Play a recording like an instrument — by its sound, not its timeline. Chop a source into a corpus of grains, describe each by a few features (how loud, how bright, how noisy), and scatter them in that feature space. Then navigate: point at "bright and loud" and the synth grabs whichever grains sit nearest and plays them; drag across the pad and you're recombining the material by timbre. It's the idea behind CataRT and audio mosaicing. Drop your own sound to build a corpus from it — nothing is uploaded.
← dark · bright → · ↑ loud · quiet ↓
press play and drag the pad — play notes (a s d f …) to transpose the grabbed grains