Manual

Getting started

Press play. Browsers keep audio muted until you have interacted with the page, so the first press is really just permission. If you hear nothing, press it again.

A new session opens with a few drum tracks already running. Click any cell in a track's row of squares to add or remove a hit. After that:

• The dropdown next to a track's name changes what it sounds like.
• + add track at the bottom gets you another instrument.
• The numbered buttons above the tracks are 32 pattern slots.
• Right-click anything you don't recognise. You get what it does and whatever movement is on it.

Transport

The bar across the top runs playback for everything.

The second row is the keyboard strip: scale, octave, chord and arp, covered under playing from your keyboard. The status line at its right end reports what just happened.

Above the transport sit new, share, and, once you are signed in, save and songs. See saving and sharing.

Knobs

Nearly every parameter is a knob, and they all behave the same way.

On a phone the same controls draw as vertical sliders and the thumb follows your finger. Two rows stay sliders everywhere: the wavetable editor's harmonic bars and the chance generator's pitch faders.

Patterns

Every session has 32 pattern slots. Click a number to switch to it; the one you are editing is outlined. Empty slots are blank, so one slot per section is an easy way to build an arrangement.

Tracks

A track is one instrument plus its pattern. The header row has its name, its engine, the pattern length and the mixer controls.

The last row of buttons (roll, filter, env, fx, eq, comp, mod, aut) opens the panels described under filter, effects and dynamics and modulation and automation.

A button wears a dot with a count when something inside it is switched on, and hovering it names what. Whatever is on also shows on the track itself: engaged effects, enabled LFO rows, a filter you have closed, an eq band you have moved. Open the panel and you get all of it again.

The step grid

The row of squares is the pattern. A filled square is a note, and every fourth one is marked so you can find the beat.

• Click an empty square to add a note, or a filled one to take it away.
• Drag right from a note to make it longer.
• Drag up or down on a note to change its pitch.
• Double-click to set a note to full velocity.
• Right-click, or press and hold on a touchscreen, for the step editor.

On drum tracks the pitch hardly matters, so the grid is usually all you need. Melodies are easier in the piano roll.

The piano roll

roll opens a pitch-by-time grid: pitch up the side, steps across. It is the best place to write a melody, and the only place to stack notes into a chord you voice by hand.

The lane underneath is velocity: drag a bar to make that step harder or softer. The buttons along the top shift every note up a step, double or halve the pattern length, or roll a new melody.

The step editor

Right-click any step for everything it can do beyond pitch and length.

The three generators

Three buttons on the track row fill the grid for you: the dice rolls a part, the ring divides the bar evenly, the die throws odds. All three can print a result into the grid as ordinary steps you can then edit. The ring and the die also run live, generating as the track plays without writing anything, so their controls can take an LFO, an automation lane or a macro pad; switch live off and the pattern is exactly as you left it. While live is on the grid shows what is playing and goes read-only. A track has one rhythm at a time, so turning one of those two on turns the other off.

The dice: roll a part

One press replaces the pattern with a new one, rhythm and pitches both. Keep pressing until something sticks; nothing is held, so each roll starts from scratch. What lands in the grid is ordinary steps, yours to edit afterwards.

Drag the dice up or down to set how full the rolls come out, drawn as the fill behind the icon. All the way down is empty, all the way up is every step, and downbeats stay likelier than the steps between them.

On a drum-kit track every hit sits on C2 and only the rhythm changes. Anywhere else it walks a melody in the session's key, mostly a step or two at a time with the occasional octave leap, starting from the last note you used on that track. It follows the root and mode even when scale quantize is off. It leaves the sound alone; rand beside the four knobs is the one that rolls that.

The ring: euclidean rhythms

A number of hits spread as evenly as possible over a cycle, which covers a lot of the world's drumming. Set the hits, the cycle length and a rotation, and the cycle tiles across the track. It decides only when notes happen; the pitches stay whatever the pattern already had.

The die: chance

Chance decides the pitches as well. You write the odds of notes rather than the notes, and it plays something that fits them. Two sections, each with its own dice.

