The open beta is live for Apple silicon Macs Download The beta opens soon for Apple silicon Macs Get on the list

The DAW with an
engineer inside.

A native macOS DAW for music production: recording, MIDI, mixing and mastering, with an AI audio engineer built in.

Download for macOS
Steel

Type a request and it measures the session, then changes the same faders, devices and clips you use. Every change is listed and can be undone.

For example: say “the vocal gets lost in the chorus” and it lifts the vocal in the chorus, and only there. Say “duck the bass under the kick” and it adds a Duck to the bass, triggered by the kick.

The workspace.
One window for the whole session.

The engineer panel, browser, transport, Loop Deck, arrangement and device chain are all in one window. The numbered points mark each area.

The engineer has no separate mode or project. It runs the same commands as the menus and shortcuts, on the session that is open.

The Zammoth workspace while a song plays: the engineer panel with a finished Auto Mix, the sample browser, the Loop Deck with four loops, the arrangement, the track controls, and a drum track's device chain along the bottom.
The engineerType what you want here. Its answers, and every change it makes, show up in this panel.
The browserSamples you can preview, the built-in devices, your plug-ins, the project library and the groove pool.
Tempo and keyDetected from the audio and shown in the transport. The engineer reads the same values.
The Loop DeckA pad for every track. Launch loops in time, then capture the jam into the arrangement.
The arrangementAudio and MIDI clips, takes and automation lanes on one timeline.
The device chainThe selected track's devices, each showing what it is doing to the sound right now.
  1. The engineer. Type what you want; its answers and changes show up here.
  2. The browser. Samples, devices, plug-ins, the library and grooves.
  3. Tempo and key. Detected from the audio and shown in the transport.
  4. The Loop Deck. A pad for every track, launched in time.
  5. The arrangement. Clips, takes and automation on one timeline.
  6. The device chain. Each device shows what it is doing right now.
One set of controlsA change from the engineer and a change by hand land on the same fader, device or clip. There is no second copy of the project to keep in sync.
The session is readableThe engineer can read every track, clip, device, route and meter, so a request can refer to any of them by name.
Changes are listedEach change appears in the panel with the value it moved and the reason.
One undo step per requestA request undoes as one step, however many moves it made. Changes you made afterwards are kept.

The Zammoth agent.
It runs all 325 of the DAW's commands.

The agent is the engineer in the panel on the left. It is Zammoth's own software, written for this DAW. A request in plain words becomes a sequence of commands with exact values: a fader in dB, an EQ band in Hz, a note at a bar and beat, a send level, a compressor's attack in milliseconds.

It stays out of the audio path. It sets the parameters, and the DAW's processors do the work in real time.

Every command in the DAW. One segment each.
  • 325 the agent runs
  • MIDI49
  • Clips34
  • View32
  • Loop Deck30
  • Plug-ins29
  • Mixer23
  • Tracks21
  • Library14
  • Recording13
  • Agent7
  • Automation7
  • Markers6
  • Time6
  • Transport5
  • Grid5
  • Tempo5
  • Session2
  • Edit2
  • Arrange2

Each request shown is a case from the agent's command tests, with the command it is graded on.

  1. ReadIt reads the session.It works from the project as it stands at that moment, including your latest edits, and from the analysis of its audio.
  2. PlanIt plans in commands.The request is broken into named operations, each with its target and its value. In Agent mode they run straight away.
  3. RunIt runs them in order.There is no fixed limit on the number of steps. A request ends when the job is done or when you stop it.
  4. MeasureA mix run repeats in passes.Each pass makes up to 24 moves and is followed by a new measurement of the mix before the next one starts.
What it runs on

The agent's decisions are made by Claude 5, the most capable family of models Anthropic makes. You choose the model and how much work it puts into a request.

Opus
Opus 5.5

The newest Opus, and the largest of the three. It is the model the app ships set to.

Sonnet
Sonnet 5.5

The newest Sonnet. The smaller model, at half the price of Opus 5.5 for the same request.

Haiku
Haiku 5.5

The newest Haiku. The fastest and lowest priced of the three; a long request costs more per token than a short one.

Effort
  1. Low
  2. Medium
  3. Highdefault
  4. X-High
  5. Max

The mixer.
Every level where you can read it.

Each strip is laid out in the order the signal travels: the level coming in, the devices it passes through, the level going out, then pan, the fader and the output it feeds. IN and OUT are held peaks in dB, so the gap between them is what the chain is adding or taking away.

The Zammoth mixer: four smoked-glass channel strips for vocals, drums, bass and other, with level meters moving and each channel's devices listed above its fader.
On every channel
  • The fader is inside the meter. It is a line across both bars on one −60 to +6 dB scale, so you read the level against the fader itself. ⇧-drag for fine moves.
  • Meters show what leaves the channel. Post-fader and post-pan, with an RMS body, a peak line that holds for 1.5 seconds and a clip light that latches.
  • Gain reduction beside the level. A column shows the total a compressor, gate or Duck is taking off. It is the dashed one on the vocals and the bass here.
One real Auto Mix pass: four tracks, five moves
auto mix
bass+6.0 dB
drums−6.0 dB
drums · eq @ 2.8 kHz−4.0 dB
other · eq @ 244 Hz−4.0 dB
vocals+1.5 dB
loudness−19.4 → −17.3 LUFS

Amber is what the engineer changed. Everything else is your own work.

