Privacy policy
This is what Selver collects, where it goes, how long it is kept and what you can delete. It describes Selver as it is built today. Selver is unfinished, and where something is not finished yet, this policy says so.
1The short version
Selver records a conversation only when you start a recording.
The recording stays on your phone. Your phone tells the voices apart, finds yours and writes down what was said. The audio is not sent to Selver.
What was said is sent to Selver as text, with when each of you spoke, so that a language model can draft your reflection. Selver does not keep that text once the reflection is made.
When nobody else can be heard, you can choose “Only me”, and Selver takes your word for it. Your phone then writes down only what it hears as your voice, and that is all of the conversation it sends Selver.
A reflection quotes you and nobody else. Selver reads the other person’s words to follow the conversation and does not keep them.
You can keep lines of your own from a reflection, in Kept. Your phone notes which, and forgets them when their reflection is deleted.
The names you give the people you talk to are kept on your phone only.
You can leave yourself something before and after a conversation, spoken or written. It is kept on your phone only.
You can delete a reflection, every reflection, what Selver knows of your voice or your whole account in the app.
Selver does not sell your data and does not collect analytics about how you use the app.
2Who “Selver” is
In this policy, “Selver” and “we” mean the makers of the Selver app and the service they run. You can reach us at kendara2023@gmail.com.
3What Selver collects
Your account. Your email address, your name and a password, or a sign-in through Google if you choose that. If you sign in with Google, Selver receives your name and email address from Google, and your profile picture if Google sends it.
A sample of your voice, on your phone. When you set Selver up, you read a short passage aloud, for up to 20 seconds. Your phone keeps what it needs from that sample to tell which speaker in a conversation is you. It is not sent to Selver. Without it, Selver cannot reflect on a conversation. There is more about the sample below.
Recordings of conversations, on your phone. When you start a recording, your phone records everyone the microphone can hear, not only you, until you stop it or it has recorded an hour. It goes on recording while your phone is locked or you use another app, and pauses while a call or another app has the microphone. The recording is not sent to Selver. If you chose “Only me”, then once your phone has written down what you said, the recording it keeps is a copy in which everything it did not hear as your voice is silent.
A few seconds of your own voice for each moment a reflection points to, on your phone, so that you can hear yourself again. They are cut where your phone heard only you, it plays one only where it can show that it heard nobody else, and they are not sent to Selver. Your phone tells voices apart by how they sound, so a voice it could not tell from yours, or one speaking at the same moment as you, can be heard in one.
What your phone makes from a recording. A transcript, and counts and timings, such as how long each of you held the floor and how often you cut in. The transcript and the timings are sent to Selver so that the reflection can be made. If you chose “Only me”, the transcript and the counts are of what your phone heard as your voice. Selver takes such a conversation only if every part of it is marked as yours, and takes none of it otherwise. It keeps nothing for what takes two people to know, such as how long anyone else spoke or how often you cut in: that is not known, and Selver does not store a nought in its place.
The reflection. It holds Selver’s sentences about the conversation, a small number of your own lines quoted from it, the counts, what it looked at most closely (your questions, for example, or how the conversation ended), one thing to try next time when there is one, and Selver’s estimates. For a conversation you chose “Only me” for, the reflection says that it heard only you, and its counts are of your own speech only. Selver stores it. Your phone makes the spoken walkthrough from it.
Lines you keep, on your phone. When you keep one of your own lines from a reflection, your phone notes that you kept it, when, and a copy of its words; which line Selver suggested from that reflection, and your answer; and whether you have asked for it not to be brought back to you. Your phone does not send any of it to Selver.
What you leave yourself, on your phone. Before a conversation you can say or write what you want from it, and afterwards something to remember for next time. A spoken one is a recording of up to 20 seconds that your phone makes when you press to speak; it holds whatever the microphone hears in that time, so it is meant for a moment when you are on your own. A written one is up to 140 characters. Your phone keeps it with who the conversation was with and when you left it, and can note which conversation it followed, if Selver listened to that one. Nothing writes down what a spoken one says, and none of it is sent to Selver.
A record of what your phone sent, on your phone. Each time your phone sends a conversation to Selver as text, it notes which conversation, when, and how many words. When it deletes the recording, it notes when and, if it kept clips of your own voice, how many. You can read this record under You, in “Privacy and your data”. Your phone does not send it to Selver.
What “Only me” notes on your phone. With a recording you chose “Only me” for, your phone keeps that you did, and when you said that nobody else could be heard. While its reflection is on its way, your phone notes that it was “Only me”. And with the name of each person you have recorded with, it keeps whether your latest recording with them was “Only me”, so that the recorder starts there next time. Only the first is sent to Selver, with the words, and with which wording you confirmed.
Your answers and settings. What you tell Selver you want to work on, whether you tried the move it suggested, and any feedback you choose to send.
A notification address. If you allow notifications, Selver stores the token your phone gives it, so that it can tell you when a reflection is ready.
A record of agreement. Selver stores a record that you agreed to the Terms and to this policy, that you agreed to your voice being used to find you in your recordings, and that you confirmed before each recording that everyone present knew it was being made. A conversation you chose “Only me” for has no record that everyone knew: you said nobody else could be heard, so there was nobody to tell.
What you said about who could be heard. For a conversation you chose “Only me” for, Selver stores that you said nobody else could be heard, when you said it, when Selver recorded it, which wording you confirmed, and which conversation it was for. It holds nothing that was said.
