The native desktop client for Tidal. Bit-perfect gapless playback, hi-res downloads, a real parametric EQ, and your listening history from Last.fm and Spotify, in one app with no web bloat. It packages a FastAPI backend, a React and Tailwind frontend, and a PyAV and sounddevice audio engine behind a single pywebview window.
This app is not made by Tidal. It talks to the same private endpoints the official Tidal apps use, but with traffic patterns that can look unusual to Tidal's anti-abuse system. Heavy use, especially mass downloads or rapid browsing, has triggered both soft rate limits and longer "abuse detected" cooldowns.
A soft rate limit pauses playback and search for about a minute. The abuse-detected variant pauses everything for thirty minutes and counts as a strike. Repeated strikes can escalate to account suspension or a permanent ban.
The app throttles itself in normal use and surfaces a banner the moment a backoff engages, but you are using your own account at your own risk. If you cannot accept any chance of a Tidal action against your account, do not use this app.
Download for macOS, Windows, or Linux »
- What's inside
- How For You works
- Support
- Install a released build
- Stack
- Run it from source
- Layout
- Notes
- Known limits
- Building a distributable
- Contributing
- License
Playback. Audio plays through a pipeline built on PyAV for decode and sounddevice for output. Tracks hit the OS audio API at their native sample rate and bit depth. There is no resampling and no hidden software mixer. Transitions between tracks at the same sample rate splice at the PCM sample boundary, so they are truly gapless. Transitions that change sample rate reopen the output stream and bridge across roughly fifty milliseconds, which is the same limitation Tidal desktop and foobar2000 have.
A parametric equalizer runs in three modes. Off bypasses the
EQ stage entirely, leaving the audio bit-perfect. Manual is a
full parametric EQ: every band carries its own filter type (peaking,
low shelf, or high shelf), frequency, gain, and Q, and you add or
remove bands freely. You shape them on a live frequency-response
graph — drag a node to set frequency and gain, scroll over it to
tighten or widen its Q, double-click empty space to drop a new band
or a node to delete it — or type exact values in the per-band rows.
The curve redraws in real time as you drag, computed with the same
RBJ biquad math the audio engine runs. A fresh install seeds six
flat bands so there's something to grab, and flat bands compile to
nothing, so the audio stays bit-perfect until you actually shape
one. Presets render as miniature response curves, a master preamp
keeps headroom for boosts, and an A/B bypass flips the whole EQ off
and on without losing your bands. Profile loads a per-headphone
AutoEQ correction file
so your headphones target neutral, Harman, or B&K. Profiles are
user-imported (ParametricEQ.txt from AutoEQ or .txt from
Equalizer APO), with optional bass, treble, and preamp tilt
sliders stacked on top, an A/B bypass to compare on the fly, a
frequency-response graph that visualises the cumulative shape,
and per-device profile mapping so different headphones plugged
into different DACs each apply their own correction
automatically. See docs/eq-and-autoeq.md
for the full guide.
Three more optional stages sit in Settings under Playback, all off by default so the path stays bit-perfect until you reach for them. Crossfeed bleeds a little of each channel into the opposite ear through a Bauer-style 700 Hz split, which pulls hard-panned mixes out of the middle of your skull and closer to a speaker soundstage. ReplayGain loudness leveling smooths the volume jumps between tracks using the EBU R128 numbers Tidal already ships and that downloaded files carry in their tags, with per-track and per-album modes. A Signal Path readout shows exactly what is happening to the audio right now, from source bit depth and sample rate through each active stage to the output device, so you can confirm a stream really is bit-perfect.
The output device picker lists every USB DAC,
Bluetooth sink, and builtin option the OS exposes, plus Chromecast
targets, Tidal Connect renderers, and UPnP/DLNA devices discovered
on the LAN. You can switch between any of them mid playback. The
DLNA path covers WiiM, newer Bluesound, Cambridge, NAD, most
network AVRs, and a long tail of cheaper Hi-Fi network bridges.
See docs/dlna-renderer.md for the
DLNA-side details. Global media keys for play, pause, next, and previous
work even when the window is minimized. There is an opt-in desktop
notification on every track change. On Linux, Tideway registers as
an MPRIS player on the session bus, so GNOME and KDE media widgets,
the lock screen, and playerctl can see the current track and
drive playback. On macOS the same job is done through the native
Now Playing integration in Control Center.