Rhythm. note value is the base rhythm, from 1/1 down to 1/32 by way of the triplets, and nothing random happens to it on its own. variation is off in the middle: turn it left for longer note values, right for shorter, and how far you turn it is both how often another value turns up and how far from the base it strays. legato is how likely a note is to be tied to the one before instead of gating again; all the way up you get a single held note. rest is how likely a note is to be dropped. Two checkboxes let variation reach the triplets and the 1/32s, which are off to start with.

Melody. The twelve faders are the probability of each semitone, not a switch on each: one at half height turns up half as often as one at full, and a single raised fader is certain wherever it sits. That is how you write a scale here, or a scale with a bias. from scale loads the session's active scale into them. Generated notes are not snapped to the session scale, because these faders are the scale. low note and high note set the range, up to five octaves; a semitone raised outside it can't play, and its label greys out to say so.

The dice. Each section's roll takes a new throw. A throw is held, so the part repeats and you can play against it. Tick realtime on a section and it stops holding: a new throw every time round, so that section never repeats. Rolling a section twice always changes something; if a throw can't change anything, the button says which knob to turn instead.

The window is the first and last step it generates over, and it tiles across the track like the ring's cycle. Moving first step slides the window without changing its length.

Six controls can be modulated while live is on: the four rhythm knobs and the two range knobs. An LFO on rest breathes the part in and out; a lane on low note walks it up the register over a bar. The picture at the top of the panel is the part it is playing now: height is pitch, width is how long a note is held, and a note in the second colour is struck more than once, which is how the triplets and the 1/32s fit a grid of sixteenths.

Playing from your keyboard

On a desktop machine your computer keyboard always plays the active track, which is the last one you clicked and the one with the outline. Typing in a text box is safe: note keys only fire when no text field has focus.

The layout is a piano. The home row is the white keys, the row above it the black keys.

• White keys: a s d f g h j k l
• Black keys: w e t y u o
• z and x drop and raise the octave.

Two ways to get what you play into a pattern.

• Record. Switch on record, start playback, and notes land on whichever step is passing underneath.
• Capture. Play freely, stopped or running, then press capture. Your last phrase is written into the track with its timing and note lengths, and the pattern is sized to fit it.

Scale and chords

Tick scale, pick a root and a mode, and your playing is locked to that key. The white keys all belong to the scale and the black keys go quiet, so there are no wrong notes. There are major and minor modes, pentatonics, blues, some exotic scales and a few microtonal tunings.

The palette button next to it colours notes by pitch across the roll and the grid, which makes repeated shapes jump out.

Chord mode

With chord on, each key plays a whole chord instead of a single note. Pick the chord type and a voicing. If a scale is active the choice collapses to on or off, because the chord gets built from the scale itself: each degree gets the quality it should have and everything stays in key.

Chords you record or capture are stored as chords rather than loose notes, so the roll shows one labelled note you can retune or move in a single go.

Arp

The arp box sits next to the chord picker. Switch it on and a held key arpeggiates as you play, at the rate, octave range and direction you choose. Anything you then record or capture carries those arp settings into the pattern.

Sound engines

The dropdown beside a track's name picks its instrument. They come in groups:

The four knobs under the name are that instrument's main tone controls: harm, timb, morph and decay on a Plaits model, something else on the next engine. The labels change with the engine, and hovering one says what it does on the instrument you have loaded.

The silverbox

An acid box modelled from its circuits, so it answers a pattern the way the original does. Its four knobs are the panel: cutoff, reso, env mod and decay. Next to them are the waveform switch, the accent depth and tuning.

Three behaviours are how you play it:

• Accent comes from step velocity. Push a step past about two-thirds and it accents: louder, brighter, filter decay forced short. With the resonance up, consecutive accents stack into a rising squelch rather than resetting, and that pile-up is the sound of an acid line.
• Slide comes from note length. Draw a step longer than one cell and the note after it slides out of it: the pitch glides across and the envelopes never retrigger.
• Plain steps get clipped short, a little over half the step, which is what makes a silverbox part drive instead of running together. Lengthen a note if you want it to hold.

The filter thins out as you wind the resonance up, same as the real one. That is why acid records run a silverbox into a distortion pedal, and you can do the same from the track's fx panel.

The contagion

A big polyphonic synth built for movement. Its four knobs are cutoff, reso, shape and decay. The four beside them are the levels of osc 1, osc 2, the sub and the noise. Everything else lives in the three rows underneath.

