your own line over that voice.

The colour language

Every control is coloured by the musical concern it answers — never by where it sits. Squint at a panel and it sorts into four families; focus, and the exact hue names the concern. That is why a control can sit wherever it reads best and still tell you, at a glance, whether it touches pitch, motion, relation, or the motif at the centre.

The four families are the squint test: pitch is warm (which notes, the shape of the line), motion is green (how the line moves — how busy, how hard, how loose), relation is cool (how a voice relates to the band and the form), and the motif is gray — achromatic, the ur-material every other concern reads. Ten concerns in all; here they are as their signature LEDs — the live proof, not a printed swatch:

Harmony pitch Contour pitch Call & Resp pitch Density motion Intensity motion Humanize motion Groove relation Listening relation Structure relation Motif nucleus
[data-facet] — the ten concerns and their signature hues: pitch warm, motion green, relation cool, the motif achromatic at the centre.

The brightness of a signal is earned, not decorative: a lit LED is only ever as bright as the music is intense at that bar. The grounds stay warm, low-chroma wood and leather; the colour lives in the small lit signals. Scan the rack as colour first, and the labels only confirm what the hue already told you.

The three voices

Anywhere in Septet, exactly one of three voices is speaking, and each has one job. Once you know which is which, you always know what you are looking at:

Silkscreen — condensed caps, tracked and printed on the metal — is the voice of the machine's labels: faceplate legends, nav, buttons. Phosphor — glowing tabular monospace on a dark screen — is the voice of values: if it is a number the machine knows, a tempo, a seed, a chord, it glows. And this serif is the third — the reading voice, the liner-note prose you are in right now.

Arp
Notes/bar12
Gate58%
One control, three voices — the silkscreen label names it, the phosphor screen speaks its value, and this serif is the reading voice carrying the explanation.

The control reference

Everything else on the rack is a variation on the two ideas above — a control coloured by its concern, reporting on a phosphor screen. Here is the vocabulary, at the depth a first read needs; each piece behaves the same way once you know the band model and the colour language.

Facets

A faceplate is a grid of facets — one knob, switch or screen per leaf of a player's configuration. The facet's concern is its colour (the ten above); like-concern controls line up across the seven panels even when each panel lays itself out its own way. Turn a facet and you change intent; the screen reports what the band resolved.

Seeds & re-roll

A whole tune is a few bytes: a per-player seed plus the settings. The seed is a phosphor figure you can read and keep; re-roll spins a fresh one for that player while every other player holds, so you can chase a better bassline without losing the drums you liked.

0x1A
.fc-seg + .fc-btn--reroll — the seed glows as a phosphor figure; re-roll draws a new one. Same seed, same tune, every time.

Transport

The global head of the rack: play, stop and loop, the live tempo and bar readouts, and the randomize-everything controls. It is where you start and stop the band as a whole, and where the phosphor tells you where in the tune you are.

Mixer

One channel per player — level, mute and solo — for balancing the seven voices once the parts are settled. The mixer changes how loud a player is, never what it plays; that is still the faceplate's job.

Sequencer

A read-only view of the song laid out in bars: the sections the Structure player drew, and the notes each voice resolved, scrolling under the transport as it plays. You read the score here — you don't edit notes by hand; you steer the players that wrote them.

The listening strip

The band model, made operable on every panel: the row of source chips you met above, plus a master listening knob (Attend Blue) that scales how much this player follows the others at all. Turn it down and the voice goes its own way; turn it up and it locks tighter to the band.

The advanced drawers

Each panel keeps a headline of the controls you reach for most, and folds the rest into an Advanced drawer — the per-source, per-dimension listening weights, the finer facet trims. The rack stays calm by default; the depth is there the moment you want it.

Model
Seven players · one shared score graph · a bounded two-pass fixed point
Seed
One per player, held fixed — a tune is a recipe, not a recording
Colour
Ten musical concerns → ten signature hues, in four families
Voices
Silkscreen labels · phosphor values · this serif, the reading voice

Ready to throw the power yourself?