Browsing and library. Search, a unified Charts page (Popular, Top, Rising, and New Releases as tabs), and dedicated album, artist, playlist, and mix pages are all there. The Home page also surfaces two AlbumOfTheYear-backed discovery rows, a top-of-year highlight reel and a new-releases row, so you can find music that isn't already in heavy rotation. Both open into full pages where you can filter by genre, so pulling up the year's best ambient or this week's new metal is one dropdown away, with the data credited back to AlbumOfTheYear. Album, artist, playlist, and mix pages have Play, Shuffle, and a More menu. Artist pages keep compilations in their own section instead of mixing them into the studio albums, and lead with a Latest releases row. Albums also show a quality badge that reflects the best version Tidal has in its catalog. That includes Max, Lossless, Dolby Atmos, and 360 Reality Audio, though see the limits section below for what the app can actually stream. Playlists can be created, edited, reordered, and extended. Any track in the app has an "Add to playlist" action attached to it. There are dedicated radio pages for artists and tracks. The Stats page is backed by Last.fm and surfaces your top tracks, artists, and albums, along with a listening activity chart that responds to period filters.
AI-generated music. Tidal tags tracks it identifies as 100% AI-generated, part of the AI policy it started rolling out in July 2026. Tideway reads that tag and, with the filter on, hides those tracks everywhere you browse: search, artist and album pages, mixes, radio, and the Home recommendations. The downloader skips them too, so expanding an album or playlist never quietly pulls in AI tracks. Your own library is the one exception. Anything you've already favorited still shows, which mirrors how Tidal's own client treats the setting. The filter is on by default and lives in Settings under Playback. It's a local preference on this install, not something stored on your Tidal account, so it doesn't sync across machines or with the official app. Existing installs get a one-time notice on the Home screen the first time they open after updating, with the option to allow AI content right there if you'd rather keep it.
Data enrichment. Last.fm powers scrobbling and per-user or global playcounts on every track, album, and artist page. Spotify contributes a separate layer of public data through its anonymous GraphQL endpoint. This includes track playcounts, artist monthly listeners, and top listening cities. The two services complement each other. Last.fm gives you your own history, and Spotify fills in the global popularity signal.
Downloads. The file format follows the quality tier. High and
Max are lossless and download as FLAC. Tidal delivers that FLAC
inside an MP4, so Tideway remuxes it into a native .flac file.
Low and Medium are lossy AAC and download as .m4a, which is the
standard AAC container, not a deprecated format. Music videos
download as well. Everything runs through a concurrent queue that
you can tune. Metadata and artwork are embedded with mutagen. Any track you have downloaded
plays straight from disk without touching the Tidal streaming
path. A filename template, toggles for per-album and per-playlist
folders, and a skip-existing option cover the common download
preferences. A downloaded playlist lands in a folder named after
the playlist and is numbered in playlist order rather than by
album track number, and the template understands {playlist_num}
and {playlist} alongside the usual title, track, album, and
artist tokens. A template like {playlist_num} - {artist} - {title}
keeps playlist downloads in playlist order on disk; on album and
single downloads the empty token and the separator next to it are
dropped, so the same template renders a clean Artist - Title.
On this device. Everything you've downloaded is browsable from a dedicated page that reads tags directly off disk. Music groups by album (or by artist, or sorted by date added), and videos sit on their own tab. Tracks play locally without round-tripping through Tidal's streaming endpoints, so they keep working offline and don't count toward any per-day stream cap.
Import. You can transfer playlists from Spotify using PKCE OAuth, which requires a Spotify Developer client id that you supply yourself. There is also support for Deezer and for plain M3U or text lists. Liked songs, saved albums, and followed artists all mirror across.
Video. Music videos play through native HLS, powered by hls.js with a CORS proxy that keeps things working inside the Chromium based WebView the packaged app uses. Picture in picture, fullscreen, a quality picker, and subtitle toggles are all supported.
Tidal attribution. Every track you play fires a playback session event at Tidal's event producer bus, in the exact wire format the official tidal-sdk-web uses. That means plays credit the artist and surface in your Tidal Recently Played. There is a diagnostic panel under Settings that shows the outgoing events and the status Tidal returned for each one.
