项目的 README,来自 github.com/SimoHypers/limusic。合并为一个 markdown 文件:README.md。
A native desktop YouTube Music client. Rust + Tauri, ad-free, no Electron.
Limusic talks directly to YouTube's internal API and plays audio through libmpv: no bundled browser runtime, no backend server, no ads in the audio. It started as a desktop rebuild of the playback engine behind Metrolist, an Android YouTube Music client, and grew from there.
|
Listening
|
Lyrics
|
|
Your library
|
Together
|
|
On your desktop
|
Make it yours
|
![]() |
![]() |
![]() |
![]() |
| Platform | File | Notes |
|---|---|---|
| Windows | -setup.exe |
Self-updating |
| Windows | .msi |
Plain installer, no auto-update |
| Linux | .AppImage |
Self-updating, libmpv bundled. Needs glibc 2.39+ (Ubuntu 24.04+, Debian 13+, Fedora 40+) |
| Linux (Ubuntu/Debian) | .deb |
No self-update. Needs Ubuntu 24.04+ / Debian 13+; apt pulls libmpv and webkit2gtk in for you |
| Linux (Fedora/RHEL) | .rpm |
Needs mpv-libs installed (sudo dnf install mpv-libs). No updates, redownload each release |
| macOS 15+ (Apple Silicon) | .dmg |
Self-updating. Unsigned, so the first launch needs xattr -dr com.apple.quarantine /Applications/limusic.app |
| macOS (Intel) | none | Build from source, see docs/BUILD-PLATFORMS.md |
Community-maintained repositories, packaged and updated by their maintainers rather than by this project:
| Platform | Source | Notes |
|---|---|---|
| Linux (Arch) | AUR | yay -S limusic-bin, or yay -S limusic-bin-prerelease for release candidates. Maintained by @xiryuudev, updates through pacman |
| Linux (Fedora COPR) | COPR | sudo dnf copr enable oguzkarayemis/limusic then sudo dnf install limusic. Maintained by @oguzkarayemis, updates through dnf |
| Linux (openSUSE Tumbleweed) | OBS | sudo zypper ar -p 100 https://download.opensuse.org/repositories/home:/itachi_re/openSUSE_Tumbleweed/home:itachi_re.repo then sudo zypper install limusic. Maintained by @itachi-re, updates through zypper. The repo carries the maintainer's other packages too, so -p 100 keeps it below the distro repos |
Last.fm lives in the title bar, next to the window controls, and ListenBrainz connects from Settings ▸ Scrobbling. Both get the same cleaned-up track (the tab's preview shows what each play counts as).
playing_now on start and one single listen at the same
halfway-or-four-minutes point. A new MetaBrainz account has to verify its email first, or
ListenBrainz turns every listen away (the Scrobbling tab says so).Building from source? Last.fm needs your own API credentials, and they are not in
the repo. Get a key at last.fm/api/account/create
and put it in src-tauri/lastfm.keys:
LIMUSIC_LASTFM_API_KEY=your_key
LIMUSIC_LASTFM_API_SECRET=your_secret
Without that file everything else still builds and runs; the Last.fm button just reports that it isn't configured.
Open the panel with the microphone button in the player bar, next to the queue button. It takes the same side of the window as the queue, so opening one closes the other.
Nine sources are asked in order: Limusic Lyrics, Boidu, Unison, LyricsPlus, LRCLIB, YouTube Music's own lyrics, then Netease, QQ Music and Kugou. Synced lyrics from any of them beat plain text from one higher up, and plain text is the fallback when nobody has a synced version. Matching is keyed on the track's exact length, because popular songs exist as several cuts and the wrong one drifts a few seconds out. Results are cached locally, so replaying a track is instant.
The first four have per-word timings, which is what lets a line highlight word by word as it's sung. Reorder the sources or switch any of them off in Settings ▸ Playback ▸ Lyrics sources, and turn on Prefer word-by-word lyrics there to have word timings win over line timings from a source above. To pick another source for one song, click the source name under the lyrics. Netease additionally supplies translations, shown under each line where it has them.
Note that YouTube Music's lyrics are licensed per region and are missing entirely in some countries. Where that's the case, the other sources do all the work.