One slot per deviceThe chain is listed between the IN and OUT readings, with a light on each slot for on or bypassed. Double-click a slot to open the device, or click an empty one to add a device there.
Faders that move togetherSelect several strips and drag one fader or pan. The others follow by the same amount, so the balance between them holds.
Route it anywhereThe key at the foot of a strip sends the channel to the master, a bus, another track or a hardware output. Sends Only takes it out of the main mix and leaves its sends running.
Track delayShift a whole channel up to 200 ms earlier or later, in milliseconds or samples, to line up a distant mic or a player who rushes.
How it hears your mix

The engineer does not receive audio. Your Mac analyses every track and the mix, and the engineer reads the measurements that analysis produces.

LoudnessIntegrated loudness in LUFS for each track and the mix, loudness range, and true peak against the ceiling.
ToneEnergy per frequency band on every track, plus composite readings for mud, harshness and clarity.
DynamicsCrest factor, overall and per band. It shows where a part is already compressed and where it still has peaks.
MaskingThe bands where one track covers another while both are playing. A common reason a part sounds buried.

Key and tempo are detected from the audio as well. All analysis runs on your Mac. The model is sent your conversation, these measurements and a description of the session. It is never sent the audio.

Music and audio questions,
answered from your session.

The engineer answers questions on theory, groove, mixing, dynamics, vocals, tone and mastering. When a question concerns your song, it analyses the relevant part first and answers with those measurements.

It answers within music and audio. Requests outside that scope are declined.

Theory

Groove

Mixing

Dynamics

Vocals

Tone

Mastering

Each of these is a case from the engineer's test set, and each links to a written answer.

How one question goes
  1. It measures first.For a question about your song, such as whether the chorus lifts or which vocal line is flat, it analyses that part before answering. The answer quotes those measurements.
  2. It gives the cause.The answer names what is causing the problem and why a given tool addresses it, as well as the setting to use.
  3. It can apply the change.On request it makes the change on your track. In Ask mode it only explains and plans, and the session is left as it is.
Reference materialFrequency bands, instrument ranges and the usual causes of “muddy” or “harsh” are always available to it. Deeper material on harmony, arrangement, vocals, sound design, groove, recording and mastering is opened when a question needs it.
Confidence is statedChords detected from audio are reported with the confidence of the detection. If the app cannot do what was asked, it says so.
Findings on openA new session opens with a list of what the analysis found, such as two parts crowding the same range. Each item has a one-click fix.
Help on every controlHovering a fader, knob or device shows a card with its name and function. There are written guides to sidechain, stems and mastering for streaming.

The DAW.
Recording, MIDI, routing and export.

Every operation is a named command. Menus, shortcuts and the engineer call the same set, and the search below lists all of them.

The Zammoth arrangement: four coloured tracks with their waveforms and the playhead, the Loop Deck to the left and each track's controls and meters to the right.
What's included
  • Record, comp, warp. Takes and comping, count-in, time-stretch, fades and splits.
  • MIDI and instruments. A piano roll, a four-oscillator synth, a sampler, a drum rack and an arpeggiator.
  • Song structure. Markers, named sections, a tempo lane and time-signature changes.
  • Routing. Group buses, sends and returns, sidechain ducking and VCA groups.
  • Your plug-ins. VST3 and Audio Units, both effects and instruments.
  • Hardware and sync. Mackie Control surfaces, MIDI Learn, MIDI clock and MTC out, and OSC remote control.
  • Export. The master and every stem in one bounce.
A vocal's device chain while the song plays: an EQ with its live spectrum, a compressor showing gain reduction, a de-esser catching each sibilant, and a reverb.
A vocal chain while the song plays: EQ, compressor, de-esser and reverb. Every device has a live display like these.

Every device and instrument, by name

  • Compressor
  • De-esser
  • Duck
  • Gate
  • Limiter
  • Multiband Compressor
  • OTT
  • Transient Shaper
  • EQ
  • Dynamic EQ
  • Tilt EQ
  • Match EQ
  • Clipper
  • Overdrive
  • Saturation
  • Multiband Saturation
  • Chorus
  • Flanger
  • Reverb
  • Delay
  • IR Reverb
  • Band Width
  • Stereo Imager
  • Stereo Width
  • Spectrogram
  • Utility
  • Drum Kit
  • Sampler
  • Four-oscillator Synth
  • Arpeggiator

Devices.
Twenty-six, each with a live display.

Double-click a device's display and it opens across the width of the window. These are those views, captured while audio played. The ranges are the engine's own. A twenty-seventh device, Match EQ, is created by the engineer during reference matching: a linear-phase correction in eight fixed bands. Each name links to that device's reference page, with every parameter's range and default.

The Compressor's expanded view: a transfer curve, a threshold dial at −32 dB ringed by a red gain-reduction arc reading 7.6 dB, and a detector scope with the threshold line.
Dynamics

Compressor

Downward compressor. The left screen is the transfer curve with a draggable knee; the right is a detector scope showing the program envelope, the output and the gain reduction, with the threshold as a line on the signal. Mix below 1 blends in the dry signal for parallel compression. Set one compressor to mid and a second to side to compress the two independently.

