Privacy notice

Meet Replay records audio. Audio is personal data. This notice says exactly what the app touches, where it goes, and how long it is kept, so you can judge it rather than take our word for it.

Last updated: 7 October 2026 (v0.1.86). Meet Replay was called Rewind until v0.1.76; where this notice names an old file or folder, that is why.

The short version

Everything you record stays on your machine. No telemetry, no analytics, no crash reporting, no accounts, no cloud. Nothing you record is sent anywhere.

Meet Replay makes exactly one kind of network request: it downloads a model file, once, from a fixed published address. There are two such files - the speech model, and since v0.1.79 a much smaller one that tells the voices in a call apart (see below). No audio, no transcript, and no filename is ever part of either request. That is the whole of Meet Replay's network use, and it is written down as decision 0017.

The model Meet Replay uses is Base, about 141 MB as your file manager counts it (148 million bytes) - large enough to matter on a metered or slow connection. It is fetched once and kept in %LOCALAPPDATA%\MeetReplayData. The app has no model picker; the only other model it knows, Small (about 465 MB), is used only if somebody names it in settings.json by hand.

Transcription is on by default, so this download happens the first time you save a recording, without being asked for. The app has no switch for it on screen. To turn it off, set "transcribe": false in settings.json (in %LOCALAPPDATA%\MeetReplayData) while Meet Replay is not running; it is honoured from the next launch, and then nothing is downloaded unless you also tick Continuous transcripts.

Nothing about that download describes you or your recording: it is a fixed URL for a fixed file, and the same request every copy of Meet Replay makes.

The second model, for telling voices apart, is about 25 MB (26.5 million bytes), fetched from the sherpa-onnx project's published releases the first time a transcript of a call needs it, kept beside the speech model and checked against a published SHA-256 the same way.

Meet Replay starts recording by itself

Say this plainly, because it is the part most worth knowing: from the first time you open it, Meet Replay launches when you sign in and begins buffering audio immediately. You do not have to press anything, and nothing asks you first. (Windows lets you overrule this: if you switch Meet Replay off under Startup apps in Settings or Task Manager, it stays off, and Meet Replay cannot turn it back on.) That is what an always-on replay buffer is — one that waited to be told would have missed the moment you wanted to keep — but it does mean a fresh install is recording within seconds of your first login.

It is never hidden while it does this. The tray icon is orange the whole time it is recording, grey when it is stopped, and red if capture has failed; the level meters move with the sound; and saving a recording raises a desktop notification naming the file. There is deliberately no way to make Meet Replay record invisibly.

Continuous recording writes its files quietly, since v0.1.87. It keeps what it records in ten-minute files, and writing one is housekeeping rather than an event, so it raises no notification - it is noted in meetreplay.log instead. That does not make recording any less visible: the tray icon is orange the whole time it records, archive or not, the main screen says the archive is on the first time you open Meet Replay, and the switch for it sits on the Continuous tab. A file the archive fails to write is still announced, because a recorder that has quietly stopped keeping what it promised is the failure that matters. (Archive files were announced from v0.1.33 to v0.1.86 under decision 0006; decision 0123 changed that for the archive alone, at the owner's request.)

There is one file that does not raise one, and it is worth naming rather than letting you find it: when Windows itself is shutting down or signing you out, Meet Replay writes whatever was still buffered and then exits. No notification is shown, because the session is ending and there is nowhere left to show it,

• it is added to the Recent recordings list, so it is there when you next

open Meet Replay. It is a normal recording beside the others. Quitting Meet Replay yourself, from the tray, is not this case and does notify.

Untick Start with Windows on the settings page to stop it launching at login. Recording itself has no switch since v0.1.88: it starts when Meet Replay does and runs until you quit it from the tray, the way a game recorder's instant replay does. (Start recording immediately and the Start/Stop button were removed then; quitting is how to stop.)

A fresh install also captures your microphone as well as system output, so the very first buffer may contain your voice. Delete the "Me" source - the microphone, called "Mic" before v0.1.77 - if that is not what you want.

What Meet Replay processes

Audio from the sources you configure. Held in a buffer in RAM so you can save the last few minutes after something happens. Two settings change what reaches the disk without your asking each time, both on the Continuous tab and both described below: Continuous transcripts, on for a new install since v0.1.78, and Continuous recording, on for a new install since v0.1.84. (In v0.1.86 and v0.1.87 they were Background transcripts and Background recording on a Background tab; before that Keep a running transcript, on the Settings page, and Keep an archive, on the Buffer tab - which is the On demand tab since v0.1.88.) With neither on, only a recording you explicitly save ever reaches the disk. The buffer keeps the rolling window you set — five minutes to six hours — and older audio is continuously overwritten and cannot be recovered.