Updates. On launch the app checks GitHub Releases. When a newer version is available a banner surfaces across the top of the UI, and clicking Install downloads the right asset for your OS and runs it.
For You is built from your own listening rather than from Tidal's recommendations. Three sources feed it: Last.fm supplies what you play, the community tags on those artists, and its similar-artist graph. AlbumOfTheYear.org supplies ratings and release listings. Tidal supplies the catalogue, your saved albums, and playback. No AI or language model is involved anywhere in the path.
Most of the page needs Last.fm connected under Settings. Without it only the rows that lean on your Tidal library can render.
Which artists count. Your Last.fm top artists are read over two windows and blended: the last month at double weight, the last six months at single weight. Each window is normalised against its own top artist first, because a six-month play count dwarfs a one-month one and the longer window would otherwise always win. The recent window is what lets the page follow what you are into now, and the longer one stops a single evening on one artist from redefining everything.
Which genres count. Each of those artists carries community tags on Last.fm, and the twelve strongest are read per artist. A tag's weight is divided by how widely it is applied across all of Last.fm, so broad labels lose to specific ones. Without that step the ranking just recovers whichever vague tag every artist carries, since breadth and tag strength rise together. It is the difference between being told you listen to "electronic" and being told you listen to IDM.
A tag carried by only one of your artists has to be that artist's defining genre, scoring at least 80 out of 100, or it is ignored. One artist's passing association is not a genre you have.
Surviving tags are matched against AOTY's genre taxonomy, which knows around 360 genres. Anything AOTY does not recognise drops out here, which also discards Last.fm descriptors that are not genres at all.
Your orbit. Your top eight artists are expanded through Last.fm's similar-artist graph into roughly four hundred artists, each carrying a score for how strongly it connects back to what you play. This is what separates a recommendation from a chart: without it, everyone who listens to shoegaze sees the same shoegaze albums.
Recommended For You. Takes the strongest remaining pick from every other row in turn. It deliberately does not score across sources, since an AOTY rating and a Last.fm match score measure different things and forcing them onto one scale lets a weak pick outrank a strong one. Round-robin needs no shared scale and makes it impossible for one source to dominate. Every card keeps the reason it came from.
New Releases For You. AOTY's recent releases for each of your top genres, interleaved. Scoped to your genres rather than re-ranking a global chart of everything released that week. This row moves as music is actually released rather than on a schedule.
From Your Genres. AOTY's highest user-rated albums of all time in each of your top six genres, not restricted to any year. Within a genre the order is not AOTY's ranking. Candidates are scored on three things: whether the artist sits in your orbit, the album's rating once thin vote counts are discounted, and a penalty for artists already in your heavy rotation. A row of albums by people you already play is a mirror rather than a recommendation. View more opens a page giving each of your top twelve genres its own row, which you can keep loading further into.
Popular in Your Orbit. AOTY's highest user-rated albums of the current year, re-ordered by how well each fits your genres. The balance between raw rating and genre fit shifts with how mainstream your listening is, measured as how much your top artists overlap Last.fm's global charts. A niche listener gets genre fit weighted more heavily, or the row would collapse into the global chart.
Fans Also Like. Artists that listeners of your top artists also play, from Last.fm's similar-artist graph, resolved to their albums. Because that graph is built from listening rather than from who has played in a band together, an artist nobody listens to never enters the pool. Picks are taken from each seed artist in turn so one seed sitting in a dense corner of the graph cannot fill the shelf.
Albums already in your Tidal favorites are never recommended back to you, in any row. At most two albums by the same artist appear in a row, and reissues of the same record collapse together.
Rows rotate weekly. Charts move slowly, and an all-time ranking does not move at all, so each row walks further into its candidates on a weekly schedule. Rotation is bucketed by week rather than randomised, so the page is stable across a refresh and within a session, and changes on a schedule instead of under your cursor. For the genre rows the rotation stays inside the best-ranked candidates, so the variety does not come at the cost of fit.
Ratings are discounted by how many people voted. A 95 from eleven listeners is weaker evidence than an 88 from three thousand, so scores are pulled toward the middle until the vote count earns them.
The assembled page is cached for 15 minutes. Underneath, a genre's all-time ranking is cached for 24 hours, recent releases for 30 minutes, the current year's chart for an hour, and the mapping from an AOTY album to a Tidal one for 30 days. A first ever load is therefore slower than later ones, and does a lot of catalogue lookups.