Threshold
−60 to 0 dB
Ratio
1:1 to 10:1
Attack
0.1–100 ms
Release
10–1000 ms
Makeup
0–12 dB
Mix
0–1
Channel
stereo, mid or side
The Multiband Compressor's expanded view: low, mid and high bands over the spectrum with a threshold line and a reduction figure in each, and four reduction meters.

Three bands split by fourth-order Linkwitz-Riley crossovers, each with its own threshold, ratio and makeup. A band's ratio is bipolar: right of centre is downward compression, left of centre is upward compression, which raises material under the threshold. The display draws the input and output spectra and fills the difference in each band.

Threshold, per band
−60 to 0 dB
Ratio, per band
1:1 to 10:1, down or up
Makeup, per band
0–12 dB
Attack, shared
0.1–100 ms
Release, shared
10–1000 ms
Crossovers
30–4000 Hz and 200–16,000 Hz
The OTT's expanded view: three bands over the spectrum reading +6.1, −5.2 and −2.6 dB of gain, a depth dial at 0.60 and four bipolar gain meters.
Dynamics

OTT

Three-band upward and downward compression. Each band is pulled toward a fixed −30 dB target: level above it is reduced and level below it is raised. Depth scales the whole effect, Time scales the attack and release, and each band has separate Up and Down amounts.

Depth
0–1
Time
0.1× to 4×
Up and Down, per band
0–1 each
Input and output gain
±12 dB each
Crossovers
30–4000 Hz and 200–16,000 Hz
The Limiter's expanded view: a ceiling dial at −1 dBTP, six seconds of gain reduction, an output sample peak of −1.8 dB and a short-term loudness of −11.9 LUFS.
Dynamics

Limiter

Look-ahead brickwall limiter. The detector estimates inter-sample peaks, so the ceiling is in dBTP. The lookahead is reported as latency and compensated. On the master it also holds the integrated loudness target and shows short-term LUFS against it.

Ceiling
−6 to −1 dBTP
Loudness target
−24 to −6 LUFS-I, master only
Release
1–1000 ms
Lookahead
0.5–3 ms
Stereo link
0–1
The Gate's expanded view: a detector meter with the open threshold at −30 dB, the gate open, a rhythm strip of its openings and the detector's spectrum.
Dynamics

Gate

Downward expander with hysteresis, hold and a range floor. The detector has its own high-pass and low-pass filters, which are not in the audio path. The display shows the detector level against the open and close thresholds, the attenuation, and four seconds of openings on the beat grid.

Threshold
−80 to 0 dB
Ratio
1:1 to 20:1
Attack
0.1–100 ms
Release
5–2000 ms
Hold
0–500 ms
Range
−80 to 0 dB
Hysteresis
0–24 dB
Detector filters
high-pass to 2 kHz, low-pass from 200 Hz
The De-esser's expanded view: six seconds of the detector band with the threshold line across it, its catches drawn in red above, and a maximum reduction of 6 dB.
Dynamics

De-esser

Split-band compressor. The detector listens to one band between 2 and 12 kHz and reduces that band when it crosses the threshold. Wide applies the reduction to the full signal instead. The tape shows six seconds of the detector band with each reduction drawn from the top. Holding SC solos the detector band.

Threshold
−60 to 0 dB
Frequency
2–12 kHz
Ratio
1:1 to 20:1
Attack
0.1–20 ms
Release
10–500 ms
Range
−80 to 0 dB
Wide
off or on
The Duck's expanded view: the drums chosen as the trigger, the trigger level above the threshold line and the track held down by 6 dB.
Dynamics

Duck

Sidechain gain reduction keyed from another track. While the trigger track is over the threshold, this track is turned down by the depth. The display plots six seconds of the trigger level against the threshold, with the reduction beneath it. It sits on a track's chain, one per track.

Trigger
any other track
Threshold
−60 to 0 dB
Depth
1–12 dB
Attack
1–100 ms
Release
20–1000 ms
The Transient Shaper's expanded view: one hit envelope with the attack raised by 4 dB and the sustain lowered by 3 dB, between an attack knob and a sustain knob.
Dynamics

Transient Shaper

Separate gain for the attack and the sustain of each detected onset. The display is one hit envelope: the attack portion and the tail are drawn at their current gain over an outline of the unprocessed hit. Speed scales the timing.

Attack
±6 dB
Sustain
±6 dB
Speed
0.25× to 4×
Output
±12 dB
The EQ's expanded view: three bands on the plot, a cut at 120 Hz and boosts at 2.8 and 9 kHz, with the summed response drawn over the live spectrum.
EQ and tone

EQ

Parametric equaliser with up to 24 bands. Each band has its own shape and acts on stereo, mid, side, left or right. High-pass and low-pass slopes run from 6 to 96 dB per octave. The plot shows the live spectrum, each band's curve and the summed response, and bands are added, moved and removed on the plot itself. Holding a band's SOLO plays only that band's passband.

Bands
up to 24
Frequency
20 Hz – 20 kHz
Gain
±12 dB
Q
0.3–8
Shapes
peak, low shelf, high shelf, high-pass, low-pass, band-pass, notch, tilt shelf
Cut slope
6, 12, 24, 48 or 96 dB per octave
Channel, per band
stereo, mid, side, left or right
Output
±12 dB
The Dynamic EQ's expanded view: one band at 3 kHz with its range outline, its threshold marker and its controls over the live spectrum.
EQ and tone

Dynamic EQ

Up to six bands whose gain changes only while the band's own level is over its threshold. Each band has a threshold, ratio, attack and release. The plot draws a band's maximum range as an outline and the gain it is applying now as the live curve.

