Drop in a track, get clean stems back. Separation runs on your Apple Silicon GPU through HT-Demucs and Ultimate Vocal Remover — no upload, no account, no cloud. OhmSplit does one thing, entirely on the machine you already own.
Silo Systems builds infrastructure for people who would rather not negotiate with their own software. OhmSplit is the same discipline, pointed at one job — pulling a track into stems — done entirely on the machine you already own. Your unreleased masters never touch a server.
Separation runs on your Mac through the audio-separator runtime on your Apple Silicon GPU. The model weights download once from public open-weights mirrors, then OhmSplit works offline for the lifetime of the app — on a plane at 35,000 feet exactly the way it works at your desk.
No analytics SDK, no telemetry, no account, no crash reporter. The PrivacyInfo manifest declares no tracking domains. The track you drop in and the stems that come out stay in your Application Support container — exactly where you put them, never transmitted anywhere.
HT-Demucs and the UVR models are open-weights. The vocals, drums, bass, and instrumental tracks OhmSplit writes carry no per-stream licensing and no platform claim. Drop them into any DAW, sample them, sell them — OhmSplit doesn't appear in your release notes unless you put it there.
Pick two-stem — vocals and instrumental, via MDX-Net and MDX23C — or four-stem — vocals, drums, bass, and other, via HT-Demucs. The model runs in audio-separator on your Apple Silicon GPU and writes real WAV files to disk. There is no submit endpoint, no job queue, no upload. The first run downloads the weights from a public open-weights mirror; every run after that is offline.
Every stem gets its own waveform, play, solo, mute, and level — all playing in lockstep through a WebAudio graph. Solo isolates a stem, mute drops it, the faders ride the gain. Audition the separation the way you'll use it before you ever export a thing.
Try the mixer below. Flip between a two-stem and four-stem split, press play, and ride solo, mute, and the level faders — the same synchronized surface OhmSplit gives you over the real separated audio.
Visual demo · the app plays the real separated audio, in sync
Export each stem on its own, or grab the whole set as a single zip to a folder you choose. They're plain WAVs — no re-encode, no watermark, no upload step between the separation and your session. Drop them straight into Logic, Ableton, or whatever you already run.
Drop a WAV, MP3, FLAC, or AIFF onto the window — or pick one. One track in, clean stems out. No queue, no project setup.
Vocals and instrumental, via MDX-Net (Kim_Vocal_2) and MDX23C. The fastest path when all you need is the acapella and the backing.
Vocals, drums, bass, and other, via HT-Demucs. Full multitrack breakdown of any mix, written as real files you can drag anywhere.
Every stem gets its own waveform, play, solo, mute, and level — all synchronized through a WebAudio graph. Audition the separation the way you'll actually use it.
Export each stem individually, or grab them all as a single zip to a folder you choose. No re-encode, no watermark, no upload.
Models download once from public open-weights mirrors with live byte-progress. After that, every separation is offline.
If the separation weights are missing, OhmSplit tells you with a typed error — it never quietly hands you a degraded fake-split and calls it a stem.
Separation runs on your GPU via MPS — no per-track fee, no rate limit, no subscription throttle. Run it as many times as you like.
Subscription stem services rent you separation by the month and run your audio through their servers. OhmSplit is a single purchase, on your Mac, with unlimited local separations — because it all runs on hardware you already own. Buy once, keep splitting.
Coming to the Mac App Store ⌘OhmSplit is in pre-release for the Mac App Store. Join the waitlist and we'll send a single email when it goes live — no marketing list, no newsletter, no follow-up campaigns.
