Features

Cadence aims to be stupid-easy for newcomers and endlessly extensible for pros. Here’s what the app does today and how capabilities map to editions.

Compose

A low-latency Web Audio composer with a piano-roll editor, transport, metronome, and a rich instrument set — keyboard-first and built to stay smooth on a canvas timeline. Notation round-trips through MusicXML (OpenSheetMusicDisplay + VexFlow).

Cadence ships 64 built-in instruments — synthesis and physical-modeling voices spanning keys, guitars & plucked, bass, strings, brass & winds, leads, pads, mallets, and percussion — plus 5 drum kits. Every one is registered through the composer’s Plugin SDK, the same typed seam a third-party plugin uses to add its own instruments, formats, and commands.

Cadence piano-roll editor: violet note blocks arranged across a horizontal timeline beside a vertical piano keyboard, with a transport bar above. — The piano-roll editor — transport, timeline, and note blocks.

AI composition assistant

Suggest chords, harmonize a melody, or continue a phrase. The assistant runs entirely on-device in your browser (Magenta.js + TensorFlow.js), so it works offline and your music never leaves your machine. The free tier includes 50 generations per day; paid tiers lift that cap. The assistant sits behind a typed provider seam (part of the Plugin SDK), so the runtime can evolve without changing the UI.

On-device only, today. There is no server-side AI generation service in the backend. All AI generation currently happens in the browser.

The Cadence AI assistant panel with actions to suggest chords, harmonize a melody, and continue a phrase, plus temperature and length controls. — The AI assistant — suggest, harmonize, and continue, on-device.

Stems

Split a mix into its constituent sources — bass, drums, vocals, guitar, keys, synth, and other — through an authenticated, owner-scoped asynchronous job pipeline backed by an ONNX/Demucs separation worker (with a deterministic band-split fallback in CI/dev). Upload a mix, the job runs server-side, and each labeled stem returns as a WAV you can preview and download. Stem separation is a paid entitlement (free users see an upgrade prompt). If you later downgrade to Free, Cadence stops new separations, but the stems you already created while subscribed stay yours — you keep listing and downloading them.

Phase 1 ships the standalone separation surface (upload → preview → download). Bringing a separated stem back into the composer as an editable mixer track with its own solo, mute, and gain is a planned Phase 2 follow-up.

Collaborate

Edit a project together in real time. The client uses Yjs (a CRDT) with awareness and presence over a WebSocket relay — a first-party endpoint inside the API, not a separate service — so concurrent edits merge without conflicts. Projects are owner-scoped and shared through links that carry an owner, editor, or viewer role, resolved and enforced server-side (viewer writes are dropped before fan-out). Collaboration is a paid, seat-limited feature and stays inert until a session is explicitly activated.

MIDI and MusicXML import/export for interop with other tools.
WAV (lossless PCM) and MP3 audio export. MP3 encoding runs on-device via the LAME encoder (credited under Acknowledgements).
Portable .cadence.json projects.
Shareable listen links — a client-side snapshot, no account required.
Offline-first local storage that syncs to your account’s cloud projects on sign-in.

The Cadence composer toolbar with import, export, and share actions, including MIDI import/export. — Import, export, and share — right in the composer toolbar.

Extend it

The composer is built on a typed, in-process Plugin SDK. A plugin can contribute instruments, audio effects, import/export formats, AI providers, commands (with their own keyboard shortcuts), and sidebar panels — through the same host the built-ins use. See the Plugin SDK guide.

Editions

The free tier is a real studio — full composer, on-device AI, and MIDI / MusicXML import-export — with an audible watermark on exports and a few limits. Paid tiers remove the watermark and lift the caps. The table reflects what the backend enforces today, mapped onto the editions shown on the pricing page.

Capability	Free	Pro
Composer, piano roll, mixer	✅	✅
On-device (offline) AI assistant	✅	✅
AI generations per day	50	Unlimited
MIDI & MusicXML import / export	✅	✅
WAV & MP3 export	✅ (watermarked)	✅
Watermark-free export	—	✅
Cloud projects	Up to 10	Unlimited
Live collaboration (seats)	1	5
Stem separation	—	✅

What’s enforced today. Cadence’s entitlement engine implements two enforced levels: Free and Pro. Pro unlocks watermark-free export, unlimited projects and AI generations, stem separation, and 5 collaboration seats. AI runs on-device in both tiers. Rendered-audio export ships as both WAV (lossless PCM) and MP3. Pricing is indicative during active development; the free tier is always free. The subscription tier is a claim (default Free) resolved through the entitlement seam described in Authentication.