Bands
up to 6
Frequency
20 Hz – 20 kHz
Q
0.3–12
Depth
±12 dB
Threshold
−60 to 0 dB
Ratio
1:1 to 10:1
Attack
1–100 ms
Release
20–1000 ms
The Tilt EQ's expanded view: the response line tilted +2 dB toward the highs around a 1 kHz pivot, over the live spectrum.
EQ and tone

Tilt EQ

Tilts the whole spectrum around a pivot frequency. The line on the display is the exact response the device applies, drawn over the live spectrum. Drag either end to set the tilt and the marker to move the pivot.

Tilt
±12 dB
Pivot
200 Hz – 8 kHz
Output
±12 dB
The Clipper's expanded view: six seconds of peak and RMS level with the peaks above the threshold marked in red, and input and output crest factor meters.
Colour

Clipper

Peak clipper with a knee that runs from soft to hard and oversampling up to 8×. The display plots six seconds of peak and RMS level against the threshold, marks each peak it removed in dB, and reports the crest factor in and out. Holding Δ plays only the removed material.

Threshold
−12 to 0 dB
Hardness
0–1, soft to hard
Output
±12 dB
Oversampling
1×, 2×, 4× or 8×
The Overdrive's expanded view: input and output spectra with the added harmonics filled in, and the transfer curve bending into its ceiling.
Colour

Overdrive

An optional resonant band-pass filter into an exponential soft clipper, followed by a tone tilt. Only the clipper is oversampled. With Mix below 1 the dry path is delayed to match the oversampler, so the blend does not comb. The display shows the input and output spectra and the transfer curve with its ceiling.

Drive
0–36 dB
Filter
off or on, 40 Hz – 12 kHz, Q 0.3–8
Tone
0–1, 0.5 is flat
Output
−24 to +12 dB
Mix
0–1
Oversampling
1×, 2×, 4× or 8×
The Saturation's expanded view: a drive dial at 8 dB, the shaped waveform between its knee lines, 10 % THD and a harmonic meter.
Colour

Saturation

Memoryless waveshaper with four curves: tanh, warm, tube and hard. Oversampling up to 8× keeps the harmonics it generates from aliasing back into the audible band. The display shows the input and the shaped waveform, total harmonic distortion, and the level of harmonics 2 to 7.

Drive
1–12 dB
Curve
tanh, warm, tube or hard
Output
±12 dB
Mix
0–1
Oversampling
1×, 2×, 4× or 8×
The Multiband Saturation's expanded view: low, mid and high bands over the spectrum with a knee line and a THD figure in each.

Three bands, each with its own drive, curve and mix, split by two adjustable crossovers. The display shows the input and output spectra, the harmonics each band adds and where they land, and the distortion per band.

Drive, per band
0–24 dB
Curve, per band
tanh, warm, tube or hard
Mix, per band
0–1
Global mix
0–1
Crossovers
30–4000 Hz and 200–16,000 Hz
The Chorus's expanded view: a rate dial at 0.8 Hz inside a ring showing each voice, two cycles of the delay sweep, and readouts of sweep, detune and period.
Modulation

Chorus

One to three delay voices swept by an LFO. The sweep is offset between left and right by the Width angle, so a mono source comes out stereo. There is no feedback path. The readouts give the delay's excursion in milliseconds and the peak pitch deviation it causes in cents.

Rate
0.01–10 Hz
Depth
0–1
Delay
1–50 ms
Voices
1–3
Width
0–180°
Mix
0–1
The Flanger's expanded view: the comb-filter response with its first notch at 111 Hz, a feedback dial, and readouts of comb depth, notch frequency and spacing.
Modulation

Flanger

One swept delay voice with a feedback path. Positive feedback puts peaks on the multiples of 1/delay; negative feedback puts the notches there. The display is the device's own frequency response, computed from the delay it is reading at that moment, and it reports the comb's depth in dB.

Rate
0.01–10 Hz
Depth
0–1
Delay
0.5–20 ms
Feedback
−0.95 to +0.95
Width
0–180°
Mix
0–1
The Auto Filter's expanded view: a 24 dB/oct low-pass at 900 Hz with its resonant peak, a triangle LFO at a quarter note, the trigger set to its own input, and the cutoff now at 1.27 kHz.
Modulation

Auto Filter

A resonant low-pass, high-pass, band-pass or notch filter at 12 or 24 dB/oct whose cutoff moves. An LFO pushes it up and down in octaves, in Hz or in note values locked to the song's tempo map, and an envelope follows a level: the track's own, or another track's as a trigger. The two add. The display draws the filter at its knobs, the band the movement reaches, one cycle of the LFO, and the cutoff as it plays.