• Shape is one continuous morph from sine through triangle and saw to pulse, and at the top the pulse width knob takes over. Sweep it and the tone changes character, not just brightness.
• Unison is the hypersaw. Each note plays up to eight detuned copies of the whole oscillator section, spread across the stereo field. Two or three thickens things up; eight with the detune wound on is the wide trance sound.
• Two filters, not one. Each can be a low pass, high pass, band pass or notch, and you choose how they connect: in series, in parallel, or split so filter 1 plays the left ear and filter 2 the right. A low pass into a high pass gives you a band pass you can sweep from both ends.
• Saturation sits between them, so filter 2 gets to tidy up whatever the saturator did. It runs from a gentle warmth through hard clipping to a bit reducer and a rate reducer.

With sync on, osc 2 is forced to osc 1's pitch, so dragging or automating the semi knob gives you the classic tearing sync lead.

The hexop

Six sine waves. That is the whole instrument: no filter, no sub oscillator, nothing else. What comes out depends on which sines are wired into which, and how hard. The panel is the same six rows whatever you are building, one per operator, each with its level, its frequency as a ratio of the note, and its own envelope.

• The algorithm is the wiring. There are 32 of them and you can't make your own. The dropdown draws each one, so 1←2 means operator 2 modulates operator 1. The panel marks the carriers, the operators that reach your ears, whose level is volume. Everything else is a modulator, and its level is how hard it bends the operator below it.
• Level is exponential. A modulator at half its knob is a sixteenth of full scale, so nearly all the useful range lives in the top quarter. Nudging one modulator level is how you program this thing.
• Every operator has its own envelope, so the timbre has an envelope. A modulator that decays fast under a carrier that doesn't is a struck sound. That is the trick behind an FM electric piano.
• Feedback is the only source of harmonics here that isn't another operator. Wind fbk up and it stops being a tone and turns into noise, which is where the breaths and cymbals come from. Which operator carries it depends on the algorithm, and the panel says which.

The four track knobs ride all six operators at once. bright raises every modulator's level together, which is the fastest way to hear what a patch can do. fbk is the feedback, mod dec scales how fast the timbre falls away, and decay how fast the note does.

Start from the voice dropdown rather than from silence: an electric piano, a bass, a bell, brass, a marimba, an organ and a pad. Load one and change a single operator level. vel makes playing harder raise the modulation index, so hard notes come out brighter and not just louder, and key scale pulls the modulators back as you play up the keyboard. Without it the top octave screams.

The guitar and the bass

Each models the whole chain a real one goes through: the string, the pickup reading it, the amp it runs into and the speaker in front of the mic.

Both start from a tone dropdown of famous rigs. Load the nearest one, move a control, hear what that control is for. Loading a tone replaces everything including the four track knobs.

The guitar

The string has the losses of a real one: the highs die before the fundamental does, so a note gets duller as it rings rather than just quieter. Where you pick notches harmonics out of it, bridge thin and cutting, over the neck round and full, and the pickup notches it again on the way out. Choosing a pickup is choosing where its resonance sits. A single coil peaks high and glassy, a humbucker low and fat.

The four knobs are drive (how hard the amp is hit, on a log taper like the real pot), tone (the knob on the guitar itself, not the amp; roll it down with a neck humbucker for the darkest sound the instrument has), bloom and sustain.

bloom is feedback. The speaker is coupled back into the strings, so past a certain point the energy coming back beats the string's own losses and the note stops decaying and starts growing. It needs volume: a clean amp barely blooms, a cranked one sings. It stops when the note ends, the same way taking your hand off the string does. Hold a long step on a high setting and listen to it climb.

The bass

Same string, wound and stiffer. Its harmonics sit noticeably sharp of where the arithmetic says they should, and that disagreement between the pitch and the clank is most of what a bass sounds like. Flatwounds take it away along with the high end, which is why records made before about 1970 sound the way they do. That is the roundwound / flatwound select.

fret is how hard the string is allowed to clatter against the fretboard; wound up with the hand control at the top, that clatter is slap. grind is distortion, but only above the xover frequency, with the clean low end put back underneath. Distort a bass whole and the bottom vanishes, which is why every bass overdrive worth owning works this way.

