Questions, answers, and where to report problems

Download the Windows .msi from the setup guide and install it, then run the one-time setup-backends.ps1 to fetch the AI models. The Quick start there walks through it step by step.

The installer isn't code-signed yet, so SmartScreen shows a blue “Windows protected your PC” screen. Click More info, then Run anyway. Code signing is planned.

The installer ships the app and engine, but not the ~10 GB models. Click See what's missing on the Generate screen for a live checklist, or run setup-backends.ps1 once (see the setup guide); it installs them into Music\Cadence\vendor, where Cadence looks. Then fully restart the app.

Open Settings → About in the app and click Check for updates. If a newer release is out, it shows a Download button; install it over the top. Your library and trained voices are stored separately and carry over.

Yes. Pick a voice (not Instrumental), open + Advanced under the prompt, and paste them into the Lyrics box - there's an Insert example template to start from. Shape the song with [Verse], [Chorus], and [Bridge] tags; keep lines short and singable, and leave genre and tempo in the prompt. Leave the box empty to let Cadence write them.

Only for the initial downloads - the app, the models, and dependencies. After that, generation and voice training run entirely offline. The exceptions are optional cloud providers (Claude for lyrics, Suno or ElevenLabs for generation), which you have to deliberately enable.

No. Recordings are saved as WAV files on your machine, the voice model is trained on your machine, and both stay there. Cadence has no server, no account system, and no telemetry.

An NVIDIA GPU with 8 GB of VRAM or more is strongly recommended - generation takes seconds on a modern card. Everything can fall back to CPU, but expect generation to take minutes rather than seconds.

Prompts and lyrics work in more than fifty languages, including mixed-language requests - for example, a prompt in English asking for lyrics in Malayalam. Quality is strongest in widely spoken languages and improves with a Claude API key for the lyric-writing step.

No. The default setup is fully local: Ollama handles lyric writing and ACE-Step handles music, both free and on your machine. A Claude, OpenAI, or Gemini key is an optional upgrade for lyric writing - pick the provider in Settings and paste the key right under it. Suno and ElevenLabs keys are optional alternatives for generation, later.

Ten to fifteen minutes of clean speech, captured in short takes with a guided reading script inside the app. The script runs to 75 lines, enough to reach the target in a single pass without reading the same sentences over and over. You review each take and re-record anything that isn't right before training starts.

Each song gets its own dated folder under Music\Cadence\output, named for when you made it, so they sort in the order you created them. Inside is track.wav, the finished song, alongside the working files it came from. There is an MP3 export option, and you can send songs anywhere you like: set a “Save to” folder under + Advanced on the Generate screen, and optionally tick Save stems separately to get the vocals and instrumental as their own files. Tracks appear in the app's Library, but they're ordinary files - copy, move, or back them up like anything else on your disk.

Yes, both ship with every release, and both are under testing. macOS gets a .dmg (Apple silicon only, not Intel) and Linux a .deb for Debian and Ubuntu. They are built and smoke-tested on their own operating systems, so the app and its engine start and run - but generating music on them is not verified yet: the models want an NVIDIA GPU, so Linux needs CUDA, and Apple silicon support is unconfirmed. Windows remains the tested path. Grab them from the setup page.