• A three-panel first-run card: add your Music folder or any other, pick the language, theme and accent, and see every feature that reaches the network before you switch it on. Settings ▸ About reopens it
• Parallel folder scanning, live folder watching, and incremental re-scans
• Cue sheets: a whole-disc rip with its .cue lists as the disc's tracks, each with its own title, artist and number, and a playlist holding the whole disc holds its tracks in its place. A FLAC carrying its cue sheet inside, as a CUESHEET tag or its own CUESHEET block, splits the same way. Opening the .cue, or dropping it on the queue, queues the disc in order, and Edit Tags on its tracks saves into the .cue itself, or into the FLAC's CUESHEET tag
• A file or a .cue opened from outside your library folders plays without joining the library, and stays as long as a playlist or the queue holds it
• Content hashing (BLAKE3), so a moved or renamed file keeps its play counts, favorites, and place in the queue
• A duplicate finder under Settings ▸ Library that groups the songs held more than once, labels how sure each match is, and picks the best copy to keep. Remove takes the spares out of the library, moving their plays, rating, heart and playlist places onto that copy. Their files stay on disk, and later scans skip them until you bring them back from Files kept out of the library
• Full-text search (SQLite FTS5) over tracks, albums, artists, genres, and the people credited on a track: accent-insensitive, relevance-ranked, with a top-result card and recent history. Every list's filter box searches the same fields
• My Library gathers everything into one page with five tabs (Songs, Albums, Artists, Genres, Playlists); opening an entity grows the tab band into its banner rather than navigating away
• Favorites and Recently Played, each a hero banner over sortable lists and browsable card grids
• Browse by folder, as a detailed list or a grid of cards
• Star ratings, play and skip counts, natural sort, resizable and toggleable columns
• Multi-selection on every list and card grid: tick one to start, click to add, shift-click for a range, and right-click to queue, rate or tag the whole set. Esc clears it
• Copy from any right-click menu: a song's title, artist, album, "Artist - Title" or file path, an album's name, or a station's name, stream URL or website, one line per selected item
• Undo for removals: tracks taken out of a playlist or the queue, deleted playlists, a smart playlist's rules once it is converted, Clear queue, a play that replaces a queue you built, favorites, radio stations, the equalizer's Reset, and recent searches' Clear all. A notice names what went and offers Undo, or Ctrl+Z while it is up, and waits while the pointer is over it or Melodia is in the background. Settings ▸ Accessibility turns undo off altogether, per kind, or for single items, and sets how long the notice stays
• Tag editing for one track or many, cover art included, written straight back to the files. Artists, genres and the ten studio credits are each an editable list, beside work and movement, key, mood, ISRC, label, catalog number, barcode and medium
• Manual and smart playlists, the latter rule-based and resolved live until you convert one to a regular playlist that keeps the tracks it holds; .m3u8 import and export, several at once as one .zip that imports back without extracting, drag-and-drop import, and reordering that moves a whole selection
• A database backup before every schema migration, three kept

• Gapless, including the AAC encoder delay and padding read back from iTunSMPB or the MP4 edit list, and the trailing padding a Matroska file states
• Crossfade (1–12 s) across two decks with a clip-safe ramp, optionally skipped between same-album tracks, and under exclusive output too wherever the next track needs no change of device format
• 10-band equalizer (31 Hz – 16 kHz) with preamp, presets, and a soft-knee limiter, from the player bar's overflow menu or Settings ▸ Effects
• ReplayGain in track or album mode, with preamp and peak-based clip prevention, reached the same two ways
• A queue with shuffle and repeat in a resizable side panel, which floats over the page once the window is too narrow to dock it. Playing from any list queues that list behind your pick
• Drag tracks, albums, artists, genres, playlists, or files from your file manager into the queue or a playlist where you let go; a selection reorders together in the queue, Up Next, and the mini-player
• Hold Previous or Next to scan through a track in audible jumps that grow the longer you hold, and hold Play/Pause to stop
• Full-screen Now Playing with an up-next list and a spectrum, mirrored, or waveform visualizer tinted to the album's own colors
• Right-click the playing song in the player bar for the menu its row has in a list
• Playback speed 0.25×–2.0×, a playback-linked sleep timer, resume on startup, media keys
• Sample-rate conversion wherever a file's rate isn't the output's or the speed isn't 1×, band-limited or a CPU-light linear (the Windows default) under Settings ▸ Output. It carries across gapless track changes and is skipped where the rates already match
• A responsive mini-player that rearranges as you resize, from a one-line strip to a column with Up Next or lyrics, and reopens as the mini-player if you closed Melodia from it

