Flowtonik · Nick Brandes

Context

Flowtonik is an AI-powered digital audio workstation for Mac. It does everything a DAW has to. It records and arranges audio and MIDI, mixes, runs instruments and effects, and it puts a conversational assistant alongside the session that can act on it directly.

The product’s own framing is human-led, AI-accelerated, and it’s positioned explicitly between two things that already exist: traditional DAWs, which give deep edit and mix control but make you do every step by hand, and AI music tools, which generate fast but leave you little session control and few professional mixing workflows.

Say who this is for, and what they were using before.

The problem

Both halves of that market fail the same person in opposite directions.

A traditional DAW asks you to already know what a 4-band EQ does, which band to cut, and by how much. The interface offers control but no opinion. AI music tools invert it: they hand you a finished-sounding result and almost no way in. When the vocal sits wrong, there’s nothing to grab.

Designing around a model

The hard part of this product isn’t the DAW. It’s that a model is making consequential decisions about someone’s creative work, and producers are not inclined to accept those on faith.

The AI’s output is ordinary session state. When the assistant mixes, it doesn’t return a rendered file. It puts a 4-Band Equaliser and a Compressor on the tracks, with real parameters, in the same plugin slots you’d have used yourself. Every AI decision is editable, automatable and removable by hand afterwards.

The 4-Band Equaliser plugin open over the Flowtonik mixer: a spectrum analyser behind an EQ curve with four colour-coded draggable band points, and matching columns of Gain, Freq and Q knobs below — The 4-Band Equaliser. Each band's point on the curve is colour-matched to its column of knobs, so the graph and the numbers are obviously the same four objects rather than two parallel control schemes.

This is the thesis made concrete, and the plugin windows are where you can check it. The assistant reports adding a compressor with a threshold and a ratio; the compressor it adds is this one, with those values, sitting on that track, with its curve redrawn accordingly. There is no separate “AI version” of the mix to reconcile against your own.

The Compressor plugin open over the mixer: an input/output transfer curve with a draggable knee point and a gain-reduction meter, beside knobs for threshold, ratio, attack, release, output and sidechain gain — The Compressor pairs a transfer curve you can drag with numeric knobs for the same parameters. The graph is for shaping, the numbers for reading back exactly what happened.

Both plugins follow the same pattern: a direct-manipulation graph and a numeric readout of the same parameters, side by side. You can pull the EQ curve into shape by eye, then read off 1304 Hz. That matters more than usual here, because when the assistant makes the move rather than you, the numeric readout is how you audit it.

It shows its work in the units of the craft. The assistant reports what it did per track and then why, in one line: gentle glue on the drum bus, fast attack and slower release, punch and sustain without squashing transients. That’s written for someone who could have made the move themselves, which is what makes it checkable rather than merely reassuring.

Original / Mixed is a single toggle. The comparison a producer actually wants (before against after) is one control, not an undo history to reconstruct. It makes disagreeing with the model cheap, and cheap disagreement is what makes trusting it possible.

Suggestions are concrete, not open-ended. The empty chat offers “Turn up the drums by 3dB”, “Balance the levels for all tracks”, “Change tempo to 130 BPM”. A blank prompt box would have tested whether users can guess the capability surface. These answer it instead, phrased as specific operations, which also teaches the register the assistant responds to.

The rest of the interface

The Flowtonik mixer: seven colour-coded channel strips with plugin slots, input and output selectors, pan, gain, level meters and mute/solo, beside the AI chat panel — The mixer inherits the arrangement's colour and track order. The chat panel is signed out here. It occupies the same dock in both views, so the thing that changes between arrange and mix is the session, not where you go to ask for help.

Colour is the identity thread. A track’s colour is set once and carries across the arrangement and the mixer. The green of Drums heads its channel strip, the magenta of Vocal Chops heads its own. Finding a channel becomes recognition rather than reading, which matters when tracks are named from generation prompts rather than by hand.

The arrangement absorbs the inspector. Input, output, volume, pan, track type and the plugin chain sit in labelled columns (Track, Plugins, I/O) on the arrangement itself, beside the waveforms. Most DAWs hide those behind a separate panel or make you switch to the mixer. Inline, the common adjustments never cost a context switch, and the assistant’s plugin additions are visible the moment they land rather than one view away.

Song structure is a first-class lane. Above the tracks, a Sections row sits alongside Bars|Beats and Min:Secs. Most DAWs leave structure implicit in where clips happen to fall. Naming it gives the producer and the assistant a shared vocabulary for the part of the song being discussed.

Provenance survives in the track name. Generated clips keep what was asked for: dark,_heavy_choppe…, synth_pad_1776116…, vocals_1776117067…. It’s unglamorous, and it means six months later you can still tell what you requested.

Where it landed

Shipping on macOS with a free tier that includes AI mixing, no card required.