Local files stay local

Audio and other source files

The working tools decode, inspect, edit, preview, and export files on your device. They do not send source audio, separated stems, artwork, lyrics, mixer settings, or generated downloads to mp3tools.io, Hugging Face, or another processing service.

Four-Stem Splitter model delivery

The Four-Stem Splitter needs a 166MB public HTDemucs model. It makes one off-origin GET request to Hugging Face only after you choose a valid file and press Download model & split four stems. That request downloads model weights; it contains no audio, POST body, account identifier, or mixer data.

The browser verifies the model's expected size and SHA-256 digest before using it. It may keep the verified model in Cache Storage so a later split can reuse it. If storage is unavailable or full, processing can continue for the current run without saving the model persistently. No song or stem is placed in that cache.

Temporary browser data

Selected files and generated results live in browser memory for the current page. Temporary download URLs are revoked when results are replaced or the page closes. Closing the page releases the worker and audio resources; clearing site data removes any cached model.

Routine requests and measurement

Like most websites, the hosting provider can record routine requests for pages and local assets, including an IP address, user agent, request time, and requested path. Hugging Face receives the separate public-model GET request under its own service practices. mp3tools.io does not include behavioral analytics in this release.

Scope of this notice

This page documents the implemented browser workflows and the Stem Splitter's model-delivery boundary. Contact, terms, and final legal wording remain subject to the project's launch approval process.