• Exclusive output on Linux and Windows, off until you choose it under Settings ▸ Output. Each file's own sample rate and bit depth go straight to the device with no system mixer in between: an ALSA hw: device on Linux, asked of the sound server so it is handed back cleanly, and WASAPI exclusive mode on Windows
• A Signal Path panel grades every stage from the file to the device and sums it up as Bit-perfect, Enhanced (only effects you chose), Converted, Lossy, or Fallback (a refused claim). Make Bit-Perfect switches off whatever you chose that changes the sound
• A quality chip in the player bar carries that verdict and the device's rate under exclusive output, and opens the panel in one click
• Output Device (Exclusive Device on Linux) picks the sound card to claim, and on Windows routes shared output too. An unplugged device hands playback to the default and is taken back as soon as it returns
• A device that refuses the claim never costs you the music: playback carries on shared, and a notification and the panel say why
• Unsupported Sample Rates can resample a file to a rate the device has rather than give up the claim, staying gapless wherever the device's rate doesn't change
• Dither wherever samples Melodia changed are narrowed to 16 or 24 bits, while untouched samples still pass bit-exact
• Hardware Volume turns the device's own volume control instead of scaling the samples, so the slider keeps the sound bit-perfect and the system's volume keys move it too
• Give the Device Back When Paused hands the device to other apps after five minutes paused; pressing play takes it again
• A choosable buffer period, a short silence for DACs that mute while switching sample rates, and on Windows a polling mode for USB devices that stutter
• In shared mode on Linux, Match the File's Sample Rate reopens the device at each track's own rate wherever the sound server allows it
• Unplugging the device or restarting the sound server picks playback up on the new output at the same position

• A timed sheet in Now Playing, in place of Up Next, that follows the song line by line; click a line to seek to it. Switch it on from that view's header or Settings ▸ Lyrics
• Off by default, and off means off: while lyrics are switched off, nothing looks a sheet up or keeps one in memory
• Read from a .lrc beside the track or from the file's own lyrics tag, so a library that already carries them needs no online lookup
• Online lookup is off until you switch it on: nothing contacts lrclib.net until you do, and its answers are cached under your data folder rather than written into your files
• Romanization under a line in a script you don't read, around a hundred scripts in all, with engines matched to Korean and Japanese. On by default, and silent for a sheet already in Latin letters
• A bilingual sheet's translation is drawn under the words it glosses
• Look up again for a wrong match, Save to file to write the sheet into the track's tag, and Edit lyrics… to fix a stamp by hand

• Off until you switch it on, under Settings ▸ Radio. Nothing contacts the directory until you do
• A worldwide directory (radio-browser.info, no account and no API key) narrowed by country, language, genre, codec, or bitrate
• Your own stream URLs, checked at the dialog rather than at the first click; .pls, .m3u, and .asx resolve to the audio behind them
• Live titles reach the player bar, Now Playing, your desktop's media controls, and Discord, with buffering and reconnection handled underneath
• Favorites, Recently Played, station pages, cached logos, and playlist-file import and export
• Station cards select like any other card, with a right-click menu to play, favorite, copy, edit, or remove them
• The titles a station announced this session, listed under it and copyable
• Segmented (HLS) stations play like any other

• Six families, each with light and dark variants and configurable accents: Catppuccin (Latte, Frappé, Macchiato, Mocha), Material 3, GNOME Adwaita, KDE Breeze, Windows Fluent, and macOS
• System dark/light is followed automatically and KDE color schemes are read from kdeglobals. A first launch starts on the desktop's own theme (Windows Fluent, KDE Breeze, or GNOME Adwaita, in the system's light or dark mode), and on Catppuccin Mocha anywhere else
• Material You derives a palette from the current artwork in seven color styles, fading the window into each new cover's colors, and the accent and window outline can follow the artwork too. Material 3 and Material You follow Google's 2025 Material color spec wherever it defines the style
• Headers paint an aurora of the album's own colors by default, or a blurred cover instead
• A custom transparent titlebar, or native decorations, which KDE opens on

MP3, FLAC, M4A/M4B (AAC and ALAC), raw AAC (.aac), Ogg Vorbis (.ogg, .oga), Opus (.opus, up to 7.1), WAV (PCM and ADPCM), AIFF/AIFF-C, Matroska (.mka), and CAF. Matroska and CAF carry no tags Melodia can read, so those tracks list under their filename, as does anything whose tags are too damaged to parse.

Sixteen locales (English, German, French, Spanish, Turkish, Greek, Italian, Brazilian Portuguese, European Portuguese, Dutch, Polish, Indonesian, Russian, Ukrainian, Hungarian, Vietnamese), switchable at runtime with no restart under Settings ▸ Language.