Saving also empties the part of the buffer it wrote, so the next recording starts where the last one ended rather than repeating it. The audio it held is gone from RAM at that point, and only the file you asked for still has it. If the write fails nothing is dropped, so a failed save never costs you the recording.

One exception, and only while Continuous transcripts is switched on: that feature writes audio to disk.

This is one of two places Meet Replay puts recorded audio on your disk without you asking each time - the other is continuous recording, below - and on a new install this one is on unless you switch it off, so it is worth being exact about. Transcribing is slower than recording - on a normal machine, noticeably slower than two people talking at once - so the audio waiting to be transcribed is written into a queue and worked through as fast as the machine can. Without it, whatever could not be transcribed in time was simply lost, which is what made transcripts go blank for hours at a stretch.

What that means in practice:

• It happens only while Continuous transcripts is ticked. Untick it and nothing more is queued. It has never applied to recording on its own.
• What is already waiting stays until it is transcribed or you delete it. Unticking the box, stopping recording or quitting does not empty the queue: the next time the running transcript is on, Meet Replay works through what was left and deletes each piece as before. If you are switching the feature off for good, delete the queue folder yourself.
• Each piece is deleted the moment it has been transcribed, usually within a minute or two, and within a few hours at worst on a slow machine.
• The queue is capped at six hours of audio. Past that the oldest is deleted untranscribed, and Meet Replay says so rather than doing it quietly.
• It lives in a queue folder inside your transcripts folder, so you can see exactly what is waiting and delete it yourself whenever you like.
• Silence is never queued, so a quiet room costs nothing.
• Nothing heard while the Timeline plays is queued, so a replay is not written down a second time. Anything said while it plays is not written down either.
• It never leaves your machine, like everything else here.

If you would rather no audio were ever written without you asking, turn Continuous transcripts off. Recording, saving and the buffer are unaffected by it.

Continuous recording

On for a new install since v0.1.84, and it changes the sentence above - an install from before keeps whatever it was set to. The main screen says so the first time you open it, until you press Got it: what is kept, for how long, and that nothing leaves your PC. Untick Continuous recording on the Continuous tab to stop it. With it on, Meet Replay keeps a compressed copy of everything it records, writes it to disk every ten minutes and once more when you quit, and deletes each day once it is older than the number of days you choose - 30 unless you change it, from 1 to 365, on the Continuous tab right under the switch. Past 365 the slider reads Forever, and then nothing is deleted: the archive grows for as long as it is on, and the tab says so instead of giving a size it will stop at.

What that means in practice:

• Where: a folder called Archive inside your recordings folder - for a new install %LOCALAPPDATA%\MeetReplay\Archive - one folder a day, one .opus file per source. You can choose another folder on the Continuous tab. When you change it, or move your recordings folder while the archive follows it, Meet Replay moves the days it kept there. It moves only its own day folders and never replaces a file the new folder already has.
• How much: speech quality by default, 16 kbps a source - about 7 MB an hour each. You can choose 8 to 128 kbps on the Continuous tab; 128 is about 57 MB an hour a source. The Continuous tab shows a day's worth for your sources and the most the archive can ever hold, which is that times the days you keep.
• What it holds: audio, and the screenshots described below.
• What is deleted: only the archive's own day folders, and only once they are older than your setting. It checks when Meet Replay starts and after every file it writes. Anything else in that folder is left alone.
• What is never deleted: recordings you save with the shortcut or the Save button. Those are full quality, in your recordings folder, and stay until you delete them, archive or not.
• Switching it off throws away what was waiting to be written and stops adding to it. What is already in the archive stays until its days run out, or until you delete the folder yourself.
• Changing a source starts recording again with it, and writes what was waiting first, so nothing is lost; quitting writes it too.

With screenshots - on unless you untick it, and only ever while the archive is on. Meet Replay also keeps small pictures of your screen in the archive: every monitor, shrunk to 480 pixels across and blurred as strongly as you choose. How strongly is yours to choose (since v0.1.86; from 0 to 16 in half-steps since v0.1.88), on the Continuous tab, with a sample window beside the slider showing exactly what that strength leaves. The default, 6, leaves no text readable - checked on a real screen, where even a lock-screen clock became two smudges. Below 2.5 the text in a screenshot can be read, and at 0 it is not blurred at all - only made small, so headings and other large text can be read. The tab says so in words under the slider: at those strengths the archive holds a legible picture of your screen every few minutes, emails and chat included. (The checkbox was With blurred screenshots until v0.1.88, when 0 made "blurred" untrue.) A new strength applies to screenshots taken from then on; those already kept stay as they were taken. The screen is looked at every 30 seconds; a screenshot is kept when it changed and one every five minutes otherwise, so a screen that never stops changing - a video playing all day - keeps at most 2,880 a day. Each is a JPEG of about 10 KB at the default strength and larger the weaker the blur; the Continuous tab shows what that comes to a day, worst case, at the strength you chose. They sit beside that day's audio and go when the day's archive goes. They show you what was on screen when something was said; nothing is read out of them, nothing is searched in them, and like everything else they never leave your PC. They still show the shape of what was there - which window, a shared screen, a face in a video call - so untick With screenshots under the Continuous recording switch if you do not want even that.

