Vommet diagnostics privacy
Vommet can send diagnostics so we can find what's slow or broken, for example why sending a photo takes a minute on one computer. It's off unless you say yes, and you can change your mind any time in Settings › Privacy. This page covers these diagnostics, the services some features talk to, Vommet's check for new versions, and Vommet on the web.
What is and isn't collected
Collected, if you say yes
- How long things take: starting up, syncing, opening a room and loading older messages, each step of sending a file, joining a call
- Sizes and counts: file size and picture dimensions, how many rooms and accounts, call length and number of people in it
- Whether something worked, and if not, what kind of error (for example "timeout" or "server said too large"), and the server's standard error code from a fixed list (like
M_FORBIDDEN), never its message - Once per session and for each signed-in account (numbered 1st, 2nd and so on, never which account), whether notifications can work: which notification system the app uses, whether it has permission, whether a push app and a push registration with your homeserver exist, whether "mute everything" is on, and how many of your rooms are unread, muted or set to mentions only. Yes/no answers and counts only, never the push address or which rooms
- On Android with UnifiedPush, so we can tell why notifications stop arriving: what kind of push app you use (ntfy, NextPush, Sunup, the built-in one or another), whether its push address is on ntfy.sh, on Mozilla's push service or somewhere else, which kind of push gateway is used, whether the push address changed, whether setting it up failed, and whether your homeserver had removed this phone's push registration (and how many times in the last 7 days). Each time Vommet checks that registration (when the app starts or comes back to the screen), what it found. And when Vommet tests the push address by sending itself an empty test message through the push gateway, what kind of gateway that was, its answer code, whether the message arrived and how long it took. Fixed words, yes/no answers and numbers only, never the push address, the push app's server or your homeserver
- How many notifications Vommet handled since the last report, and what happened to them: shown, shown quietly (for example because you were chatting in that room on another device), or not shown and why (that room was open, Do Not Disturb, notifications turned off, or another rule), plus how long handling them took. On Android, also how many arrived while the app was running and how many arrived in the background, and how often Vommet had to start its background notification handler because nothing was listening. Counts only, never which rooms, who sent them or what they said. These counts are kept on your device until they're sent, only while diagnostics are on, and are deleted when you turn diagnostics off
- Which experiments you have switched on, by their internal code names (for example
vommet_experiment_forward_messages), so we can tell when a feature is ready to leave the Experiments page - Which of the settings that used to be experiments you have switched off (noise suppression, automatic gain control, newest unread DMs first, sliding sync, secure messaging setup at sign-in, show deleted messages, and the Layout options under Rooms and spaces: shrinking banners, Online and Offline member list, who is in the call, member panel tabs, resizable panels, space home page room list, unread DMs on the Home button, and the member list's role sections: headings for owners, admins, moderators and the room's own roles, and keeping offline role holders in their section), by the same kind of internal code names, so we can see how many people turn each one off
- When the app crashes: the error's type and where in Vommet's code it happened (on Android, as positions inside the Vommet app itself plus the build's identifier, which we look up afterwards against that build)
- In encrypted chats: how many messages couldn't be decrypted when you opened the room and when you left it, how many were unlocked while you were there, why they couldn't be read (the sender never sent the key, the secure channel to them is broken, the key can be asked for again, or the sender didn't share it with people who joined later; counts per reason), and whether a "retry decryption" worked (counts only, never the messages)
- When earlier messages of an encrypted chat are shared with someone you invite, or with you when you accept an invite: whether it worked or why not (from a fixed list of reasons), how many message keys were shared and how many weren't, and to how many of their devices. Each time you start a new encryption key for a chat: whether people invited later may receive it (yes, no, or unknown). Counts and fixed words only, never which chat, who was invited, or any keys
- Once per account, how your encryption is set up: whether encryption, cross-signing and key backup are on and their secrets are available on this device, whether this device is verified, how many devices your account has and how many of them are unverified or blocked, and how many of your rooms are encrypted. Yes/no answers and counts, never keys, device names or IDs
- Each time you send a message, reply, edit, thread message, reaction, deletion or sticker: how long it took, whether it worked and what kind of error if not, how long the message was (in characters), whether it had formatting and how many people it mentioned. Never the text, the sticker, who you mentioned or which room
- Every 15 minutes, for each kind of request to your homeserver (syncing, sending, downloading and uploading media, fetching keys, profiles, room state, older messages, members, push and account settings): how many, how many failed, how long they took and how much data came back, and whether it was your own server. The kind is worked out from the shape of the address only; the address itself (with its room, user, media and message IDs) is never kept