• Scrobbling to Last.fm and ListenBrainz, each independently, with loved-track sync that catches your existing favorites up on connect, and a durable offline queue. A ListenBrainz love needs the track's MusicBrainz recording ID in its tags
• Discord Rich Presence (off by default): title, artist, album, and optionally a cover looked up on Deezer, sent to your running Discord client. Nothing leaves the machine while it is off
• OS media controls (MPRIS2 on Linux, SMTC on Windows) and media keys
• A system-tray icon with playback controls, on by default, and an optional close-to-tray; both under Settings ▸ Window
• A daily update check, named on first run and switchable under Settings ▸ Updates. It asks GitHub for a signed manifest and nothing else; package-managed installs skip it entirely
• Set Melodia as your default player and double-click a track; it runs as a single instance, so files open in the window you already have
• Always-on-top on Windows, on any X11 desktop, and on Wayland under KDE, or GNOME with the Window Calls extension
• On Linux, a self-deploying desktop entry and AppStream metadata for KDE Discover and GNOME Software
• Window, queue, and navigation state persisted across sessions

Measured with scripts/measure-footprint.py on KDE Plasma under Wayland, with the default Band-Limited sample-rate conversion. Startup is sampled from the moment Melodia launches, and every other scenario starts 90 s after the launch or 70 s after the switch into it, once the database connections either opened have closed. The Songs tab is My Library ▸ Songs, opened with the playlist still playing; double-clicking its first track queues the whole library, and the queue sheet opens over that queue.

Native titlebar

With both libraries, 51,012 tracks. The higher PSS and RSS at startup are the library's database, which every connection that read it keeps mapped until it closes; RSS counts it once per connection.

Custom titlebar

With both libraries, 51,012 tracks:

Linux keeps no commit charge per process, so Anonymous is the figure to compare: the memory Melodia allocated itself, in RAM or in swap, which nothing another process does can move. PSS is KDE System Monitor's Memory column. It adds a share of the file-backed pages, mostly the binary and the shared graphics stack rather than anything Melodia allocated, and that share shrinks whenever another program maps the same libraries. USS is the part no other process maps, what closing Melodia gives back, and System Monitor's Private column. RSS counts every shared page in full. GPU memory is the video memory the NVIDIA driver charges to Melodia, where FemtoVG keeps its textures and framebuffers, counted apart from the rest. CPU (1 core) counts one full core as 100%, and CPU (all cores) spreads that across all 16 threads, as System Monitor's CPU column does. GPU is the utilization the NVIDIA driver reports for Melodia, in whole percents each second.

Measured with scripts/measure-footprint.py, on an output device Windows mixes at 192 kHz, so every track passes through the sample-rate conversion. Startup is sampled from the moment Melodia launches, and every other scenario starts 90 s after the launch or 70 s after the switch into it, once the database connections either opened have closed. The Songs tab is My Library ▸ Songs, opened with the playlist still playing; double-clicking its first track queues the whole library, and the queue sheet opens over that queue.

Native titlebar

With both libraries, 51,012 tracks. The higher commit and working set at startup are the library's database: every connection that read it holds a page cache of its own, which commit counts, and maps the file, which the working set counts, until it closes.

Both tables use the default Linear sample-rate conversion. Band-Limited costs more CPU while playing: with the 512-track library, about 1.9% of one core in the list view and about 7.9% with the visualizer live.

Custom titlebar

With both libraries, 51,012 tracks:

Both tables use the default Linear sample-rate conversion. Band-Limited costs more CPU while playing: with the 512-track library, about 2.1% of one core in the list view and about 6.2% with the visualizer live.

Windows keeps no PSS, so Commit is the figure to compare: the private memory Windows has set aside for Melodia, in RAM or paged out, which doesn't fall when Windows trims the process. Private WS is Task Manager's Memory column, the private memory in RAM at that moment. Working set adds the pages Melodia shares with other processes, mostly system DLLs. GPU dedicated and GPU shared are the video memory Task Manager lists per process, where FemtoVG keeps its textures and framebuffers, counted apart from commit. CPU (1 core) counts one full core as 100%, and CPU (all cores) spreads that across all 16 threads, as Task Manager's Details tab does. GPU is the busiest GPU engine, Task Manager's per-process figure.

A source build keeps its own library under Melodia-dev, beside an installed copy's folder rather than inside it, so a schema migration still on a branch can't leave an installed Melodia unable to open its database, and the two can run at once. MELODIA_DATA_DIR points either build at a directory of your choosing.

winit/ is a checked-in copy of winit 0.30.13 plus an unmerged Wayland file drag-and-drop fix (winit#1881) and drag events that say where the pointer is, which is how a dropped file finds the queue or the open playlist. It is wired by Cargo.toml's [patch.crates-io], so a fresh clone builds with no setup.

Last.fm scrobbling needs an API application's key and shared secret, which identify the app rather than any account. They are read at compile time from LASTFM_API_KEY / LASTFM_SHARED_SECRET, so nothing secret lives in the repo; for local builds, copy .env.example to .env and build.rs bakes them in. A build without them is fully functional, ListenBrainz included, and the Last.fm Connect button reports "not configured in this build". A fork wanting its own Discord presence sets MELODIA_DISCORD_APP_ID.