Type
low-pass, high-pass, band-pass or notch
Slope
12 or 24 dB/oct
Cutoff
20 Hz–20 kHz
Resonance
0–1
LFO
0–6 octaves; sine, triangle, saw up, saw down, square or random
Rate
0.01–20 Hz, or 1/32 to 8 bars
Envelope
−6 to +6 octaves
Trigger
its own input or any other track
Mix
0–1
The Delay's expanded view: a time dial at 250 ms on a note-value scale, five repeats falling 12 dB each, and readouts of tail length and tone cutoff.
Space

Delay

Feedback delay with a one-pole low-pass on its output. The time dial carries a note-value scale at the session tempo. The scope plots each repeat's level, the drop per repeat in dB and the number of repeats before the tail has fallen 60 dB.

Time
20–2000 ms
Feedback
0–0.85
Tone
low-pass, 1.2–18 kHz
Mix
0–1
The Reverb's expanded view: a decay dial at 1.8 s, the tail curve with pre-delay and decay handles, a tone strip, and readouts of wash, gap, tone and dry.
Space

Reverb

Algorithmic reverb with four parameters. The scope draws the tail on a logarithmic time axis, from the pre-delay to −60 dB. The readouts give the time to −40 dB and the share of the last three seconds in which the reverb was louder than the dry signal. Like Delay and IR Reverb, it runs as an insert with a dry/wet mix or fully wet on a return.

Decay
0.2–12 s
Pre-delay
0–250 ms
Tone
low-pass, 1.2–18 kHz
Mix
0–1
The IR Reverb's expanded view: the hall impulse as a spectrogram with the RT60 marked at 2.0 s, and the room, hall and plate selectors.
Space

IR Reverb

Convolution reverb with three generated impulse responses: room, hall and plate. Decay is scaled per space, and the face prints the resulting RT60. The display is the impulse as a spectrogram, time against frequency, with the tone cutoff and the RT60 marked on it.

Space
room, hall or plate
Decay
0.2–8 s
Pre-delay
0–250 ms
Tone
low-pass, 1.2–18 kHz
Mix
0–1
The Band Width's expanded view: the 2–4 kHz band marked on the spectrum, a width dial at 1.00 and a vectorscope of that band.
Stereo

Band Width

Stereo width on one frequency band, placed anywhere in the spectrum. The band is set on the live spectrum or snapped to one of eight named bands. A vectorscope shows the processed band alone with its correlation, and a lamp turns red when the correlation falls under +0.2.

Width
0–2, 1 is unchanged
Band
eight named bands, or free low and high edges
The Stereo Imager's expanded view: low, mid and high wedges on a top-down stage, three width dials and a correlation history.
Stereo

Stereo Imager

Stereo width in three bands with two adjustable crossovers. Each band is a wedge whose opening is the width that was set; the points inside it are the band's measured correlation. A strip at the foot plots the last twelve seconds of all three correlations.

Width, low, mid and high
0–2 each
Crossovers
30–4000 Hz and 200–16,000 Hz
The Stereo Width's expanded view: the signal plotted on a top-down stage between two speaker marks, a width dial at 1.35 and the correlation meter.
Stereo

Stereo Width

Mid/side width from mono to double. The stage view plots the signal as points in front of the listening position, with the speakers at ±30°. The field view shows side against mid per frequency band. Two keys play the mid or the side alone, and the foot plots twelve seconds of correlation.

Width
0–2, 1 is unchanged
Correlation meter
−1 to +1
The Utility's expanded view: input and output meters, a gain dial at −3 dB, width and balance dials, a vectorscope and the DC, polarity and mono keys.
Utility

Utility

Gain, polarity per channel, width, bass mono, balance and a DC filter, applied in this order: DC filter, gain, polarity, width, bass mono, balance. The meters read this device's own input and output, so one placed late in a chain measures what the devices before it left.

Gain
±24 dB
Width
0–2
Bass mono crossover
50–500 Hz
Balance
−1 to +1
Switches
mono, invert left, invert right, bass mono, DC filter
The Spectrogram's expanded view: the tape picture with the loudest partial traced, eight band histories beneath it, and peak, tilt and LUFS readouts.
Utility

Spectrogram

Analysis only: the audio passes through unchanged. The picture is either a flat tape with a note ruler and the loudest partial traced, or the last twelve seconds in perspective. Under it run eight band-energy histories with momentary LUFS, or a twelve-lane pitch-class display with the detected key. Enhanced mode uses frequency reassignment for a sharper picture.

Picture
tape or ridge
Lower panel
bands, notes or off
Hover readout
frequency, note, cents, level and bar.beat
Modes
full, freeze, enhanced

Racks, automation
and modulation.

A rack holds parallel device chains. An automation lane changes a control over time. A modulator moves a control continuously. All three can be built and edited by hand or by the engineer.

