36 modelled sources — trumpet, violin, cello, sax, guitar, piano, drums — that respond to how the part is written. Not samples triggered by a keyswitch. A model that learned the instrument.
An in-page player for all 36 sources is coming here. Until then, each instrument with a published sample has its own page below.
A sampled instrument is a set of recordings. Someone played a trumpet at three dynamics across its range, and the plugin picks the nearest recording and transposes. That is why programmed brass gives itself away: turn a soft sample up and you get a quiet recording played loudly, not a player pushing harder.
Modelling the instrument instead means learning how it behaves — how the harmonic balance shifts as a note is pushed, how the attack changes with articulation, what the top of the range costs the player. A crescendo becomes one continuous change rather than three recordings crossfading into each other.
So if you have been looking for a realistic guitar VST, a convincing strings patch, or an orchestral library that does not need 200 GB on disk and a keyswitch map to sound played, this is the other approach to that problem. Nothing to install, nothing to load.
Playing louder is not turning it up. A forte note has more high-frequency content, a harder attack, and on brass an edge that only appears under pressure. Velocity here moves timbre, which is the difference between a part that sounds played and one that sounds triggered.
The same is true of articulation. Pizzicato is not arco with a short envelope — it is a different excitation of the string, and it is trained as its own instrument. So is muted trumpet, for the same reason.
Bring a MIDI file and pick a source, and it is rendered as written — your notes, your timing, your velocities. Or describe the line you want and let the model write it, then keep the MIDI and change your mind about the instrument later.
Either way the result is audio you can drop into a session, or a part that travels with the rest of the arrangement into Logic or Ableton.
There is a native Audio Unit — dø1 — that opens the session inside a plugin window in Logic, so the work does not have to live in a browser tab. Parts rendered here land in a Logic or Ableton project natively rather than as a bounced stem, which means they arrive as tracks you can still edit.
The sources run in the browser and in the app rather than as an instrument VST you install. There is a native Audio Unit, dø1, that brings the session into a plugin window in Logic. Parts you render arrive in a Logic or Ableton project as editable tracks.
A sample library plays back recordings and crossfades between dynamic layers. A model learns how the instrument behaves, so a crescendo is continuous rather than three recordings fading into one another — and there is nothing to download or load.
36 sources today: strings (violin, viola, cello, double bass, arco and pizzicato), brass (trumpet, muted trumpet, trombone, french horn, tuba), woodwinds (flute, clarinet, oboe, bassoon, saxophone), keys, guitars, voices, drums and percussion, plus string, brass and wind ensembles.
The list is the model's trained vocabulary, so a missing instrument is a training question rather than a setting. It is worth asking for — the architecture is not specific to these instruments.
Yes. Velocity moves timbre rather than only level, and articulations that genuinely change the sound — pizzicato, mute — are separate sources rather than effects applied afterwards.
There is a free tier. Heavier use and saved sessions need an account; pricing is on the pricing page.