- During a call, once a minute: the connection quality word (excellent, good, poor, lost), how many people are in it and how many of them report a poor connection, audio packets received and lost, jitter, how much audio had to be patched over (the glitches you hear), bitrate, round-trip time, video frame rate and size, whether a screen share is on, and how many other people's audio streams are being received, muted or not received; in encrypted calls also how many of those streams decrypt fine, are waiting for or missing their key, or fail to decrypt, and how many people's keys have arrived; how many video, screen share and audio tiles the call view has to show; and your microphone's noise suppression: whether its experiment is on, the mode and engine it uses, whether the built-in suppressor is on, your microphone volume setting, the sample rate it receives and how many audio blocks it cleaned, skipped or converted from another sample rate; and whether the filter has switched to the lighter RNNoise filter because processing was too slow or its output stayed silent despite speech-like input, or has not switched. This is the state at the time of the sample, not a count of switches. Numbers, yes/no answers and fixed words only, never who is in the call or any audio or video
- When you join a call: whether your microphone started and how long that took, whether it joined open, muted or on push to talk, whether it used the microphone you picked in Settings, the default one, or the default because yours was missing, what kind of microphone it is (built in, USB, Bluetooth, virtual or HDMI), how many microphones a computer has, and if it failed, the kind of error and its type name. Never the error's message, your microphone's name or any audio
- When you record, send or play a voice message: for recording, whether you tapped or held the button, how long the recorder took to start, how long the recording is, its size and format and how loud its loudest moment was (a number, to spot a silent microphone), or why it ended without one (no microphone permission, too short, canceled and similar); for sending, how long it took, its size, whether the room is encrypted, and if it failed, the kind of error and its type name; for playing, how long fetching the file and opening the player took, how long until playback started, the length of the message, its file type and whether the room is encrypted, or which step failed or never finished. Never the recording itself, its transcript, a file name, which room or who sent it
- When a voice message, video or file takes more than 10 seconds to load, and when such a load ends: which kind of media it is, which step it was on (looking it up in the local cache, finding its address, waiting for the server, downloading, decrypting, saving, or the player opening it), how many bytes had arrived and the file size, how long it took, whether the room is encrypted, whether the newer or older media address was used, and whether it worked, failed (the kind of error, its type name and the server status code) or was closed first. Never the file, its address, a file name, which room or who sent it
- When the media player behind a voice message, video, GIF or sound takes more than 10 seconds over something (starting, opening a file, playing, pausing, changing the volume or speed and similar): which kind of player it is, what it was doing, what it was waiting for (the player starting, the video output starting, another task on the same player, or the player itself), how many players of each kind were open, how many minutes Vommet had been running, and whether it finished late (and how long it took) or never finished. The same player details are added when a voice message never finishes opening. Never the file, its address, a file name, which room or who sent it
- Why the app was last closed. On Android, the reason the system gives (for example a crash, low memory, or you closing it). On a computer, only when the last session didn't end normally (a crash, a force-quit or a power loss): how long before that it was last seen running (Vommet notes this about once a minute while it's open), which Vommet version it was, and whether a call, a screen share or the camera was on at the time (yes or no). Never file paths, error messages or which call
- When your microphone, camera or screen sharing fails to start: what kind of device it was (built-in, USB, Bluetooth…) and how many there are, never their names
- When you share your screen on a computer, once, a few seconds in: which audio you picked (none, all apps, or one app, but never which app), how many apps were playing sound, and on Linux how many sound streams were recorded, whether they came from app playback or another source, and how long the shared audio had sound in it. Never app or device names, and never the audio itself
- If you use the "Nickname per space and room" experiment: each time Vommet updates your name in a batch of rooms, how many rooms it updated, left as they were or couldn't update, whether the server asked it to slow down, and at which step and with what kind of error it failed. Never which rooms or spaces, and never the names