The knobs are drive, tone, comp and sustain. Wind comp up and the part sits perfectly still under everything else.

Subby

A monosynth for the bottom two octaves and nothing else. Two notes a third apart at 40Hz beat at a rate you feel as lumpiness rather than hear as harmony. Last note wins, and glide slides into it, either always or only when a note arrives while another is still sounding, which is the 808 slide.

Most people cannot hear a 35Hz sine. A phone speaker starts around 500Hz and a laptop around 180Hz. The fix is harmonics: generate them and the ear rebuilds the fundamental it cannot hear. drive is that path. At zero you get a genuinely pure sine, which is the right choice when something else in the mix is already carrying the note.

• The shaping is parallel and high-passed. xover is where it starts; nothing below it is ever distorted, so the bottom stays clean however hard you drive.
• edge biases the signal before the drive, which makes the clipped pulse uneven, and an uneven pulse is the even harmonics. Four shapers: tube, fold, fuzz and rect.
• drop is the pitch envelope, the 808 beater. It spans 40 semitones and lands exactly on the note. click adds the band of noise that is often the only part of the note a small speaker reproduces.
• reso hangs the silverbox's ladder filter above the crossover, with its own envelope and its own accent. The acid never reaches the fundamental. At zero the whole stage is bypassed.

The four knobs are drive, tone, shape and decay. The tone dropdown loads a complete patch: the trap 808, a distorted one, a pure sine, a reese, an acid sub and a dub sub.

Samples

Choose sampler as a track's engine and it asks for a source: a file from your machine, or one of the bundled drum kits. After that the waveform button in the track header opens the sample editor.

• Trim the start and end, and fade either edge.
• Slice a loop and play the slices from the grid, one per note.
• Fit a loop to the tempo, or leave it at its natural speed.
• Keep pitch locked so a tempo-fitted loop stays in tune, or unlock it and play the sample melodically.

Files you load stay in your browser. They only go anywhere if you save the session to an account or make a share link.

The wavetable editor

With the wavetable engine selected, the waveform button opens an editor where you draw the sound. A wavetable is a series of frames, each a single cycle of a waveform, and the wave knob sweeps between them.

• Draw straight onto the canvas, or build the shape with the harmonic sliders underneath.
• Load a starting point: basic shapes, or any of the bundled waveforms.
• Add frames and the wave knob morphs smoothly between them.
• Unison stacks up to 7 copies of the voice per note, spread apart by the track's detune knob.
• Wave scan sweeps through the frames on its own. Set the speed (free, or in time with the tempo), the direction, and how much of the table it covers. Retrigger restarts the sweep on every note.

Filter, effects and dynamics

Each track has its own chain, opened from the buttons under its name.

The effects chain

Effects run in the order they appear in the panel, starting with amp. drive pushes the signal into everything that follows, so fuzz, the wave shaper, tape saturation and the crusher all bite harder, and out trims the level back afterwards. Both sit at unity in the middle.

Then come vinyl and cassette (wear, warble and noise), fuzz, ring modulation, a wave shaper, a bit crusher, auto-wah, chorus, phaser, flanger, pitch shift, delay and reverb. Each has a wet or amount control that does nothing at zero, so you can poke around the panel safely. The vinyl crackle and cassette hiss only sound while the track is playing, so a muted or stopped track is silent.

The bit crusher has a rate as well as bits: the sample rate it runs at, from 48k down to 250Hz. Bits alone is a noise floor under the sound; rate is where the sampler grit comes from, because anything above half of it folds back down out of tune with the track.

glide lives here too. It slides the pitch between notes instead of jumping, for portamento leads and basses.

Fx buses

Everything in the panels above belongs to one track. So does the mod matrix, and so do the automation lanes. An fx bus is how several tracks come to share them: a track with no instrument, that other tracks are routed through on the way to the master.

Press + add fx bus under the tracks, or pick fx bus from any track's engine dropdown. An out control then appears on every track: point it at the bus and that track arrives there with its own sound intact, like an output assignment on a mixer.

In every other respect the bus is an ordinary track. Its filter, effects, eq and compressor act on everything feeding it at once. Its mod panel sweeps the whole group with one LFO, its aut lanes draw per-step movement across all of it, and it can be p-locked, so the group is drenched in one pattern and dry in the next. A bus can feed another bus. Sends that would loop back on themselves are refused.

