Every release, newest first
Colada updates often, and each release is written to say what actually changed for you rather than what changed in the code. The app shows you this list when it updates — this is the same history, out here, so you can link to a version or see what you have been missing.
Colada stops reporting things it did not do
Three separate places where Colada told you something had happened when it had not.
Deleting a song now tells you if it could not go to the Recycle Bin. When you delete a song or a folder, Colada asks first and says the files go to your Recycle Bin so you can put them back. On some locations — a network drive, a NAS folder, some USB sticks — Windows has no Recycle Bin to move them to, and Colada deleted them outright while still telling you they were recoverable. It now says so in the confirmation you get afterwards, so a file that could not be recovered is never reported as one that can. Nothing about where your music lives has changed, and songs on your own drive still go to the Recycle Bin exactly as before.
A library sync that was refused no longer counts as sent. If the server was busy, or was restarting for an update, it would turn away a batch of tracks — and Colada advanced its progress counter anyway and recorded the sync as having succeeded. The tracks were never lost; the next sync sent them again. But while you were watching it, the progress bar and the “last synced” time were both describing work that had not happened. Only an accepted batch counts now.
Your profile's privacy setting is no longer guessed when Colada is offline. The Account page shows whether your profile is public, friends-only or private. If Colada could not reach your account — a slow connection, or the service restarting — it showed “public profile” regardless of your real setting. It now says the setting is unavailable rather than naming the most open one. The button that opens your public page is also hidden until Colada knows your profile link, instead of opening the Colada home page.
Colada is also stricter about what it will fetch and where it can be embedded. The Discover screen accepts a link and imports from it; it now checks that the link points at a music source it actually supports before fetching anything. Colada also refuses to be displayed inside another web page. Both close ways a malicious page could have used your running copy of Colada to reach things on your behalf.
And playlist descriptions from Spotify are now shown as plain text. The description of a playlist you browse in Discover was being rendered as though it were part of the page rather than as somebody else's words. It is text now, and reads the same as before.
"Check for updates" tells you when it couldn't check
When Colada can't reach the update server, it now says so. Before, a check that failed looked exactly like a check that succeeded: the button turned green and told you that you were on the latest version. If your connection dropped, or the server was briefly unavailable, Colada would confidently assure you there was nothing new — and there might have been.
Press it now and you get one of three honest answers: a new version is available, you're on the latest, or Colada couldn't reach the update server and you should check your connection.
The automatic check in the background is deliberately quieter. It runs every half hour, and a single missed connection there won't put an error in front of you — it simply keeps whatever it last knew and tries again. Only the button you pressed yourself reports a failure, because only that one was a question you asked.
Sort the DJ track list by any column
The track list in the DJ booth now has headings you can sort by. Press Track, Artist, BPM, Time or Played to put the list in that order, press the same heading again to reverse it, and press it a third time to get the folder's own order back. An arrow marks the column you sorted by and which way it is pointing, and only one column sorts at a time.
The order you choose stays with you as you move around: pick BPM once and every folder you open, and every search you run, arrives in BPM order until you turn it off. The track under the highlight stays under the highlight when the list re-orders, so the row your LOAD buttons are aimed at never changes out from under you.
Tracks with nothing in the column you sorted by — no BPM worked out yet, never played — gather at the bottom instead of counting as zero, so reversing the arrow to find your fastest tracks shows your fastest tracks. Sorting by BPM covers the whole folder rather than just the rows on screen, so the first time you sort a very large folder that way it can take a moment. The columns also line up with each other now, which they did not always do before.
Your listening totals say when they are estimates
The lifetime figures on your Collection page were counting some plays twice. The sync fault fixed in the last release inflated them as it went: total listen time, total plays, and the play counts behind your top tracks and artists all grew a second time as your own listening came back from your account.
Those stored numbers are kept exactly as they are rather than rewritten. Nothing can rebuild them honestly — plays from the DJ decks and listening from before you signed in are not recorded in a form that allows it — so a rebuild would only trade a figure that is too high for one that is too low, and you would have no way to tell which. Instead the page now says so, in one line under the totals.
That line appears only if your own history was actually affected; a fresh install never shows it. Everything you play from now on is counted correctly, and your session history and per-track list were never wrong — only the running totals over them.