- How smooth the app runs (frame rate summary, split into the time spent preparing and drawing each frame), how often and for how long the app briefly freezes (and, the moment a freeze passes 5 seconds: which part of the app stopped, whether you were in a call, how many seconds since your last call ended and how many people that call had at most), how long it takes to process each batch of new messages from the server (split into saving it and the follow-up steps, such as fetching other people's encryption keys), how many rooms and messages each batch contained, how often the server asked the app to start syncing over (and, once per launch, whether the app had a saved place to resume from) and how many rooms it then sent again from scratch (counts only, never which rooms), how often each kind of database operation ran and how long it took, memory use, and how much of it the app's cache of decoded images holds (sizes and counts, never which images), and after the app comes back from the background: how long it was away and how long until new messages arrived; also how long new messages were ever silent while the app was in front (the longest wait between two syncs and how many waits passed 90 seconds), how many room timelines are open, the size of the app's database on your disk, how far your device's clock is from the server's, and when your system asks the app to free memory: how much it was using at that moment; and how long each startup step took (including, once per launch, when the first request for new messages went out, when its answer arrived and when it was processed, and how long the app was too busy to respond until then)
- Your device in broad terms: Windows, Linux, Android…, the system version number, Vommet's version, number of CPU cores, total memory (and on Android whether the system calls it a low-memory device), screen size and pixel density, your text size setting and whether bold text, reduced motion or high contrast are on, and on Linux which desktop family (GNOME, KDE, Sway…) and whether software rendering is forced. No device names, serial numbers or graphics driver names
- A frame counter, summed up every 15 minutes for each part of the app you had open (home, a space, a room or settings; just which kind of view, never which room or space): how many minutes it was showing, how many frames the app drew in that time and in its busiest minute, and how long the slower frames took to prepare and to draw; on Linux also how often the system redrew the app's window. Counts only, never what was on screen
- When you unlock your messages on a new device (with your recovery key, phrase or password, or by approving it from another device): which of those three ways you used, how long each step and the whole "Unlocking your messages" screen took, and if it failed only what kind of failure (wrong or mistyped key, nothing stored to unlock); then, for the download of your older messages' keys from your online backup that follows in the background: how long it took, how many keys the backup holds, and whether it worked. Never the key, phrase or password, and never which rooms
- If an older recovery key of yours can still open copies of your secret keys (left behind when you replaced your key in an earlier Vommet version): once per launch while that cleanup is waiting, how many secrets still have such copies; and when you enter your current key and Vommet removes them: when it happened (unlocking or checking your key), whether it worked, and how many copies were removed, kept or failed. Counts only, never the keys, the secrets themselves, or which secrets
- Once per launch, for the first batch of new messages from the server: how big it was (in bytes), how many rooms it covered and how many of them had new messages, how many messages, room settings and members, typing and read notices, account settings, online-status updates and encryption messages for your devices it contained, for how many people the server said their devices changed or that you no longer share a room with, and how long saving it took, including the total time spent in the database. Counts and times only, never which rooms, people, devices or messages, and never what they say.
- When Vommet sends out a key for your encrypted messages in a room (when it starts a new one, or gives the current one to devices it has just learned about): how many people in the room it was for, how many of them Vommet didn't yet have an up-to-date list of devices for, how many of those lists it fetched first and how long that took, how many people still had no known devices, and how many devices were blocked, left out by your key-sharing setting, got the key, couldn't be reached, or failed to send. Counts and times only, never which room, people, devices or keys.
- When you verify one of your devices or another person (for example approving a new device from your phone with the emoji): whether this device asked or answered, whether it was your own device or another person, the method (emoji or QR code), whether it finished or was cancelled (and the standard cancel reason, like "timed out" or "declined", and which side cancelled), how many milliseconds after the request each step happened (the other side getting ready, the keys exchanged, the emoji shown, your confirmation, the final check), how long the request dialog took to appear and how long you took to tap Accept, how old the request was when it arrived, whether it arrived as the app was just starting, and how busy the app's syncing with your server was meanwhile (how many sync rounds, how long they took to process, and how long verification messages waited before being handled). Numbers only, never which people or devices, and never the emoji, codes or keys.
- When you open someone's profile: how many milliseconds after your tap it appeared and its pictures and colors arrived (the small and the full banner and avatar, or that one failed), how long your server took to answer for their profile and their online status (and whether that worked, or the kind of error), where the profile's colors came from (their own color choice, a picture, remembered from earlier, or the default), whether the profile was already known to the app, whether they are on your own server, whether you closed it before it finished, and whether they have a banner, an avatar or a shared timezone. Never who they are, their server, their name, their pictures or anything on their profile.