Nothing else changes: the tray icon and the meters behave exactly as they do otherwise, and meetreplay.log notes how many screenshots went with each archive file. Recording never becomes less visible than it already is.

(Before v0.1.77 this option was called Save automatically and wrote full-quality recordings into your recordings folder, never deleting them. Those files are still where they were, and the archive does not touch them.)

Saved recordings. A save writes several files that belong together, in a folder of their own named meetreplay-<timestamp>, and each file's name begins the same way:

• meetreplay-*.mkv - the clock picture and a track holding all sources mixed together, which is what a player uses when you press play.
• meetreplay-*.<source>.opus - one file per source, holding that source alone. These are the individual recordings, in a format any audio editor opens.
• meetreplay-*.srt sidecars - transcripts, when transcription is on, plus a .clock.srt timeline written on every save.

(Recordings saved before the rename begin rewind- rather than meetreplay-. Those saved before v0.1.70 sit loose in the recordings folder rather than in a folder each. Older ones still, and rewind-*.mka from before the clock became a picture, are single files with every source as a track inside.)

That mix is worth knowing about before you share a recording: the .mkv contains every source, so deleting a source's .opus file does not remove that source from it, and neither does muting a track in a player. To share one source, share that source's .opus and not the .mkv. This is easier to get right than it used to be - each source is already a separate file - but it is also easier to get wrong in the other direction, because the .mkv is the file that looks like "the recording".

They are written to the folder you chose — by default, since v0.1.80, %LOCALAPPDATA%\MeetReplay (for most people C:\Users\<you>\AppData\Local\MeetReplay), which you can change on the settings page. It used to be Documents\MeetReplay, and Documents\Rewind before the app was renamed: if you recorded into either and never chose a folder, Meet Replay goes on saving there, where your recordings already are, and the settings page shows that folder as your choice. They stay until you delete them; Meet Replay never deletes them and never uploads them.

Your settings (settings.json). Buffer length, quality, sources, hotkey, output folder, and a list of your ten most recently saved recordings - for each, its path, the date and time you saved it, and how long it runs. That list is what the main screen shows, and it is worth naming plainly: ten paths with timestamps say more about when you were recording than a single path did. Deleting settings.json clears it; deleting a recording does not, so a row can outlive its file. For the package - the only build published since v0.1.46 - it is in %LOCALAPPDATA%\MeetReplayData, since v0.1.83 - its own folder, beside yours rather than inside it (v0.1.82 briefly used MeetReplay\App). (Before that it was in %APPDATA%\MeetReplay, and before the rename %APPDATA%\Rewind; the first launch after upgrading moves that folder there whole, the models and the log with it, so nothing is left in the old place and nothing is downloaded again.) A copy run from a folder you unpacked yourself keeps it beside the executable instead, which is what a build from source does; which one applies depends on how Meet Replay was installed, never on how it was launched. It stays until you delete it, and contains no credentials.

Device and process names. Used to list your audio devices and to capture a named application. Held in RAM, and the ones you pick are written to settings.json.

A set-aside settings file (settings.json.broken), only if your settings file ever fails to load - a hand edit that leaves invalid JSON, or a truncated write. Meet Replay keeps the old file under that name instead of overwriting it, so a typo costs the typo rather than every source you had configured. It holds exactly what settings.json holds, the list of recent recordings included, and it stays until you delete it.

A diagnostic log (meetreplay.log, and meetreplay.log.old once it has filled), in the same folder as settings.json. It exists so that a crash or a failure can be explained afterwards, and it never leaves your machine - nothing sends it anywhere, and it reaches us only if you choose to send it. It records the version, when Meet Replay started, problems and errors as they happen, the graphics adapter used to draw the window, and the work of the transcription catch-up: the names of your recordings, the names of your sources and the folders they are in, so like the recent-recordings list it says when you were recording. It holds no audio and no transcript text. It is bounded: at 256 KB it becomes meetreplay.log.old and a new one starts, so the two together never pass half a megabyte. Delete either at any time.

