What happens to your music and your data
Last updated 15 October 2026. Draft: text in orange boxes must be filled in before launch.
Short version
- The file you upload is deleted as soon as it has been converted. In preview mode we then keep only a 20-second excerpt and its mastered copy; in whole-track mode a converted copy of your track and its mastered version (or, if you ask for variants, up to four), so that you can play and analyse them. Everything is deleted one hour after it is ready.
- We do not train any model on your music and we do not send it to anyone else. Mastering runs on our own server.
- No accounts, no cookies, no analytics, no advertising, no third-party fonts or scripts.
Who is responsible
[Legal entity name, address, registration number] (“MediaAtlas”). Contact for anything on this page: [privacy contact email].
What we process, and for how long
| Data | Why | Kept |
|---|---|---|
| The audio file you upload | To convert it for mastering (whole track) or to cut the excerpt (preview) | Deleted right after that step (normally within a minute or two). If the server crashes first, deleted when processing resumes. |
| The audio we work on (a converted copy of your track, or a 20-second excerpt), the mastered copy or copies, and the measurements made from them | To play and analyse the result in your browser | One hour after the result is ready, then deleted. The page loads them with a link that only your browser knows. The spectra and spectrograms you see are computed in your browser. |
| The prompt you type | To tell the model what to do | With the preview data, one hour. |
| Your file name and IP address | Queue and abuse limits (one track at a time per visitor, a few previews per hour) | File name and IP are removed from the job as soon as it finishes. The IP is also held in memory only for up to 24 hours for the rate limit, and is lost when the server restarts. |
| Stems of one song you upload (several files or a zip) | To analyse them, correct them and add them up into one mix that is then mastered | The stems are kept on our server so that you can change levels and mix again without uploading them once more, together with an analysis file made from them (measurements and a low-resolution copy of the signal, used for the mono check). Both are deleted one hour after your last upload, analysis or mix, or immediately when you press "Delete stems now" (if a mix of them is still waiting, as soon as it is finished). Each result keeps its report, its mix excerpt and the same excerpt as the plain sum of the stems for one hour. |
| A count of each finished job: time, mode (preview or whole track), length of audio, how long it took, and whether it worked | To see how busy the service is and how much hardware it needs | Kept for 90 days, then deleted. It holds no IP address, file name, prompt or audio. |
| API customers: the name of the key, and for every job its time, mode, minutes of audio, duration and outcome | To apply the limits of the key, to invoice, and to see whether the service works | Kept while the key exists and for the period tax law requires for invoices [period]. Never the audio, the file name or the prompt. |
| A daily and an alert message to the operator (Telegram) | To notice outages and see how busy the service is | Only counts and times, and the names of API keys. No audio, file names or IP addresses are in them. Telegram keeps its own messages. |
| Proof of mastering: the ID of the master, the time, a fingerprint (hash) of the mastered audio and of the source audio, the settings used (the wishes and the loudness target), and our signature | To be able to show that a file was mastered by Master Blaster and when. The ID and the signature are also written into the tags of your master. A fingerprint cannot be turned back into music. | Kept in a register: 5 days for the free demo, 10 years for masters made with an API key (the register entry then ends). It holds no audio, file name, IP address or account. Whoever has a file or an ID can ask the verify page whether it is in the register. |
| Tags inside your file (artist, title, …) | Copied into the mastered copy so they are not lost. We add our credit line and the proof of mastering to them | Same as the excerpt: one hour. |
| A line per finished job: time, preview or whole track, length of the audio, how long it took, whether it worked and, if not, a short error text | To see how busy the service is, what a track costs us and what fails | Kept without a limit. It holds no IP address, file name, prompt or audio and cannot be linked to you. |
We do not write request logs with IP addresses. The web server in front of this site is configured without access logs [confirm in the deployed Caddy config].
What we never do
- Train or fine-tune any model on your uploads.
- Share, sell or publish your audio or your data.
- Keep a copy of your track beyond the hour above.
Cookies and tracking
This site sets no cookies and keeps only a few display preferences in your browser (player volume, waveform or spectrogram view). They stay on your device and are never sent to us. Fonts are served from our own server, so your visit does not send your IP address to Google or anyone else.
Payments
Checkout is not live yet. When it is, payments will be handled by a payment provider [provider, e.g. Stripe], who will receive the details needed to take payment. We will update this page first, including how long order records are kept (tax law requires us to keep invoices [period]) and how long paid masters stay available for download.
Legal basis and your rights
We process the data above to provide the preview you asked for and to keep the service running and abuse-free [legal basis, e.g. performance of a contract and legitimate interest; have this checked]. Under the GDPR you can ask us for access to, correction of or deletion of data about you, object to processing, and complain to the Information Commissioner of Slovenia (ip-rs.si). Because previews are deleted after an hour, there is usually nothing left to access or delete; write to the address above with the time of your upload if you want us to check.
Who runs the model
Mastering uses SonicMaster, an open model run on our own hardware. Your audio is not sent to its authors or to any cloud service.
Changes
If this changes, the date at the top changes. We will not make the policy less protective for audio you have already uploaded.