- In the web version only (Vommet in your browser), once per session: whether the WebAssembly or the JavaScript build is running and, if JavaScript, why (your browser lacks a WebAssembly feature it needs, or the browser isn't yet on the list allowed to use it); the browser engine family (Chromium-based, Firefox-based, Safari/WebKit, or other) and the browser's major version number (like 141); whether the page runs in the browser's isolated mode (cross-origin isolation); and which WebGL version is available (2, 1 or none). Never the browser's full user-agent text, your operating system, screen or graphics card name
- For each account, whether it's on nether.im (Vommet's default server) or on another server, never which other server
- When a reply quotes a message that isn't already loaded: every 15 minutes (or when the app goes to the background), how many quoted messages were found on your device, fetched from your homeserver, or couldn't be found anywhere, how many of those lookups failed and of what kind (not found, not allowed, timed out, other), how many were in rooms that live on a different server than your account, and how long the slowest one took. Counts only, never which messages, rooms or servers
- When loading older messages fails: the error's type (a name from Vommet's code, as in crash reports), and whether the room lives on a different server than your account (yes or no, never which server)
- If you encrypt the data Vommet keeps on this device: once per launch, whether encryption is on and whether you chose that yourself, whether the next start still has to encrypt or decrypt files, where the key is kept (system keyring, Flatpak portal, Windows account, a key file, a key file locked with a password you chose, or none yet), whether a system keyring answered, the key's state at start-up, and how many of the app's databases are encrypted or not. When you turn it on or off: the new setting and where you did it (first-run page, settings or the notice). For each start-up conversion: encrypt or decrypt, whether it finished, ran out of space, failed (and at which step) or was interrupted, roughly how long it took, how many files it worked on, and whether the database keys were removed afterwards. For the start-up screens about the key (keyring locked, key missing, keyring not answering, stored messages that don't open, or asking for your password): which screen, what you did, roughly how long the screen was up, and whether the old files were kept; on the password screen also how many passwords didn't open the key and whether it was shown in a Steam Deck's Game Mode. Never file paths, account names, keys, passwords or error messages
- Your name tag, only if you type one in (see below)
- The file type of things you send (like
image/jpeg), and whether the chat it happened in is encrypted (yes or no, not which chat)
Never collected
- Your messages, in any chat, encrypted or not
- Your voice messages, or the audio or video of your calls and screen shares
- Your photos, videos or files, or previews of them
- File names
- Names or IDs of people, rooms, spaces or servers
- Your Matrix account or password
- Error message text (it can quote what you wrote)
- Your contacts or who you talk to
- Your IP address (see below)
How "never" is enforced
We don't rely on promises alone. Every report must match a fixed list of fields, and each field can only hold a number, a yes/no, a choice from a fixed list, or a location in Vommet's source code. There is no field that can hold free text, so there is nowhere for a message, a file name or a name to go. Vommet checks every report against this list before sending it, and our server checks again and rejects anything that doesn't match.
The list is public: proxy.nether.im/telemetry/schema. The code that does the checking is in Vommet and vommet-proxy.
You can see exactly what is sent
Settings › Privacy › What's been sent shows every report your device sent since Vommet started, exactly as it left your device.
Who receives it, and for how long
- Reports go to
proxy.nether.im, a server run by the Vommet maintainers. They are read only by the people working on Vommet, to fix problems. They are not sold, shared or used for advertising. - Reports are filed under a random ID created on your device, not under your Matrix account. You can find this ID in Settings › Privacy, and share it with us if you want us to look at your reports when you report a problem.
- Diagnostics are pseudonymous, not anonymous. They don't contain your account or anything you wrote or shared, but they do contain exact details such as a photo's file size, when it was sent, how long a call lasted and what device you use. Someone who also runs the server could match those details to your activity, for example to a particular upload or call, and in a small group of testers your device details alone may single you out. We don't do this to snoop. We may use it, for example, to line up your diagnostics with our server logs when you report a problem.
- Our server does not store your IP address. Like the rest of nether.im, it sits behind Cloudflare, which necessarily sees your IP address to deliver the request.
- Reports are deleted automatically after 30 days.
Optional name tag
Under the diagnostics switch you can type a name tag, for example the name you go by in the testers' chat, so we know which reports are yours and can group them across your devices. It's empty unless you fill it in. Vommet never fills it in for you from your Matrix account or display name.
- With a tag, your reports are not anonymous: the developers will know they're yours. Leave it empty to stay pseudonymous.
- It can only contain up to 32 letters, numbers, spaces,
-and_, and it's saved in capitals (so "dark" and "Dark" are the same tag). Both Vommet and our server reject anything else. Settings › Privacy shows exactly which tag is being sent. - The tag is sent with every report while it's filled in. Clearing it stops tagging new reports. Reports already sent keep their tag until they're deleted after 30 days, or until you use Delete my diagnostics.
- Use the same tag on each of your devices if you want them grouped. Deleting is per device: Delete my diagnostics removes everything sent from that device, tagged or not, but not reports from your other devices. Use it on each device.
Stopping and deleting
- Turn diagnostics off in Settings › Privacy. Nothing more is sent, and anything waiting to be sent is thrown away.
- Delete my diagnostics in the same place asks our server to delete everything sent from your device, including reports with your name tag, then starts over with a new ID.