That is the complete list. There is no identifier, no usage counter, and no profile of any kind.

What Meet Replay does not do

• No network connection other than downloading the speech models, and only while transcription is on - which it is by default, see above.
• No telemetry, analytics, crash reports, or update checks.
• No account, sign-in, or licence check.
• No audio, transcript or filename ever leaves your machine, including while transcribing. Speech recognition runs entirely on your computer, using a model file on your disk.
• No content analysis beyond the transcript you asked for. Meet Replay does not summarise what you recorded, and builds no index of it: the Timeline tab reads your transcript files and lists your archive when you open it, searches them on your PC, and keeps nothing of what you searched for or played.
• No continuous recording while it appears stopped. See below.

Transcripts of saved recordings

On by default (see the top of this notice for how to turn it off). Worth reading, because it changes what is on your disk rather than what leaves it:

• Meet Replay writes .srt transcripts beside each recording, in its folder — one per source, plus a combined one. A transcript is a searchable text record of what people said, which is easier to skim, copy and quote than the audio, so treat it as at least as sensitive as the recording. Deleting the .mkv alone does not delete its transcripts; deleting the recording's folder does.
• The speech model is downloaded once to %LOCALAPPDATA%\MeetReplayData and checked against a published SHA-256 before it is used.
• It runs when you save a recording, not continuously. The separate setting below is the one that runs continuously.

The running transcript

On for a new install since v0.1.78, because a transcript you can hand to an AI is what the app is for; an install from before keeps whatever it was set to. Untick Continuous transcripts on the Continuous tab to stop it. With it ticked, Meet Replay transcribes what is being said about once a minute while it is recording and appends it to a text file - one file per day, in Transcripts inside your recordings folder - by default %LOCALAPPDATA%\MeetReplay\Transcripts - unless you choose another folder on the Continuous tab. If v0.1.70 renamed that folder to _transcripts on your machine, Meet Replay moves it back on the next launch and says so in meetreplay.log; if the move cannot be done, your transcripts are left exactly where they are and it says that too.

• What it keeps is words. The audio waits in the queue described above only until it has been transcribed, and then it is deleted; the day file holds only text. That is less on your disk than a recording - and, in the way that matters here, something more exposing: a searchable, greppable, quotable record of everything said near the machine, with a wall-clock time and the source's name on every line. It is easier to read a year of these than to listen to an hour of audio.
• It runs only while recording is running, because it takes its audio from the recording. Quitting stops it. Unticking the box stops it without stopping the recording.
• They are kept Forever unless you choose otherwise (since v0.1.86), on the Continuous tab under the switch: 1 to 365 days, or Forever past the end. Forever is the default, so nothing you already had is deleted by upgrading. With a number chosen, Meet Replay deletes day files older than that when it starts and once an hour - only files named the way it names them (meetreplay-<date>.log); anything else in that folder is left alone. Uninstalling never deletes them.
• It needs the speech model, so the note above about the download applies to this too.

Telling voices apart

Always on since v0.1.80, with no switch - it was one in v0.1.79. Each voice is written as its source and a random tag, such as Them #AAECAwQFBg (since v0.1.82; they were numbers before). In the running transcript every voice gets a tag - yours too, on Me, since v0.1.83 - in the running transcript and in a saved recording's transcripts alike, so the app never guesses which voice is you. Switching transcription off is how to stop it.

• How it tells them apart. For each stretch of speech, a model on your PC turns the sound of the voice into a list of numbers - a voice embedding - and compares it with the voices already heard. Close enough is the same voice; otherwise it is a new one.
• A voice embedding is a kind of voiceprint, which some laws treat as biometric data. Meet Replay keeps them only in memory and never writes one to disk: the running transcript forgets every voice at midnight, and a saved recording forgets them once its transcripts are written. What reaches your disk is only the label - Them #AAECAwQFBg - beside the words.
• Nobody is named and nobody is recognised. A tag is random and means only "the same voice within this transcript": it is never reused, so the same person in two meetings has two different tags, and nothing links them. Nothing is sent anywhere, as with everything else here.
• It needs the second model described at the top of this notice: WeSpeaker's ResNet34-LM speaker model, trained on VoxCeleb, by the WeSpeaker authors, published as ONNX by the sherpa-onnx project, used unmodified under the Creative Commons Attribution 4.0 licence.

Recording other people

This is the part that deserves your attention. Meet Replay captures whatever your system plays, which can include other people's voices — a call, a meeting, a stream. Laws on recording other participants differ by country and by US state, and many require that everyone knows, or consents.