A rack named Vocal Rack: eight macro knobs, four of them mapped, and three parallel chains named Crush, Space and Width (Compressor and Saturation, Reverb and EQ, Chorus and Delay), each with mute and solo buttons, a level fader and a pan cell.
Select, then ⌘G

Racks: parallel chains in one device

Select a few devices and press ⌘G. They become one chain of a rack, sounding exactly as they did. Add more chains beside it, each with its own fader, pan, mute and solo, and blend them. Built-in devices and your own plug-ins sit in the same rack.

group the EQ and compressor into a rack
Automation Mode in the arrangement: a volume ride drawn on the vocal track with seven points, every rise and fall bent into a curve, and a flat volume lane on the drums below, beside each track's controls.
Press A

Automation: draw, record or curve

Click a lane to add points, draw steps with the pencil, or arm a track and ride the fader while the song plays. Touch, latch and write decide how a ride ends, and ⌥-drag bends any segment into a curve. Volume, pan, every device control, send levels and rack macros all have lanes.

ride the vocal up in the last chorus
MacrosEight knobs on every rackMap a macro to any control inside the rack, on a plug-in or a built-in device, and one knob can move several at once. Macros have automation lanes of their own.
PresetsSave the whole rackChains, devices, macros and plug-in settings are stored together. The preset shows up in the browser, ready to drag onto any track.
ModulatorsFour sources for any knobRight-click a device and put an LFO, an envelope follower, a random source or a step pattern on any of its knobs, free-running or locked to the tempo. It starts centred on the knob's current position, so the value does not jump.
OverridesManual overrideMoving an automated fader by hand overrides its lane from that moment. A lamp in the transport stays lit while a lane is overridden, and one click on it returns control to the lane.
Drum rackA chain on every padEach pad can carry its own effects, built-in or plug-in, so a reverb on the snare does not touch the kick.
EngineerBuilt on requestA rack or a ride the engineer builds is made from the same chains, lanes and points. It can be opened and edited afterwards like any other.

Three built-in instruments.
Synth, drum rack and sampler.

Each one plays from your keyboard or a controller, and each can be programmed by the engineer.

The WOOLLY synth on its Warm Pad patch: four oscillators with wave displays, a filter with its response curve, filter and amp envelopes, two LFOs, two mod envelopes, a mod matrix, and drive, chorus, delay and reverb along the bottom.
Synth

WOOLLY, a four-oscillator polysynth

Four oscillators with up to eight unison voices each, a filter with its own envelope, two LFOs, two modulation envelopes, and a mod matrix you rewire by clicking. Drive, chorus, delay and reverb are built in, and there are seven factory patches. Every knob can be automated, modulated or mapped to your hardware.

make me a warm pad with a slow filter sweep
The Drum Rack editor: a four by four page of pads holding a kick, rim, snare, clap, toms, hats, crash and ride, a map of all 128 notes on the left, and the selected kick pad's gain, pan, tune and choke cells on the right.
Drum rack

Every pad is a sampler of its own

Sixteen pads to a page, across all 128 notes. Drop a sample on a pad, set its gain, pan and tuning, and put the hats in a choke group so the closed one cuts the open one. Open any pad and it is a full sampler, with its own envelopes, filter and velocity layers.

tune the kick down two semitones
The Sampler in Slice mode with a drum loop loaded: sixteen numbered slice markers across the waveform, a keyboard beneath it, and the amp envelope, filter and LFO modules on the right.
Sampler

Chop a loop across the keys

Drop in any sound and play it looped, as a one-shot, or sliced. Slice detects up to 64 hits and assigns each one a key. One more press writes a MIDI clip that replays the loop in its original timing, which you can then edit note by note. The sampler can also follow the song's tempo and layer up to sixteen samples by key and velocity.

slice this break and give me the MIDI

MIDI.
Notes, steps, controllers and timing.

A MIDI clip holds notes and controller curves. It is edited in the piano roll, programmed in the Step Grid, recorded from a keyboard or written by the engineer. The parts in these pictures were written by the engineer, on request.

The piano roll on a four-bar clip in A minor: four chords held a bar each in the lower octaves, a melody above them, note names on every note and on the keyboard, and a velocity handle for each note in the lane underneath.
Piano roll

Notes on a grid, held to the song’s key

Double-click to add a note one grid cell long. Drag a selection to move it in pitch and time, drag an edge to resize it, and ⌥-drag to copy it. The grid runs from one bar down to 1/32, with triplets. SCALE marks the song’s key on the keyboard and snaps notes to it, and FOLD shows only the pitches in use. The arrow keys nudge by a grid step and transpose by a semitone, or by an octave with ⇧.

write four bars of chords in A minor with a melody on top
The lower part of the same clip with the lane switched to Cutoff: a CC74 curve of five points rising in a straight line across the four bars.
Lane

Velocity and four controller curves

The lane under the notes switches between velocity, mod wheel (CC1), pitch bend, expression (CC11) and cutoff (CC74). Velocity is one handle per note. The other four are point curves: click to add a point, drag to move it. A new synth patch already responds to all of them: CC74 moves the filter cutoff, CC11 the oscillator level and CC1 the vibrato depth, and pitch bend covers twelve semitones. The sampler responds to the same four.

open the filter slowly over the four bars
The piano roll on a drum track, its header reading Drum Kit: fourteen rows named for the kit pieces instead of a keyboard, a kick and snare pattern, closed hats on every sixteenth, an open hat at the end of each bar, and a velocity handle for every hit underneath.
Drum view

One named row per kit piece

On a drum track the piano roll replaces the keyboard with the kit, one named row per piece. Notes are added, moved and deleted the same way, the lane edits each hit’s velocity, and clicking a row’s name plays that piece.

add ghost snares and an open hat at the end of each bar
The Step Grid on a sixteen-step pattern: rows for kick, snare, closed hat, open hat and clap, lit cells in different brightnesses for velocity, and four closed-hat cells hatched to show a 50 percent probability.
Step Grid