Listening sessions stop resetting
**A listening session was being thrown away every ten minutes, so a long evening of music never showed up as one.** If you are signed in with sync on, Colada fetches your listening history from your account every ten minutes so your computers agree with each other. It was not recognising the plays it had itself just sent up — so it filed them as though they had come from a different computer. That both counted them a second time and quietly closed the session you were in the middle of. A new one opened, ran for ten minutes, and was replaced in its turn.
Colada only lists sessions of half an hour or more, so none of them ever qualified. Two hours of listening could show as ten minutes, and the session history looked empty even while your play counts kept climbing.
Sessions now stay open for as long as you keep playing, and plays made on this computer are recognised when they come back from your account.
One thing to know about the totals already recorded. Total listening time and play counts gathered while this was happening are too high, because the same plays were added more than once. Those stored figures are left as they are for now rather than rewritten — what you played and when is intact; it is the totals over it that are overstated.
When a download fails, it tells you why
Failed downloads now explain themselves. Instead of *“download engine crashed”*, the download list shows what actually went wrong — the source was rate-limiting you, YouTube asked for a sign-in check, no source had the track, the connection dropped — and an ⓘ next to it opens the one thing to do about it, along with the lines from the engine that led to that conclusion.
And there is a Repair button, under Settings → Downloads. Downloading relies on two tools that talk to YouTube and Spotify. Those services change every few weeks, and the copies on your machine are the ones that arrived the day you installed Colada — so downloads can start failing without you having changed anything at all. Repair reinstalls both and re-fetches the helpers they need, in about a minute. It also tells you which versions you are running, and warns you if a piece is missing.
If a download has been failing for days for no apparent reason, that button is the thing to try.
The Collection page stops going blank
The statistics panel could still fail, for a second reason, and did — on every refresh. The last release fixed one cause of a blank Collection page: a single unreadable entry in your playlist file. This fixes another that had been there since the previous update. Colada keeps its own small note inside the genre list it builds while tagging your library — a marker recording how far it has got. The Collection page read that marker as though it were a track's genres, failed, and showed nothing; because the page refreshes itself every half a minute, it failed again as soon as you looked. The genre breakdown went the same way for the same reason. Both now skip the marker, as the rest of the app already did.
Nothing to do, and nothing was lost — the numbers were always there, and the page simply could not finish drawing them.
Renaming a track keeps its beatgrid and cues
Renaming a track quietly threw away its analysis — including hot cues. Colada remembers a track's beatgrid, waveform and cue points against the file's contents, so that moving or retagging it never costs you the work. But renaming also rewrites the artist and title inside the file, which changes those contents, and the analysis was left behind under the old ones. The beatgrid and waveform would have been recomputed eventually. Hot cues are placed by hand and nothing can work them out again, so they were simply gone — silently, and usually not noticed until a deck. Renaming now carries the analysis across, and refuses rather than guesses in any case where it cannot be certain.
The Collection page could show nothing at all. A single unreadable entry in your playlist file made the whole statistics panel fail, so every tile stayed blank for as long as the app was open. The page now skips what it cannot read, keeps working, and reports how many entries it had to skip instead of pretending they were not there.
The notification bell no longer says "You're all caught up!" when it simply could not check. Being unable to reach Colada and having nothing waiting used to look identical. It now says so, and the unread dot no longer drops to zero because a check failed.
One part of the app failing at startup no longer takes the rest with it. Start-up steps are isolated from each other, and error reporting starts first, so a problem is reported rather than silently cancelling everything after it.
Searching for a band shows their albums
Searching for an artist listed almost none of their records. Looking up a band with their whole discography on disk would show one album — or none — under ALBUMS, because an album only counted as a result when the search words were inside the *album's own title*. A band's name usually is not: "The Division Bell" does not contain "Pink Floyd".
An album is now a result when the search matches the album or the artist who made it, so searching a band brings up their records, ordered by how much of each one you have.
Searching for a song still does not list the album it happens to be on — looking for "Money" gives you the song, not every record it appears on.
Correcting a song's name keeps what everyone else knows
When Colada tidies up a downloaded song — dropping "(Official Video HD)" from the title, or crediting everyone who is actually on the track — the tidied version used to look like a completely different song to the shared catalogue. It lost the lyrics, artwork and genres other people had already found for it, and started again as something nobody had ever heard of.