A bus plays no notes, so its step grid and roll are gone. mute on a bus cuts the audio passing through it, and solo on a bus keeps whatever feeds it.

p-lock

A track normally has one sound and 32 patterns of notes. Move its filter and it moves everywhere, because the sound belongs to the track and only the notes belong to the pattern.

The p-lock button changes that for the pattern you are on. Locked, that pattern keeps a sound of its own; every pattern you leave unlocked carries on sharing the track's. So the bass can be bright and drenched in delay for the chorus while the verse and the middle eight are left alone, and those two still move together when you tweak them.

It is per pattern, so the button changes as you move around, lighting up on the patterns you have locked. In chain mode the sound changes arrive with the arrangement, on the bar.

Everything about the sound comes along:

• the engine's own controls, meaning the four knobs and whatever panel it has
• the filter and its envelope
• the effects rack, the eq and the compressor
• the modulation assignments in the mod panel

What stays put is the instrument: the engine, and any sample loaded into it. A locked pattern is one instrument played differently, not a different instrument. Routing, macro pads and the generator settings stay put too.

Modulation and automation

Two ways to make a sound move, both per track.

mod: continuous

Assign an LFO to a parameter and it sweeps on its own. Pick the target, a shape, an amount, and either a free speed in hertz or, with sync on, a length: how long one cycle takes in sequencer steps, from half a step up to 64. Filter cutoff is the obvious target, but effect amounts and the instrument's own tone controls work just as well.

The ± switch beside the amount says where that movement sits. On, the modulation swings either side of wherever the knob is. Off, it only lifts the parameter above the knob and lets it fall back: the same amount of movement, all on one side. Use that for anything that should idle where you left it and only open up.

phase is where in the cycle the shape starts. A quarter turn between two LFOs at the same rate gives you a circular pan.

Two of the six shapes are not waveforms. rnd square holds a fresh random value for a whole cycle and then jumps to another, which is sample and hold. euclid taps the parameter in a euclidean rhythm instead of sweeping it. Its length is the step length, so “1 step” means one ring step per sixteenth, and you set the pulses, the cycle length and a rotation the way you would on a track. decay at zero holds each tap for its whole step, which is a gate; turn it up and each tap falls away, which is a pluck. It locks to the transport, so the rhythm it plays is the rhythm the sequencer is playing. Put it on a filter for a gated sweep, on reverb for throws that land on the offbeats, or on a live euclid track's rotate and have one euclidean pattern turn another.

aut: per step

Automation draws a value for each step of the pattern instead. Choose a parameter and you get a lane under the grid where every step holds its own value: a filter that opens over 16 steps, a delay that only shows up on the last beat. Lanes belong to the pattern, so each pattern can move differently.

Right-click any parameter

To see what is moving one control, right-click the knob, the switch or its label. It works anywhere: the instrument row, the filter, effects, eq and comp panels, the sample and wavetable editors. You get a small window for that one parameter, holding what it does, whatever LFO or automation is on it, its macro assignment, and a button to add any of them.

A parameter with something on it wears a dot next to its label: green for an LFO, blue for an automation lane, grey for a lane you have switched off.

A knob being driven grows a second needle in that same colour, at wherever the LFO or the lane has pushed the parameter right now, while the knob itself stays where you left it. The two needles together tell you how far it is travelling and where it comes back to.

Macro pads

macro in the transport opens a set of XY pads. Each axis drives a list of parameters, and that list can span tracks: one thumb opening the bass filter while ducking the lead's reverb.

Two ways to assign a parameter to an axis:

• Learn. Press learn on an axis, then touch any control anywhere in the app. The panel gets out of the way while an axis is armed.
• The parameter menu. Right-click a control and assign it there, next to its LFO and its lane.

Each assignment has a from and a to, so one axis can open one filter across its whole range while barely moving another, and flip reverses it.

The assigned knobs move as you play the pad. By default the move is momentary: let go and everything ramps back to where it was, so nothing is committed. latch instead leaves the parameters where you put them, exactly as if you had moved the knobs yourself, so they go into the patch and into a save.

Pads are global rather than per track, and they don't follow p-lock. They are saved with the session.