Velocity, gate and probability per step

A step clip opens as a grid of 8, 16 or 32 steps with one row per kit piece. Click a cell to switch it on, drag it vertically for velocity, and ⌥-drag for a gate length from 0.1 to 1.5 beats. Right-click sets the probability that the step plays: 100, 75, 50 or 25 %. Cells below 100 % are hatched, and the four hatched hat steps in the picture are at 50 %. + Channel adds a row on any MIDI note.

give a few of the hat steps a 50% chance
The Arpeggiator while the song plays: its display shows the up-down pattern it is playing from four held notes at sixteenths, above a row of twelve styles, the rate buttons, seven knobs from gate to shift, and the scale and key cells.
Arpeggiator

Twelve styles, in time with the song

A MIDI effect that sits before the instrument and turns held notes, from a keyboard or a clip, into a pattern. It has twelve styles, rates from 1/1 to 1/32 (straight, dotted or triplet), a range of one to eight octaves, gate, swing, up to eight repeats per step, chance, velocity decay and a shift of up to 24 semitones either way. SCALE and KEY hold its output to one of twelve scales. The display shows the pattern it will play. It follows tempo and time-signature changes, runs while the transport is stopped, and plays the same notes in every export.

arpeggiate these chords up and down in sixteenths
RecordMerge, replace or new clipA recorded pass can merge into the clip already there, replace it, or land as a new clip each pass. Input quantize is optional, on any grid.
InputsSettings per MIDI inputEach input has its own channel filter, forced channel, note range, fixed velocity and timing trim. A note range splits one keyboard between two instruments.
Computer keyboardPlay notes from the letter keysTwo rows of letter keys play the selected instrument. Z and X change the octave, C and V change the velocity, and chords work.
OutputsTo external hardwareA track’s MIDI can be sent to a hardware output instead of its built-in instrument.
VelocitySet, scale or humaniseThe engineer can set, scale, offset or humanise velocities across a whole clip, one pitch or a range of beats.
ArpeggioWritten into the clipAn arpeggio can also be written into a clip once, as ordinary notes that can be edited afterwards.

The Loop Deck.
A loop pad on every track.

Each track has a pad beside the timeline. A pad holds loops and launches them in time with the song, and a performance on the deck can be captured into the arrangement as clips.

The Loop Deck playing four loops: drums, bass, other and vocals. Each pad shows its waveform, a playhead, a Match Key button and its loop length of four bars.
LaunchQuantised launchClick a pad to start it and click again to stop it. While the song plays, a launch waits for the grid you choose, from half a bar to eight, and counts down on the pad.
Tempo and keyFollows tempo, matches keyEvery loop follows the song's tempo. Match Key retunes one to the song's key in a single press, and each pad has its own tune and gain.
ShelfSeveral loops per padDrop a second loop on a pad and it joins the shelf underneath. Click it to swap, on the grid.
SnapshotsStore and recall a combinationStore which loop every channel is playing, then bring the whole combination back with one click.
CaptureInto the arrangementArm Capture and play the deck. When you stop, everything you performed is in the arrangement as real clips, as one undo step.
RecordStraight into a padRight-click an empty pad to record a set number of bars from your input. The take starts looping the moment it ends.
Follow actionsNext, repeat or stopAfter 1, 2, 4 or 8 loops, a pad can go to the next clip, repeat or stop.
TimelineDrag in either directionDrag a pad onto a lane or a clip onto a pad. Play All and Stop All are in the header, and the engineer can load, launch, snapshot and capture for you.

Three example requests:
sketch, mix, deliver.

One request for each stage of a production, with what the engineer does in response.

01 · SKETCH

Tempo, key, chords, instruments.

82 BPM in D minor, an 8-bar jazzy ii-V-i, then drums, keys and bass

It sets the tempo and key, writes the chords onto the timeline, loads a drum rack with a kit and builds the synth patches. It can also resample a patch into a new sample.

02 · MIX

Auto Mix, then adjustments.

auto mix, vocal-forward and club loud

Auto Mix labels every track and builds a starting chain: levels, high-passes, EQ where tracks mask each other, gentle compression and a limiter at your loudness target: −14 LUFS for streaming, −16 for Apple and podcasts, −8 for the club, or a ceiling alone. Give it a direction by name (vocal-forward, punchy drums, warm master, wide pop, club loud, podcast clean), then steer in plain words: brighter, tighter, less of that. Press and hold to compare before and after.

03 · DELIVER

Reference, loudness, export.

match my reference, master to −14 LUFS, then bounce the stems

Load a reference track, or several, weighted for tone, width or loudness and compared section by section. Set a loudness target. Then bounce the master and each track through its own chain. Files go out as WAV, AIFF or FLAC, up to 32-bit float, and each one is read back and checked before the export is called done.

“I deliberately skipped an extra warm tilt: sub, bass and low-mid are all within 0.6 dB of each other at the top of the spectrum already, and tilting further would have been mud, not warmth.”
The engineer, explaining a mastering decision on a lo-fi track it had written and arranged in the same session.
master it for lofi, around −14 LUFS, warm with the top rolled off, ceiling −1 dBTP
master eq @ 8.0 kHz−2.5 dB
master limiterceil −1.0 · tgt −14.0

Single-step operations.
By click or by request.