Meet Replay is deliberately built so recording is never hidden:

• The tray icon is orange while recording, grey when stopped, and red if capture has failed, so it can never silently look idle while running.
• Saving a recording always raises a desktop notification, even when the window is closed - the one exception being the save made as Windows itself shuts down, described above. Continuous recording's own files are written quietly, as described above; a failure to write one is announced.
• Live level meters show which sources are actually capturing.

We will not add a hidden or silent recording mode. If you use Meet Replay around other people, tell them.

Your data, your control

Because nothing leaves your machine, there is no data of yours for us to hold, export, or delete — the usual data-subject rights have nothing to act on at our end. In practical terms:

• Delete recordings: delete the recording's folder, meetreplay-<timestamp> (or rewind-<timestamp> for one saved before the rename), not just the .mkv inside it. A save is a group - the .mkv, one .opus per source, and the .srt sidecars - and the .opus files hold each source's audio on its own, so deleting only the .mkv removes the mix and leaves the individual recordings behind. (A recording saved before v0.1.70 has no folder: delete every file beginning with its name, which rewind-<timestamp>.* selects. Older single-file recordings are just the one .mka or .mkv.)
• Delete settings and the log: delete settings.json, meetreplay.log and meetreplay.log.old, and settings.json.broken if one is there, all in %LOCALAPPDATA%\MeetReplayData. (A copy you built or unpacked yourself keeps them beside MeetReplay.exe instead, under the same names.)
• Clear the buffer: quit, or change a source (which starts recording again). The buffer is RAM-only and does not survive. Saving a recording also clears the part it wrote, and shortening the buffer length discards whatever no longer fits, both while recording continues.
• Remove autostart: untick Start with Windows, or switch Meet Replay off under Startup apps in Windows Settings. It is part of the installed package, so uninstalling removes it too. (A copy you built or unpacked yourself uses a MeetReplay value under HKCU\Software\Microsoft\Windows\CurrentVersion\Run instead; untick the box before deleting such a copy's folder, or the value is left pointing at an executable that is no longer there.)

Uninstalling leaves behind the recordings you saved, their transcripts, your running transcript's day files if you kept one, any audio still waiting in its queue folder, the archive in %LOCALAPPDATA%\MeetReplay\Archive if you kept one, your settings and log in %LOCALAPPDATA%\MeetReplayData, and the speech model - ggml-base.bin, 141 MB, in that same folder - with, if voices were told apart, the 25 MB speaker model wespeaker_en_voxceleb_resnet34_LM.onnx beside it. The uninstaller does not remove the recordings, transcripts, queue, archive or settings because they are yours; delete them as described above if you want them gone.

The model is a different case and worth being plain about, because it is by far the largest thing that stays: it is left so that reinstalling or upgrading does not download it again. It contains nothing about you - it is the same published file everybody downloads - and deleting it is safe at any time. Meet Replay fetches it again the next time it needs it. Measured after an uninstall: 141.1 MB remains, and all but 0.1 MB of that is this one file.

A caution about the recordings themselves

Once you save a recording, it is an ordinary file: it is as private as wherever you put it. If your output folder is inside a synced folder (OneDrive, Dropbox, Google Drive), your recordings will be uploaded by that software. Meet Replay cannot know or prevent this — choose the output folder deliberately. The same is true of the transcript and archive folders, and the archive holds sound and screenshots from the whole day.

The default is chosen so this does not happen. Since v0.1.80 Meet Replay saves to %LOCALAPPDATA%\MeetReplay, which OneDrive and Windows backup never sync. It used to save to Documents\MeetReplay, and on many Windows PCs Documents itself is backed up to OneDrive - Windows 11 switches that on during setup with a Microsoft account, often without asking. If you used an earlier version, Meet Replay kept your folder where it was, so your recordings, transcripts and the transcript queue may still be in Documents. Since v0.1.98 the archive follows them there: before, it was always in %LOCALAPPDATA%\MeetReplay\Archive, and Meet Replay moves it into your recordings folder the first time it starts. Meet Replay does not check whether a folder is synced; it did from v0.1.77 to v0.1.97. To check yourself, open File Explorer: a cloud or tick icon beside the Documents folder means it is synced. Choose a folder outside it - recordings on the Settings tab, transcripts and the archive on the Continuous tab - if you do not want that.

Contact

Questions about this notice, a privacy concern, or a request about your data: email support@meetreplay.com. It reaches the developer directly. Because Meet Replay sends nothing to us, we hold nothing about you to look up; we can explain what the app keeps on your PC and how to remove it. Security problems can go to the same address - see SECURITY.md.