Undo

Ctrl/⌘ Z steps back, Ctrl/⌘ shift Z steps forward, and the two arrows in the transport do the same. It covers the whole session, not just the grid: notes, knobs, tracks added and removed, generator settings, a pattern you cleared, a session you loaded by accident.

A drag counts as one step, so a knob moved in three goes comes back in one. Which pattern you are looking at is not part of it, so an undo changes the song and never the view. Text fields keep the browser's own undo: retyping a track name is the field's history, not the song's.

Jamming with other people

jam in the top bar starts a room from the song that is open and gives you a link. Everyone who opens it edits the same song with you, live: steps, knobs, tracks added and removed, generators, a song opened from the menu. Nobody needs an account, only the link.

Each of you plays it on your own machine, so press play wherever you are; the transport is not shared, and neither is which pattern you are looking at. Everything else is one song. Undo steps back over the others' edits as well as your own, and an undo goes out to the room like any other edit.

Saving, sharing and export

Leave the name field blank and the song gets named after what is in it, something like basement squelch or soft vapour. The adjective comes from the tempo and the key, the noun from whichever engine the song is mostly made of. The name appears in the field when the save box opens, so you can read it and type over it.

A saved song is a tree, not a file that gets overwritten. Every save hangs off the version you are working from, so the history keeps a straight line while you go forward and branches the moment you open an older version and carry on. Forking leaves the tree and starts a new song. Sharing a song shares the state you saved, never the versions behind it.

Without an account everything lives in the tab, which means a reload or a clearing of site data takes it with it. An account is the way to keep it.

Letting an AI agent write a song

seqbaby ships an MCP server, which is the plug that lets an AI agent such as Claude use a tool. Connected, the agent can start a song, add tracks, write their rhythms and notes, shape their sound, put movement on them, and hand you a share link. It writes the same session the studio saves, so whatever it makes opens here like any other song, and you carry on by hand from there.

The agent works in the same vocabulary as this manual. A rhythm is a step string, so x..x..x. is a hit on steps 1, 4 and 7; X is an accent,_ a tie; a string shorter than the pattern repeats across it. Notes are names like C2. Every engine, every panel control with its range, every preset and every modulation target is served to the agent as reference, and each value it sends is checked against the engine's own tables, so a wrong engine name or a control outside its range comes back as a correction rather than a silent track. A guide tells it which engine suits which job and how each one likes to be played: that a silverbox accent is a hard hit and a slide is a tie, that subby needs drive to be heard on a phone, that a reverb belongs on a bus.

Connecting it

In Claude Code, register the server once. It runs straight from the repository; the first start installs it, later starts are instant.

Claude Desktop and other MCP clients take the same command in their configuration file:

Working inside a checkout of the repository needs no registration at all: Claude Code finds the server in the project and the composing guide with it.

Asking for a song

Say what you want in plain words, and ask for a link at the end:

The reply is a link that opens the song here. Keep talking to change it: busier hats, a different bass preset, a filter sweep across four bars. Each request edits the same song and each share is a fresh link. If you would rather nothing was published, ask for the song as a file instead and load it from the session menu's import.

If something sounds wrong

• No sound at all. Press play once more, since browsers hold audio back until you interact with the page. Then check that nothing else is soloed and that the track's volume is up.
• One track is silent. Look for a soloed track elsewhere, a muted one here, or a filter cutoff closed all the way down.
• Playback stutters. Reverb, granular and pitch shift are the expensive ones, and a lot of them at once on a slow machine will do it. Closing other tabs helps more than you would think.
• The playhead looks out of time with what you hear. Safari can't report its audio delay, so the display works from an estimate. Add ?vlat=0.2 to the address to tune it.
• An arp isn't audible. The note is probably too short to fit more than one arp note in. Lengthen it, or choose a faster arp rate.
• A knob moves but nothing happens. Check whether an LFO or an automation lane owns it. A driven knob is the base the movement swings around, and a lane rewrites it every step.
• Sound cut out after switching apps. Click anywhere on the page. The audio engine reconnects on your next interaction.

Credits and licences

seqbaby is built on other people's work. Thank you to everyone below, and if you use the sounds in something you release, these are the licences that travel with them.

The emulator engines are original models of those instruments rather than recordings of them, and the trademarks belong to their owners.