
Polaris is a music production app for Windows, built for electronic music: house, tech house, bass house, breakbeat, dubstep and drum & bass. Native instruments and FX, a piano roll with an AI copilot, and a team of AI agents that plan, build and argue about your track where you can see them.
Most AI music tools generate a finished file and hand it over. Polaris works the way a studio does. You brief the orchestrator; it briefs the team — Drummer, Percussionist, Bass Architect, Sound Designer, Composer, Hook / Topline, Arranger, Automation Director, Mixing Engineer, Mastering, Genre Stylist, DJ / Crowd Flow, Reference Match and Producer’s Review — fifteen agents in all, each with its own reference material and its own tools. Two of them are auditors rather than makers: Producer’s Review checks the work against the brief before it reaches you, and Reference Match measures your track against one you love. They read up before they plan, they place real notes on real tracks, and when two of them disagree — the kick and the sub fighting over the low end, say — the disagreement is shown to you with the rule that settles it. Nothing is hidden behind a spinner.
Models run locally on your own machine, or through your own API keys. Every tool call, every reference consulted and every decision is visible in the transcript.
Most AI music tools hand over a finished file. Polaris hands over a finished session: real notes on real tracks, in a project you can open, edit and undo. The agents plan the arrangement and build it section by section — intro, drop, breakdown and the fills between — and every step arrives as a proposal you keep or drop. Take the whole thing, or only the parts you like.
An AI that cannot hear is guessing. Polaris loads an audio model that listens to the rendered sound — not the MIDI, not the parameter values, the actual audio — and answers for it: is the kick punchy, is the sub fighting the bass, does the drop hit. It measures alongside: frequency balance, loudness, stereo width and tempo. So a verdict always has numbers behind it.
It auditions by ear, too. Ask for a sound and Polaris can render the same patch across a range of one control, listen to every version, and bring back the one that fits the brief. How Polaris listens.
Type dusty tape kick, short, no ring and Polaris generates the audio — not a sample-library search, but sound made on your own GPU, on your own machine. Then it listens to what it made and checks it against what you asked for, and generates again if it missed. Kicks, subs, risers, impacts, textures.
Point Polaris at a web page, a YouTube video or a document and it distils the technique and keeps it. Every note it saves carries where it came from and how far to trust it — a ladder from what you told it, through what it measured, down to what it read on the web — plus a revision number and a dated change log. Anything time-sensitive gets a review date and asks before it refreshes.
It learns you as well. Polaris studies your own projects and builds a profile of how you produce, which you can keep, compare and weight. It ships knowing a lot already: 244 written technique packs and 67 machine maps for house, tech house, bass house, breakbeat, dubstep, 140, drum and bass and neuro. How Polaris learns · the knowledge it ships with.
Every instrument and effect is built into Polaris itself — no plugin hunting, and every parameter can be driven by the AI, so a patch can be designed from a sentence and auditioned before it lands on your track. Each instrument comes with its own reference guide, so the agents know how to play it.
Two oscillators with warp, FM and wavefolding, plus a sub and a noise source; a filter pair you can run in series or parallel; one envelope you can assign to whatever needs to move; eight macros and a deep preset library to start from. A live display draws the sound as you shape it, and the AI can build a patch from a sentence and play it back before it touches your track.
Some basses sit in a mix; Infall’s are meant to take it over. Three oscillators fed through a wavefolder, so a clean tone grows teeth as you push it, and a display that shows the fold happening in real time. Built for transformer bass, neuro and any drop that has to move air. In development — the display is finished, the rest of the panel is on its way.
A step sequencer with performance, probability and timing lanes, a drum synthesiser underneath, and layering per pad. The AI Drummer plays it as MIDI: it generates candidates, checks them against the kick-and-bass rules, picks the best and commits a groove the rest of the arrangement shares.
Left-click adds a note. Swing and groove are baked into the MIDI you commit, and every edit is undoable, just like moving a clip. The copilot proposes onto your open piano roll instead of writing over it: a dashed overlay with keep or drop for each phrase, and never more than two takes — the first attempt and one alternative. Nothing lands until you keep it. Pitch, pressure and pan curves ride on the notes.
Levels, pan, mute, solo and bus sends, and automation you can draw on any parameter. All of it drivable by the AI, all of it yours to override.
Record audio and MIDI. Pitch bend and pressure from your keyboard record as curves on the notes rather than being thrown away.
Swing and groove bake into the MIDI you commit, and save to a library that follows you between projects rather than living in one.
Search, download and manage AI models inside the app, or point Polaris at your own API keys — there is no separate AI subscription. Polaris also speaks MCP, so you can add your own tools and give each agent its own set.
The Constellation shows a session as it happens: every agent an orb in its own colour, every message a light travelling a wire, every disagreement a card on the wire between them, and the orchestrator at the centre. You can pin an agent, read what it said, and replay the whole session afterwards.
A digital audio workstation is the software a track is made in: recording, sequencing, synthesis, mixing and arrangement in one place. Polaris is a DAW with an AI team built into it rather than bolted on.
A whole track. The agents plan the arrangement and build it section by section — intro, drop, breakdown and the fills between — as real MIDI and real audio on real tracks. Every step is a proposal you keep or drop, so you can take all of it or only the parts you want.
Yes. Polaris loads an audio model that listens to the rendered audio and answers for what it heard, and it measures spectrum, loudness, stereo width and tempo alongside. It listens to its own work before it shows it to you.
In two ways. It learns technique from pages and videos you point it at, saving each note with its source, a confidence rank and a change log you can read. And it learns your style from your own projects, as profiles you can compare and weight.
Windows 10 and later. A dedicated GPU is recommended for the visualisers and for local model inference.
No. Polaris hosts models on your own machine, and you can point it at your own API keys instead if you prefer a cloud model. There is no separate AI subscription.
Yes. Every AI action is a proposal you accept or reject, and the instruments, piano roll, mixer and arrangement work exactly as you would expect without touching the AI at all.
Electronic dance music and its bass-driven family — house, tech house, bass house, breakbeat, dubstep and 140, drum & bass and neuro. The knowledge the agents work from is written for those styles.
Polaris is in development and the beta is coming. Put your email in the form above and you will hear when it opens.
Polaris — music production software for Windows. Sign up for the beta.
We’d like to use Google Analytics to see how many people find Polaris. Nothing is stored unless you agree, and the site works exactly the same either way.