Colada now remembers what the song was filed under before the correction, so the tidied-up version stays attached to everything already known about it. It works in both directions: a song someone else cleaned up will match yours even if you have the older, messier name for it.
Genres are also honest about where they came from. A genre found for the exact song is now kept separate from one guessed from the artist's other work — so a better answer can replace a weaker one later, instead of both sitting there forever with no way to tell them apart.
Nothing to turn on, and nothing changes about the songs already on your machine.
Downloads back off the right thing, and say what went wrong
When a music service decides you have asked for too much too quickly, it stops answering for a while. Colada handles that by putting the source aside and trying another one — but it was treating YouTube and YouTube Music as two separate sources, when they are one service that counts your requests together. So after being turned away by one, the very next attempt went straight back to the same service and used up more of the same allowance. Sources that share a service are now set aside together, which means being throttled starts later and clears sooner, and downloads move on to a source that can actually answer.
Downloads also used to fail silently as far as you were concerned: something did not arrive, and nothing said why. Settings → Download providers now shows what actually went wrong for each source — whether it was being rate-limited, a sign-in that needs attention, a track no source carries, or something on your connection. Those need completely different things from you, and until now they all looked the same.
If Colada does not recognise a failure it now says so rather than guessing, so a source quietly breaking in a new way shows up instead of being filed under something that sounds close.
Genres and lyrics that are right, or honestly absent
Songs by more than one artist were being looked up wrongly. A credit written as "Billie Eilish/ROSALÍA" was sent off as one long name that matches nothing, so the song came back with no genre at all — and where the credit did get split, only the first name was used, which for a duet can be a completely different artist. One song in your library was tagged Minimal and Techno because it shares a name with a techno producer.
Both are fixed. Collaborations are now looked up properly, and where two artists make different music, Colada only keeps the genres they actually agree on rather than picking one of them. Bands whose name contains an "&" or a "+" — Above & Beyond, Florence + The Machine — are treated as the one band they are.
Lyrics are checked before they are used. If the best match disagrees with the song's length or names a different artist, it is no longer accepted — previously the first result was taken whatever it said.
Colada also used to remember a failed lookup as "this song has no genre", so a moment of bad connection could hide a song's genre permanently. Failures are no longer written down as answers, and the ones already saved are repaired automatically the next time you open Colada. Nothing to press.
Deezer downloads work again. If you had connected a Deezer account, Colada was asking Deezer for a sound quality your account does not include, and every Deezer download failed because of it. It now uses the best quality your account actually allows, so Deezer works as a source alongside YouTube — useful when YouTube is being slow or unhelpful.
Colada also no longer tries one download source that has stopped working everywhere, so when a download needs to fall back it goes straight to something that can actually answer instead of waiting for that one to time out first.
Where nothing trustworthy can be found, you now get nothing rather than a confident guess.
Downloaded songs get their real name
Downloading a song found by search used to save it under the video's title, so a track could land in your library called "Artist - Song (Official Video HD)" with the YouTube channel as the artist. The name is now cleaned before anything is saved: the tags, the file on disk and the download box all show the real artist and the real song title.
Songs whose titles genuinely contain words like "Official" are left alone — an official World Cup song keeps its name, in every language.
Where the video credits more artists than the search did, everyone is now credited. Songs from a "- Topic" channel get the artist's real name instead.
You can still edit both fields before downloading; they simply start out correct.
Pass a post on to your own followers
Community posts have a Repost button. It passes the post to the people who follow you — their feed shows what was written, under a line saying who wrote it and a link to them.
Reposting a repost credits whoever wrote it first, rather than the person you happened to see it from — so a post that travels does not slowly lose its author.
If you repost something only some people can see, everyone still sees that your post is a repost, but only the people who could have seen the original are shown who wrote it and what it said. Passing a post on never shows it to somebody it was not meant for.
Reposts show up on profiles too — yours in the app, and your public one on colada.fm.
One thing that does not work yet: if you repost a post that was only a shared song, playlist or badge, with nothing written alongside it, your followers see who it came from but not the thing itself. Reposting a post that has words in it works as you would expect.