AOTY is scraped rather than read through an API, since it does not offer one. Requests can be rate limited, in which case rows come back shorter than usual or drop out entirely until the limit clears.
If Tideway is useful to you and you'd like to support its development, you can buy me a coffee on Ko-fi. Tideway is free and will stay that way; donations are appreciated but never expected.
Grab the latest from the Releases page for whichever fork of this repo you are installing from.
On macOS, download Tideway-<version>.dmg, double-click it, and
drag the Tideway app into /Applications. Apple Silicon only:
the binary is arm64-native and won't launch on Intel Macs. This
has been the case since the earliest signed release; the README
just hadn't called it out before.
On Windows, download Tideway-setup-<version>.exe and run it. The
installer drops the app under your user profile, registers a Start
Menu entry, and offers an optional desktop shortcut.
On Linux, the recommended install is the Flatpak. It bundles GTK
and WebKit2GTK from the GNOME 49 runtime so the app gets a real
native window without any host-package fiddling, and flatpak update keeps it current.
# Subscribe to the auto-updating repo (one-time)
flatpak remote-add --user --if-not-exists tideway \
https://j-m-punk.github.io/tideway/tideway.flatpakrepo
# Install
flatpak install --user tideway com.tidaldownloader.Tideway
# Run
flatpak run com.tidaldownloader.Tideway
# Update later
flatpak update --user com.tidaldownloader.TidewayIf you'd rather not subscribe to a remote, download
Tideway-<version>.flatpak from the latest release and install the
single-file bundle:
flatpak install --user --bundle Tideway-<version>.flatpakThe bundle won't auto-update — grab a newer one from the release page when you want to upgrade.
Global media keys require an X11 session (Wayland will degrade — the player still works, the global hotkeys do not). ARM Linux (Raspberry Pi etc.) is not built today.
If you were running the v1.10.x or earlier AppImage, install the Flatpak with the commands above to keep getting updates. The AppImage build was retired in v1.11.0 because it had no native window and fell back to opening the app in the system browser.
The builds are not code signed. Signing costs 99 dollars a year from Apple and upwards of 200 dollars a year from Microsoft, which is not a cost worth paying for an open source hobby project. Both operating systems show one scary looking warning the first time you open an unsigned app. After that they remember the choice and the warning never comes back.
On macOS, right click (or Control click) the .app, pick Open,
and confirm Open in the dialog that appears.
On Windows, click More info in the SmartScreen dialog, then Run anyway.
The Linux Flatpak doesn't carry a Gatekeeper-style first-launch warning; the sandbox is the trust boundary.
If you would rather verify the build yourself, clone the repo and
follow Run it from source below. PyInstaller produces the same
macOS / Windows bundle you download from Releases; the Linux
Flatpak is built with flatpak-builder against the manifest at
flatpak/com.tidaldownloader.Tideway.yaml.
The backend runs on FastAPI with tidalapi and mutagen. All mutable state, which includes your settings, your Tidal session, the download queue, and the Spotify and Last.fm caches, lives in a per-user app data directory.
The audio engine is PyAV for demuxing and decoding (libav is bundled inside the Python wheel), sounddevice for output, and scipy for the equalizer biquad cascade.
The frontend is Vite, React, TypeScript, Tailwind CSS, and a set of shadcn style primitives, with React Router handling navigation and React.lazy splitting each route into its own chunk. Realtime updates flow over Server Sent Events, both for player state and for download progress.
The desktop shell is pywebview, which uses WKWebView on macOS and WebView2 on Windows. Global media keys are delivered by pynput.
First time setup:
python3 -m venv .venv
.venv/bin/pip install -r requirements.txt
(cd web && npm install)For day-to-day frontend work, start FastAPI and Vite together:
./run.shThe FastAPI server listens on http://127.0.0.1:8000 and serves
the JSON API and /docs. The Vite dev server is on
http://127.0.0.1:5173. Hot module reload picks up frontend
changes instantly.
Anything that depends on the native pywebview window (macOS chrome, window drag, traffic lights, global media keys) only exercises in the desktop shell. To launch it from source:
(cd web && npm run build)
.venv/bin/python desktop.pydesktop.py serves whatever is in web/dist/, so rebuild the
frontend any time you change React code or the desktop window will
show stale UI. If web/dist/ doesn't exist yet, desktop.py
refuses to start and tells you to build it, since the window would
otherwise open blank.