GIF search
New installs start with GIF search on. You can turn it off during first-run setup or later in Settings › General. When you search for a GIF:
- Your search words go to
proxy.nether.im, a server run by the Vommet maintainers. It passes only the words on to KLIPY, the GIF service, and fetches the GIFs for you. KLIPY sees what you searched for, but not your IP address or your Matrix account. - Like any website,
proxy.nether.imsees your IP address and when the request came. - With GIF search off, Vommet makes no GIF requests at all. You can still save GIFs people post and send them from My GIFs.
Link previews
Settings for this are in Settings › Privacy › Link previews, and each room can choose its own.
- Links you send: on desktop, Android and iPhone, Vommet makes the preview on your device and attaches it to your message, so the people you send it to don't have to fetch the link themselves. To build it, your device opens the page, so the site sees your IP address, as if you had opened the link. Nothing goes through us or your homeserver for this. In encrypted chats the preview, including its picture, is encrypted like the message. This is on by default. Tap × above the message box to send one message without a preview, or turn off "Attach previews to links I send".
- Links you receive: by default, in encrypted chats you only see previews the sender attached, and nothing fetches the link. In other chats your homeserver makes the preview, so it sees those links. "Bundled only" stops that everywhere, and "Off" shows no previews at all.
Translation, transcription, spellcheck, search and camera blur
These run on your own device. Your messages, your voice messages and what you type never leave it for them. Some of them need a file downloaded once, and Vommet asks before downloading it:
- Translation language models come from Mozilla's server for Firefox Translations (
firefox-settings-attachments.cdn.mozilla.net). Delete them in Settings › App › Translation. - Voice message transcription uses a Whisper speech model from Hugging Face (
huggingface.co). - Spellcheck dictionaries (Linux and Windows) come from LibreOffice's dictionary collection on GitHub (
raw.githubusercontent.com). Pick them in Settings › General › Spellcheck. - Each download is a plain file request with nothing about you in it. Like any website, the server sees your IP address, when the request came and which file (and so which language) you asked for. Vommet checks each file's fingerprint before using it.
- Search in encrypted chats and in all your direct messages builds its index on your device and keeps it there. Choose Off, Wi-Fi only or Always in Settings › Privacy › Search. Deep Search, on the Overview page, searches that index on your device. For unencrypted group rooms it also sends your search words to your own homeserver, the same way searching inside one of those rooms always has. Your last five searches are kept on your device only, and you can clear them there. With usage data on, Vommet reports only how long a Deep Search took, never what you searched for.
- Camera background blur finds the outline of you in each video frame on your device, using a small model that comes with Vommet, so nothing is downloaded for it. Only the blurred video goes into the call, the same way your camera always does. With usage data on, Vommet reports only whether the blur is switched on, never anything from your camera.
Privacy-friendly frontends
Off by default. In Settings › Privacy you can have links to YouTube, X, Reddit and other sites open on alternative frontends such as Invidious, Nitter and Redlib, on instances you pick from a built-in list or add yourself. This only changes what your own device opens. The instance you pick sees your IP address and what you open there, instead of the original site. With YouTube on Invidious, videos in the chat play from your Invidious instance. We don't run any of these instances.
Checking for new versions
This is separate from diagnostics and doesn't depend on them. Once each time Vommet starts, and when you tap Check for updates in Settings › About, release builds ask nether.codes for the list of Vommet releases, so they can show a bar at the bottom when a newer one is out. If you choose to update, the file comes from nether.codes too, and Vommet checks its signature before installing it.
- The request carries nothing about you: no account, no ID, nothing you wrote. Like any website, nether.codes sees your IP address and when the request came.
- Choose what happens in Settings › General › New releases: Off, Notify and ask, or Download automatically. With Off, Vommet doesn't check unless you tap Check for updates. Installing always takes a tap.
- Vommet never asks Commet's update servers.
Vommet on the web
On web.vommet.app, the app is delivered by Cloudflare Pages, the same way as this website. Everything above applies unchanged: diagnostics are off unless you say yes, and the web version asks the same question the first time you sign in. It doesn't check for new versions (the website always serves the latest one).
- Cloudflare delivers the app's files, so like any website it sees your IP address, your browser, and which files were loaded and when. It does not see your messages: once the app has loaded, it talks to your homeserver directly, not through web.vommet.app.
- We don't add analytics, cookies or trackers to web.vommet.app.
- Your messages, encryption keys and settings are stored in your browser, for web.vommet.app only. Signing out or clearing the site's data removes them.
- If you host your own copy, your server delivers the app instead of Cloudflare, and none of it goes through us.
Questions
Ask in #vommet:nether.im or open an issue.