Also fixed: three colours that were never actually being applied — your friend code, the highlight on search suggestions, and the highlight on message actions.
Long recordings can use stems now
Stems used to need memory in proportion to how long the track was: fine for a single, heavy for an extended mix, and impossible for anything really long. A twenty-minute recording took over 800 MB per deck, and with both decks loaded you could run a machine out of room without doing anything unusual.
The deck now keeps the part of the track around where you are playing rather than all of it, and moves that as you move. Stems cost the same whatever the length — so extended mixes, live sets and long recordings work like anything else, on both decks at once.
Nothing changes for ordinary tracks. Anything short enough to fit is kept whole exactly as before, which is almost everything you play, and it sounds identical because it *is* identical.
There is nothing to switch on. It simply does this when a track needs it.
Very long recordings — an hour or more — can still be too big to load at all, for a different reason that has nothing to do with stems, and Colada says so plainly when that happens.
Stems stay clean when you pitch a track up
With stems on and the pitch above zero, playing into a part of the track that had not been separated yet made the sound dip for a moment. The faster you were running, the longer the dip: a couple of milliseconds at the sort of pitch you use to beatmatch, longer if you were pushing a track hard. At exactly zero pitch it never happened, which is why it went unnoticed for so long.
Playing into an unprepared part now hands over to the ordinary track smoothly at any pitch you would actually play at. This is a different fault from the one fixed in 2.53.1, which was about jumping rather than playing — they looked identical and had nothing in common.
Separately: very long recordings — a two-hour set, a full live rip — could fail to load and blame the file for it, saying it could not be decoded. Nothing was wrong with the file. The player simply needs more memory than a browser will hand it: roughly a gigabyte for every twenty minutes, twice over while the track is being prepared.
Now Colada works out what a track will need before it starts, and if the answer is beyond what any browser allows it says so plainly, with the actual numbers, instead of downloading the whole thing first and then failing. If a load runs out of memory some other way, it says that too, rather than claiming the file is broken.
The same is true of stems on a very long recording: rather than trying and failing, the deck now says the track is too long for them. Nothing changes for tracks of normal length — both limits only refuse what could never have worked.
Seeking no longer punches a hole in the sound
If you jumped to a part of a track that had not been separated yet while stems were on, the deck went almost silent for a thirtieth of a second before the normal sound came back. Short, but on a loud system in the middle of a set it is exactly the kind of thing you hear and cannot explain.
It happened every time, on both decks, jumping forwards or backwards. The deck was fading out of stems that were not there yet instead of simply playing the ordinary track, which is what it already does when you play into an unprepared part.
Now the sound is continuous through the jump. Everything else about the way stems fade in and out while you play is unchanged — that part was already right, and it is what made this hard to spot.
Your graphics card can do the heavy lifting
Separating stems is the slowest thing Colada does, and on some machines the graphics card can do it about four times faster than the processor. There is a new setting for it under Settings → Playback.
It is a large one-time download, about 4 GB, and it is kept completely separate from the normal stems engine — so nothing you already have changes, and nothing gets slower if it does not work out. Colada checks that your card can genuinely do the work before using it, rather than taking its word for it: a card can look perfectly capable and still fail at the first real task.
Most cards are not supported yet — today this only works on some recent AMD cards. If yours is not one of them, stems carry on using the processor exactly as they do now, and Settings will say so plainly instead of just showing a tick. It is worth checking there after your first separation: "using your graphics card" and "installed, but using the processor" are different messages on purpose.
Nothing downloads unless you ask for it, and it never arrives with an update.
The deck you can hear comes first
If stems sometimes never caught up — the pill stuck on "catching up" while the percentage crawled — this is why, and it was not your computer being slow.
Asking for stems on a track put it at the very top of the separation queue, which is right while you are playing it. But it stayed there after you swapped the track out. Do that two or three times across a set and the track actually on the deck was queued behind two or three tracks nobody was listening to any more, getting a fraction of the machine — less than it needed to keep up with playback, so it never did.
A track that leaves the decks now steps aside for the one that is on them. It keeps everything it had already separated and stays in the queue, so putting it back costs nothing.