On Linux the native window needs a system webview backend that pip can't install portably. Install your distro's GTK bindings (preferred, it's what the Flatpak uses):
# Debian/Ubuntu
sudo apt install python3-gi gir1.2-webkit2-4.1
# Fedora
sudo dnf install python3-gobject webkit2gtk4.1
# Arch/Manjaro
sudo pacman -S python-gobject webkit2gtk-4.1or the Qt fallback via pip: pip install qtpy PySide6. Without
either, desktop.py falls back to opening Tideway in your default
browser and prints these same instructions.
On first launch, click Login with Tidal. The login uses PKCE.
Paste the redirect URL back into the app and Tideway exchanges it
for a session that's entitled for hi-res streaming. The session
and every other piece of app data live in the per-user data
directory. On macOS that's
~/Library/Application Support/Tideway. On Windows it's
%APPDATA%\Tideway. On Linux it's ~/.local/share/Tideway.
app/ shared Python logic: tidal client, downloader,
metadata, play reporting, Spotify and Last.fm
clients
app/audio/ audio engine: decoder, segment reader, player,
equalizer, gapless splicing, Cast and DLNA
output
server.py FastAPI entry point
desktop.py pywebview shell, which is the entry point for
the packaged app
web/ Vite and React frontend
run.sh one command dev launcher
Tideway-mac.spec PyInstaller spec for macOS
Tideway-win.spec PyInstaller spec for Windows
flatpak/ Flatpak manifest + generated deps for Linux
scripts/ build helpers: icons, DMG, Inno Setup .iss
tests/ pytest suite
.github/workflows/ CI (test.yml) and release pipeline (release.yml)
No external binaries are required. Audio and video are both handled by PyAV, which ships libav inside its Python wheel, so there is never a reason to install ffmpeg or VLC on the side.
Your settings live in settings.json in the per-user data
directory. You can edit everything from the Settings page inside
the app. Your last volume and the window's size and position are
remembered there too, so a relaunch comes back the way you left
it.
Spotify enrichment works anonymously out of the box and needs no login. The Spotify importer is different. It needs a Spotify Developer client id, which you paste into Settings > Import. Last.fm enrichment uses a bundled API key; connecting your own Last.fm account is optional, and you only need to do it if you want scrobbling and your own per-user playcounts.
Dolby Atmos, 360 Reality Audio, and MQA do not play or download in this app. Tidal only serves those immersive streams to a short list of certified partner devices. The PKCE client id we ship with is the Android Automotive one that tidalapi exposes, and it is not on that list. Every streaming request we make comes back as a stereo FLAC downmix, even on albums that advertise Atmos or 360 in their catalog metadata. The quality badge on album pages still shows those tags, because knowing that the immersive version exists on Tidal itself is useful information to have.
This is a choice, not a hard wall. Other projects such as Dniel97/RedSea do support Atmos downloads by signing in with client ids and secrets extracted from specific Android TV or mobile apps that do have the immersive entitlement. We chose not to go down that road. It means impersonating a certified partner device, which is a different category of Tidal ToS violation than simply using the public PKCE flow. The client ids involved get revoked periodically and have to be re-extracted, which turns a shipped feature into a running maintenance burden. And Atmos playback on a desktop needs either a Dolby-capable renderer on the OS side or a receiver on the other end of HDMI, which most setups do not have. If any of that changes, we can revisit.
Very old x86 CPUs can crash the audio math libraries. The
prebuilt numpy and scipy wheels on PyPI now target the x86-64-v2
baseline, which needs SSE4.2. CPUs from before roughly 2009, such
as the AMD Athlon II or first-generation Core, lack it and die
with an illegal-instruction fault on import. That kills both the
Flatpak and a stock source install. If you run source on hardware
like that, install your distro's own numpy and scipy packages,
which are compiled for the local baseline, and create the venv
with --system-site-packages so they're picked up.
Network audio output is mostly there, with caveats. Chromecast, Tidal Connect, and UPnP/DLNA all ship in the output device picker.
The DLNA path in particular has not been bench-tested against real renderers. The unit tests cover the SOAP wrappers, session lifecycle, and audio plumbing, but device-side behaviour (does your specific WiiM accept the FLAC stream, does pause respond fast, does the device cleanly disconnect when you switch) is unverified. If you hit a renderer that misbehaves, please file a GitHub issue with the device's manufacturer and model and any log output from the diagnostic panel.
