Open public beta · August 5, 2026
Privacy
This beta collects only what is needed to deliver and protect a training run: a random browser journey ID, a pseudonymous browser-scoped participant ID, source audio, selected settings, job status and diagnostics, generated model files, timestamps, and basic request information used for security and rate limiting.
Owner activity monitoring
The beta owner can see a private timeline showing pseudonymous visits, public pages viewed, broad referral categories such as Google or ChatGPT, audio selection, upload, training stages, completion, failure, and model-file downloads. This timeline does not contain source audio, IP addresses, raw audio filenames, full referring URLs, or search-query text. Journey events are retained for up to 90 days to evaluate reliability and understand where users need help.
How the data is used
The data is used to train your requested RVC model, show progress, provide downloads, prevent abuse, diagnose failures, notify the beta owner about important service milestones, and measure beta reliability and cost. It is not sold and is not used to advertise to you.
Storage and deletion
Successful source audio and temporary training files are removed after validation. Failed jobs may retain source audio long enough to support a retry. During the public beta, the job record and generated .pth, .index, and ZIP package are scheduled for deletion seven days after creation. Signed download links should not be shared. Your private model library is tied to this browser; clearing its site data may remove your access before the files expire.
Processors
The service uses cloud hosting, private object storage, cloud GPU providers, and—when enabled—a private owner-notification endpoint to process and store the data. Data may therefore be processed in locations selected by those providers.
Your choices
Do not upload sensitive audio that you are unwilling to send to cloud providers. Download your package before expiry. To request early deletion or report a privacy concern, contact the service operator and include the model name and approximate creation time; do not send the audio again.