Two smaller faults in the same area are fixed with it: asking for stems again while the separator had gone idle could be accepted and then quietly start nothing, and seeking at the exact moment a track finished could throw away that seek without saying so.
Stems land in seconds
The last version made the separation follow you when you jump. This one makes it arrive faster once it gets there.
Separation happens in chunks, and each chunk only becomes playable once the whole of it is done. Those chunks were long enough that landing somewhere new still meant waiting about ten seconds for the first usable audio. They are now short enough that the first chunk after a jump lands in about five seconds and every one after it adds a couple more.
The trade is that there are more joins between chunks. They are crossfaded, and on everything tried so far they are inaudible — but if you ever hear a faint click or a moment of smearing at a regular interval, particularly on long sustained notes or reverb tails, that is what it would be, and it is worth telling us.
Stems follow you when you jump
Separating a track into stems happens in pieces, starting wherever the deck is playing and working forward — so you can start mixing seconds after asking for it instead of waiting for the whole track.
Jumping to a different point did not move that starting position. The separator carried on working through the part of the track you had just left, and the "catching up" percentage climbed over audio you were never going to reach. Land somewhere late in a track and it looked like it had started over.
Now a jump moves it. Drop the needle anywhere — the waveform, a hot cue, a scratch — and the separation re-aims at where you actually are and carries on forward from there. Anything already prepared is kept, so jumping back and forth costs nothing.
The storm is calmer, and waits to be asked for
The last version stopped the Lightning visualiser flashing for people whose system asks for reduced motion. This one deals with everyone else.
Lightning strikes on the beat, and each strike lit the whole screen. At a busy tempo that worked out at five or six flashes a second — and if you had turned the strike rate up past halfway, there was no limit on it at all. Flashing that fast can trigger seizures. It is now capped at three a second, whatever the strike rate is set to, which is the point where that stops being a risk.
You will notice the storm is a little slower at its default setting. That is the change, not a side-effect of it. Everything else about Lightning — the bolts, the branching, the colour that moves with the music — is untouched.
Lightning has also been taken out of the standard mix, so it no longer appears on its own. If you like it, turn it back on in the visualiser style list and it behaves exactly as before, capped.
The storm stops flashing if your system asks it to
Windows, macOS and Linux all have a setting that says *reduce motion*. People turn it on for motion sickness, for migraine, and because flashing light can trigger seizures.
Colada was not listening to it. The Lightning visualiser flashes the whole screen on the beat, and at a busy tempo that is several times a second — which is exactly the kind of thing that setting exists to stop. It is also in the standard mix, so you could meet it without ever having chosen it.
If you have that setting turned on, the lightning still strikes and the storm still looks like a storm — but it no longer flashes the room. Turn the setting on or off and Colada notices straight away; there is nothing to restart. If you do not use that setting, nothing has changed.
If flashing light is a problem for you and you would rather not rely on a system setting, you can also switch Lightning off on its own: open the visualiser style list and turn off its *Mix* toggle.
Listening together can finally be told apart from nothing
Listen Along has shipped since early August and Colada had no way of knowing whether anyone used it. Not "few people use it" — no way to tell the difference between a quiet feature and a broken one.
It now records four things, and only these: that a broadcast started, that it ended and how long it lasted, that somebody tuned in, and that they tuned out, how long they listened and whether they left or the session ended around them.
Each of those carries a random label that is made up fresh every time and thrown away when the app closes. It is not your account, it is not your device, and it is not the same twice — so these numbers can say "a broadcast lasted nine minutes" and can never say who was listening to whom. That was the point of doing it this way rather than the easier way.
Usage sharing is still off if you turned it off, and none of this is sent when it is.
The library stops waiting for a count it does not need
When Colada opens it does two things: it loads your folder tree, and it counts what is in your library for the numbers in the corner. The count went first, and the tree waited for it to finish.
On a big library that count is real work — it walks every file. Nothing about the tree depends on it, so now they start together. If your library is large, the folders should appear sooner; if it is small you will not notice, because there was nothing to wait for.
Showing the 25 most recent of 473 releases. Colada has been shipping small, frequent updates since the beginning — the app itself always tells you everything that changed since your version, however long ago that was.
Get the latest one
Free, local-first, and yours to run on your own machine.