Play reporting to Tidal Recently Played is best effort. Every
play fires a playback_session event at Tidal's event producer
bus in the exact wire format the official SDK uses, and the bus
accepts the event. Whether it surfaces in Recently Played on the
official Tidal app depends on whether the play_log consumer
accepts events from our client id. Sometimes it does and
sometimes it does not. The Settings page has a diagnostic panel
that shows every event we send along with the status Tidal
returned for it.
The builds are not code signed. First launch will trigger a Gatekeeper warning on macOS and a SmartScreen warning on Windows. See the install section above for the one time steps to get past them.
Drop a 1024 by 1024 PNG at assets/icon-source.png, then run:
scripts/build_icons.sh
That produces assets/icon.icns for macOS and assets/icon.ico
for Windows. Both PyInstaller specs look for those files. If they
are missing the build ships a generic placeholder icon.
npm --prefix web run build
.venv/bin/pyinstaller Tideway-mac.spec --noconfirm
scripts/build_dmg.sh
The DMG lands at dist/Tideway-<version>.dmg. Users drag the
.app into Applications and launch. The CI builds on macOS 14
(Apple Silicon), so the DMG produced by the release workflow is
arm64. Intel Mac builds aren't part of the release pipeline today.
npm --prefix web run build
.venv\Scripts\pyinstaller Tideway-win.spec --noconfirm
"C:\Program Files (x86)\Inno Setup 6\ISCC.exe" scripts\Tideway.iss
The installer lands at dist/Tideway-setup-<version>.exe. Users
run it and walk through Next, Next, Install. The release pipeline
builds both an x64 and an arm64 installer (Windows 11 ARM64
runner), and the auto-updater picks the right one for the user's
machine.
Linux ships as a Flatpak built against the GNOME 49 runtime (Python 3.13, GTK 3, WebKit2GTK 4.1). The PyInstaller-frozen AppImage path was retired in v1.11.0 — it couldn't import the host's PyGObject / WebKit2GTK and pywebview fell back to opening the app in the system browser.
# One-time: install flatpak + flatpak-builder + GNOME 49 runtime
sudo apt-get install flatpak flatpak-builder ostree elfutils
flatpak remote-add --user --if-not-exists \
flathub https://flathub.org/repo/flathub.flatpakrepo
flatpak install --user -y flathub \
org.gnome.Platform//49 org.gnome.Sdk//49 \
org.freedesktop.Sdk.Extension.node20//25.08
# Build
flatpak-builder --user --ccache --force-clean \
--repo=repo .build-flatpak \
flatpak/com.tidaldownloader.Tideway.yaml
# Install locally
flatpak-builder --user --install --force-clean \
.build-flatpak flatpak/com.tidaldownloader.Tideway.yaml
flatpak run com.tidaldownloader.Tideway
See docs/linux-flatpak.md for the full
architecture, the --prefer-wheels / flatpak-node-generator
trade-offs, and the GPG-signing setup.
The app hits the GitHub Releases API at /api/update-check once
per launch and caches the result for an hour. When a newer tag is
out a banner surfaces across the top of the UI. Clicking Install
now downloads the correct asset for the user's OS and architecture
from the latest release, opens it, and quits the app so the installer
can replace the bundle.
The release asset names must match these patterns:
Tideway-<version>.dmg (macOS, Apple Silicon)
Tideway-setup-<version>.exe (Windows x64)
Tideway-setup-<version>-arm64.exe (Windows ARM64)
Tideway-<version>.flatpak (Linux Flatpak bundle)
The auto-updater also detects when it's running inside a Flatpak
sandbox and points the user at flatpak update instead of trying
to install a bundle the running app can't replace.
Cutting a release is a tag push, not a manual gh release create.
See CONTRIBUTING.md for the full release flow:
deploy branch, preflight, tag, draft release review, publish.
PRs welcome. Read CONTRIBUTING.md before
opening one. The short version: branch off main with a
fix/, feature/, or chore/ prefix, run the four pre-PR
checks (pytest, tsc, lint, vitest) locally, open a PR
against main. Bug reports and feature requests go in
GitHub Issues.
MIT.