Limusic Lyrics is the project's own lyrics service, filed by video, so a lyric is timed to the exact upload you're playing. It is filled by the people who use the app.
To add a song, click the source name under its lyrics and pick Sync these lyrics, or use Settings ▸ Contributing ▸ Open the lyric editor. Start from the lyrics another source already found, paste the text (pasted LRC or TTML keeps its timings), or open a TTML or LRC file. Then the song plays one line at a time, on a loop and slowed down if you like, and you tap a key as each word starts. Duets get a singer per line and background vocals can be marked word by word. A preview shows the result at full speed before you send it.
No account is needed: your first lyric makes an anonymous one. Link Discord or GitHub in Settings ▸ Contributing to put your name on your lyrics. Depending on the account, a lyric goes live at once or after a reviewer approves it. Everyone can vote on a lyric or report it from the source picker, and the lyrics credit their authors under the last line. Contributors earn points on the leaderboard.
Karaoke takes the vocals out of a song right on your computer, using AI models from Ultimate Vocal Remover. Nothing is uploaded. It is still marked experimental while it's being tuned.
Turn it on with the switch under the lyrics. The first time, Limusic asks before downloading a model (Light, Balanced or Best, swappable later in Settings ▸ Playback ▸ Karaoke) and the runtime that runs it. After that the playing song and the next two are prepared in the background, and the vocals drop out the moment a song is ready, right where you are in it. Prepared songs are kept on disk up to a size you choose, so they play straight away next time.
Light and Balanced keep up with playback on a processor; Best is meant for a graphics card. The Processor setting picks one: on Windows Auto uses the graphics card through DirectML, on Linux it's an opt-in through Vulkan (Intel, AMD and NVIDIA), and on Apple Silicon an opt-in through CoreML. Songs over 15 minutes keep their vocals, and karaoke is off while a proxy is set.
Synced listening with friends. Everyone streams their own audio from YouTube; the room only relays play/pause, seeks, track changes and the queue. Rooms have join codes and the host approves every join and every track suggestion.
Rooms run on the project's server by default. To host your own relay:
cargo run -p sync-server # plain WebSocket on 0.0.0.0:8080
Front it with something that terminates TLS (Tailscale Funnel, Cloudflare
Tunnel), then paste the wss:// URL into the Listen Together panel in the app.
Limusic is translated on Weblate, who host it free for libre projects.
Nineteen languages ship in the app today, and the badge above shows everything else in flight.
Translate on Weblate, not in a pull request. Weblate keeps its own copy of
the catalogs, so a hand-edited fr.json merged here puts the two out of sync
and the next batch of real translations arrives as a merge conflict. Weblate
also shows you the English original beside each string, flags translations that
went stale when the English changed, checks that placeholders like {count}
survived, and opens the pull request for you. Anything untranslated falls back
to English in the app, so partial work is safe to submit.
en.json is the exception: it changes by hand, in whichever pull request
changes the UI. Switching a finished language on in the picker takes a small
code change too, see CONTRIBUTING.md.
Fedora:
sudo dnf install mpv-libs mpv-libs-devel webkit2gtk4.1-devel \
gcc gcc-c++ make openssl-devel librsvg2-devel
cd ui && pnpm install && cd ..
cargo tauri build
Windows and macOS instructions live in docs/BUILD-PLATFORMS.md.
n parameter) and by BotGuard attestation. Limusic runs that
JavaScript where it expects to run, in a real webview, hidden, and never lets
any of it touch the UI process.Limusic is free and stays free. If it earned a coffee, ko-fi.com/simohypers is where to leave one.
This project is not affiliated with, funded, authorized, endorsed by, or in any way associated with YouTube, Google LLC, or any of their affiliates and subsidiaries.
All trademarks, service marks, and intellectual property rights referenced in this project belong to their respective owners.
Karaoke uses the UVR-MDX-NET-Inst_HQ_3, UVR-MDX-NET-Inst_Main and UVR_MDXNET_3_9662 models from Ultimate Vocal Remover by Anjok07 and aufr33 (MIT) and, on Windows, ONNX Runtime and DirectML from Microsoft. See THIRD-PARTY-NOTICES.md.