Each operation below runs from the interface or from a request to the engineer. Both produce the same result and the same undo step.

The Duck device on a bass track, triggered by the drums: each drum hit shows as a spike, and the bass dips under it by 6 dB.
One press

Sidechain ducking

Drop Duck on the bass and press SC Auto. The engineer finds the track it is fighting with, sets that as the trigger and tunes the depth and timing. If you would rather choose, pick the trigger from the menu and it is ducking straight away.

duck the bass under the kick
Four tracks side by side after a split: vocals, drums, bass and other, each with its own waveform and controls.
One right-click

Split a mix into stems

Right-click any audio track and choose Split into stems. Vocals, drums, bass and everything else come back as four new tracks. It runs on your Mac. A song takes a few minutes in the background, and the first run downloads the model (about 166 MB).

split this into stems, then clean them up
freeze the synth tracks

Renders a track's whole chain to audio and switches its live processing off, which lowers the CPU load of a large session. Unfreezing restores it.

Freeze · on the track
tune the vocal

Finds the off-pitch notes in a take and corrects each one by its own error, with the joins faded. If a take can't be fixed cleanly, it tells you.

Tune Notes · one undo step
warp this loop to my tempo

Locks a clip to the grid and keeps it there when the tempo changes. A file with its BPM in the name does it by itself on import.

Warp to Tempo
tighten the hi-hats to sixteenths

Snaps a MIDI clip to the grid you name: straight, swung or triplet, as hard or as gently as you like.

Quantize · ⌘U
put the whole song in A minor

Changes the key of the entire song from the transport. The master ceiling still holds afterwards.

Transpose · key picker
give the vocal a reverb send

Adds a return track with a reverb and sets the send level. You hear it immediately, and you can rename it, adjust it or remove it.

Sends and returns
group the drums

Select the tracks and press ⌘G. They get a bus with its own fader and its own device chain.

Group to bus · ⌘G
keep what I just played

⌥⌘R turns the last 30 seconds of your input into clips on the track, even though record was not running.

Capture Last Take · ⌥⌘R
comp the best take and tidy up

Flattens your comp into a single clip and clears out the takes you didn't use.

Takes and comping
write a drum part at 92

Loads a drum rack with a 14-piece kit and writes a MIDI clip you can edit note by note.

Drum rack · piano roll
swing it a little

Apply one of 176 groove templates, or lift the groove from a clip you like, at any strength from 25 to 100 percent.

Grooves
print the chorus guitars to a new track

Renders the range you select, through the track's own devices, onto a fresh track beside it.

Bounce to New Track · ⌘B

Questions,
answered.

Requirements, the API key, privacy and limits. For everything else, read the manual.

What is Zammoth?

Zammoth is a digital audio workstation (DAW) for Apple silicon Macs with an AI audio engineer built in. It records, arranges, mixes and masters music. Every feature of Zammoth works by hand, and the engineer operates the same controls from requests typed in plain language.

What do I need to run Zammoth?

A Mac with Apple silicon running macOS 14 (Sonoma) or later. Zammoth is a native macOS app, and there is no Intel, Windows or Linux version. To use Zammoth's engineer you will also need an Anthropic API key.

Why does Zammoth need my own Anthropic API key?

Zammoth's engineer is an AI. It runs on Anthropic's Claude models, through your own key, so you pay Anthropic directly for what you use. You can watch the cost as it works, and a run stops at a spend limit that you set.

Create a key at console.anthropic.com and paste it once in Zammoth's Preferences ▸ Agent. It is stored in your Mac's keychain. New Anthropic accounts start with low rate limits, which rise as you use them.

Does my audio leave my Mac when I use Zammoth?

No. In Zammoth, recording, playback, rendering, analysis and stem separation all run on your Mac, and your project files stay where you keep them. A message to Zammoth's engineer carries the conversation, the measurements, and a description of the session: the names and settings of your tracks, clips, devices and markers, file paths (your home folder is written as ~), the names of the plug-ins installed on your Mac, and the names of your audio and MIDI devices. It never carries audio.

Can Zammoth's engineer wreck my mix?

Its changes are bounded. In a mix run, Zammoth's engineer can move a level by at most 6 dB per pass and boost an EQ by at most 6 dB. Every pass is measured again, and if one comes out worse the run goes back to its best. The master ceiling cannot go above −1 dBTP, and it can never switch off the safety limiter.

And what it does to your mix is one ⌥⌘Z away.

Do I have to use Zammoth's engineer?

No. Every feature of Zammoth works by hand, as in any DAW. When you do ask the engineer, Agent mode (the default) applies the change straight away. Ask mode plans it and touches nothing until you say “do it”.

Will my plug-ins work in Zammoth?

Yes. Zammoth hosts VST3 and Audio Unit plug-ins, both effects and instruments. Scanning runs in its own process, so a plug-in that crashes during the scan is set aside instead of taking the app down. If one crashes a session, it is quarantined on the next launch until you trust it again, and your unsaved work is offered back.

Does Zammoth work offline?

Everything in Zammoth except the engineer works offline, because the audio never depends on the network.

Join the beta list.Download Zammoth.

Download Zammoth and work with the engineer free for 3 days.

Download for macOS
esc
Try
325 commands · the engineer can run 325↑↓ to move · ↵ copies the key