Skip to content

Roadmap

What's open, organized by what it would feel like.

Tier 1 — listening experience wins

  • Lyrics via LRCLIB — read embedded \xa9lyr (MP4) / USLT (ID3); fetch from lrclib.net for missing ones (free, no API key, supports synced LRC). Cache per-track as <track>.lrc sidecar. Server's getLyrics / getLyricsBySongId are wired but only return embedded text; remote-fetch + sidecar cache is still open. Symfonium + Amperfy display synced lyrics live.
  • Album cover art in the TUI — small thumbnail in the now-playing header. Tried once with textual-image (Unicode halfblock fallback was too low-res); a chafa subprocess approach would give crisp output but adds a brew install chafa system dep. Deferred until the cost/benefit is clearer.

Tier 2 — bigger directions (~3-5 sessions)

  • Web UI — small Vue/htmx frontend mounted at / (replacing the JSON probe response). Same Subsonic backend; lets you play from any browser without installing an app.
  • Podcast support — Subsonic spec already defines getPodcasts / getPodcastEpisode / createPodcastChannel. Add an RSS feed list, fetch episodes on schedule, store position. Symfonium has decent podcast UX out of the box.
  • iTunes / Apple Music import — read the local Apple Music database to import play counts, ratings, and playlists. One-shot migration tool.

Tier 3 — server hardening / scale

  • Recorded-session integration tests — VCR-style HTTP fixtures from real Symfonium / Amperfy / Feishin sessions; replay against TestClient. Catches "client X probes new endpoint" regressions.
  • systemd / launchd plist — for users wanting serve to autostart on boot.
  • MQTT / webhook scrobble forwarding — push play events to Home Assistant / Last.fm / external systems.
  • Bigger getCoverArt cache — currently it's recomputed per request; an LRU keyed on (album_id, size) would help on slow disks.
  • TUI /-bar filter on top of FTS5 — /search3 is FTS5-backed (see Done); the TUI's incremental / filter is still in-memory casefold substring. Wiring it through the same in-memory FTS5 index would give ranked / prefix-matching results in the TUI.

Tier 4 — convert-pipeline polish

  • Better folder-fallback for live-venue parens — currently (Live in Madrid) is preserved (correctly), but Some Album (Live, 2003-04-15, Madrid) could be cleaned up while still preserving the live-ness signal.
  • AcoustID auto-enable — currently you have to pass --acoustid-key per run. Read it from ~/.config/musickit/serve.toml once and apply automatically when an album has tagless tracks.
  • Album merge tool — when the same album exists with different tags as two folders, an interactive merge.
  • --dry-run with rich diff — show exactly what tags would change, what files would move.
  • In-TUI rename when album/artist tags change — the tag editor patches tags in place but doesn't move the on-disk folder. retag --rename does this from the CLI; the TUI editor could too.

Tier 5 — speculative

Things that would be interesting if anyone ever asked for them, but not pursued speculatively:

  • BPM / key analysis (needs librosa, big dep weight)
  • AI-generated playlists with audio-feature similarity (current musickit playlist gen is tag-based; an audio-feature pass would need fingerprinting / librosa)
  • Multi-user serve (right now: single-user)
  • Sonos / Chromecast / DLNA output (AirPlay covers the Apple-ecosystem case)
  • Cross-fade between tracks
  • Listening rooms / sync-play across clients
  • Voice control

Done

  • Convert pipeline — covered in Convert
  • Library audit + fix — covered in Library
  • TUI: local + radio modes
  • TUI: Subsonic-client mode (lazy-loaded)
  • TUI: AirPlay output (CLI + in-TUI picker, persists across launches; pause + volume routed to the device)
  • TUI: in-place tag editor — e opens a track or album-wide editor that writes via mutagen + patches the in-memory model
  • TUI: / incremental filter on the focused pane
  • TUI: single-click cursor / double-click play (Spotify-style); cursor snaps back to playing track on focus loss
  • TUI: ReplayGain normalisation on local playback
  • TUI: bordered-panel UI polish (boxed sections, now-playing card, warm accent for active playback)
  • TUI: dynamic title-column width that re-flows on terminal resize (debounced)
  • TUI: audio engine in a separate process — PyAV decoder + sounddevice callback live in their own interpreter, communicate via multiprocessing.Queue + shared memory. UI reflows / focus changes / GC pauses can no longer stall the audio callback into a buffer underrun. Replaces the prior GIL-contention mitigations (500ms / 1s PortAudio buffer, resize-debounce, focus-change short-circuit), now that the engine has its own GIL.
  • Subsonic-compatible serve (~30 endpoints)
  • XML response format (Subsonic spec default) + POST + form-body credentials
  • ffmpeg-on-the-fly transcoding (format=mp3 / maxBitRate)
  • OpenSubsonic extensions advertised: formPost, transcodeOffset, multipleGenres, songLyrics
  • Multi-genre indexing — byGenre and getGenres honour track.genres[], not just the legacy single-genre field
  • Per-track recording MBIDs (one MB release/<mbid>?inc=recordings lookup at enrich time)
  • mDNS / Bonjour autodiscovery (server advertises, TUI auto-detects)
  • Filesystem watcher with debounced auto-rescan
  • Persistent SQLite library index — <root>/.musickit/index.db removes the cold-start filesystem walk + tag read; library.load_or_scan hydrates from rows then runs a delta-validate pass to pick up filesystem changes (added / removed / tag-edited albums); musickit library index status|drop|rebuild for management
  • Genre indexing (model + scan + getGenres + byGenre)
  • cover-pick semi-automated workflow
  • Folder-name edition-annotation strip
  • Continuous-numbering across-discs detection
  • Real-world tested against Amperfy / Symfonium / play:Sub / Feishin
  • Push to GitHub + CI (make check && make test on push/PR)
  • This documentation site (mkdocs + Material)
  • musickit playlist gen|list|show — tag-based auto-generated .m3u8 mixes anchored to a seed track (similarity scorer over artist / genre / year / compilation flag, per-album + per-artist caps; output is standard extended M3U)
  • TUI g keybind — generate-and-play a 60-min mix from the highlighted or currently-playing track
  • TUI Mixes browser entry — saved .m3u8 files appear in the right pane; selecting one resolves paths against the live index and plays it as a virtual album with stale-path graceful degradation
  • Persistent stars / favourites — Subsonic clients' heart buttons are now real. /star, /unstar, /getStarred, /getStarred2 backed by <root>/.musickit/stars.toml (TOML, hand-editable, survives index rebuilds).
  • FTS5 search backend — /search3 and /search2 use an in-memory SQLite FTS5 index built fresh on every cache reindex. Sub-ms ranked results with prefix matching (bey matches Beyoncé), diacritic folding (unicode61 remove_diacritics 2), bm25 ordering, and multi-token AND across title + album_artist + year body text. Falls back to casefolded substring scan when SQLite was compiled without FTS5.