What Selver does not collect. Selver does not collect analytics about how you use the app. It does not ask who a conversation was with: the names and colours you give people, and which conversation was with whom, are stored on your phone and are not sent to Selver. Which of your kept lines came back to you, and when, is noted on your phone only and is not sent to Selver. When you opened Before for someone, and when you waved away a question on Today about seeing someone or leaving yourself a word, are noted on your phone only and are not sent to Selver. Which plan you chose while Selver cannot keep one, and when, and that you said “Not now” to its question about a plan, are noted on your phone only and are not sent to Selver.
4What happens to a recording
It stays on your phone. The audio is not uploaded, to Selver or to anyone else.
Your phone works out who spoke when. It separates the voices, compares them with what it learned from your voice sample to find which one is you, and writes down what was said. It does this with speech models that run on the phone.
Selver does not try to identify anyone else. Everyone who is not you is treated as one other speaker, with no name and no voice sample.
“Only me”. When nobody else can be heard, because the others are in your headphones on another device or you are on your own, you can choose “Only me” before you start, and confirm that nobody else can be heard. Selver relies on your word for this. It cannot listen while your iPhone itself is on a call, because the call has the microphone.
Your phone then writes down only the stretches in which it hears your own voice. It silences the rest of the recording before it writes anything down, and once it has written down what you said, it keeps only that silenced copy. It tells voices apart by how they sound, so a voice it cannot tell from yours, or one speaking at the same moment as you, can still be written down as yours.
If it hears other voices for 20 seconds or more in all, and for a tenth or more of all the speech in the recording, it writes nothing down, deletes the recording at once and tells you so. Nothing of it is sent to Selver.
The words go to a language model. The transcript, which includes the other person’s words, the timings and the counts, and summaries of your earlier reflections, including what your most recent ones looked at most closely, are sent to Selver’s service and from there to Amazon Bedrock, a service of Amazon Web Services. There a language model made by Anthropic drafts the reflection. If you chose “Only me”, the transcript holds only what your phone wrote down as yours, and the only timings and counts that go with it are of your own speech. Selver tells the language model that only you were heard, and checks what it writes before anything is kept: a sentence it finds speaking of anyone else is left out.
Only your words are kept. The model is told to quote you and nobody else. Selver then checks what it wrote against the transcript and removes any quoted line that is not yours, before anything is stored. If you chose “Only me”, every line your phone wrote down counts as yours here, so a voice it could not tell from yours can be quoted as yours. The transcript itself is not stored by Selver once the reflection is made.
The walkthrough is spoken by your phone. Your phone makes it from the reflection and speaks it in Selver’s own voice, which a speech model on your phone makes as it speaks; until that model is on your phone, or if it cannot speak, your phone speaks in one of its own voices. Nothing Selver says aloud is sent anywhere, and no audio of it is kept, on your phone or anywhere else. If you have made a Personal Voice on your iPhone and allow Selver to use it, your phone can speak in that voice: the line a walkthrough suggests you try, and a line you kept when your phone has no clip of your own voice for it that it can play. That happens on your phone too.
You are told. A notification that the reflection is ready is sent through Firebase Cloud Messaging, a Google service. It carries no words from the conversation. Selver never puts a line of yours in a notification.
If Selver cannot reflect on a recording, for example because your phone cannot find your voice in it, it makes no reflection and tells you so in the app.
5The companies Selver uses
Selver uses other companies to run the service. Each receives what it needs for its part. None of them receives the audio of a conversation.
Amazon Web Services stores your account and your reflections, and runs Selver’s programs. Its Bedrock service runs the language model that drafts reflections, and receives the transcript for that.
Google delivers notifications through Firebase Cloud Messaging, and handles the sign-in if you choose to sign in with Google.
Apple distributes the app, under Apple’s own terms.
Hugging Face serves the speech models to your phone: the ones it listens with and the one it speaks with, the first time Selver needs them, and again if you remove them under You and fetch them again. It receives nothing about you and, as with any download, sees your phone’s network address.
Formspree receives your email address if you leave it on Selver’s website to be told when Selver is ready.
What each of these companies keeps, for how long, and whether it uses it to improve its own services is governed by its terms and by the settings of Selver’s account with it. A delete in Selver removes what Selver holds. It does not reach a copy that one of these companies may still hold.
Selver does not sell your data. Beyond the companies named here, Selver passes it on only where the law requires it.
6Where it is processed
The audio is processed on your phone. The text that is sent to Selver is processed in Amazon’s data centres in the United States, where Selver’s storage and programs are set up. If you live somewhere else, that text and your reflections leave your country.
7How long things are kept
On your phone, the recording of a conversation and the transcript made from it: until the reflection is ready, and then your phone deletes both, apart from a few seconds of your own voice for each moment the reflection points to. If Selver could not make the reflection, your phone deletes them when Selver says so. If your phone could not write the conversation down, the recording stays until you try again or discard it, unless you chose “Only me” and your phone heard someone else in it, as described above: then it is deleted at once. Your phone deletes a recording it has not finished with once it is seven days old, the next time Selver is opened and signed in.
On your phone, those few seconds of your own voice: until you delete the reflection they belong to, every reflection, your account, or the app. They are not included in your phone’s backups.
On your phone only, what you leave yourself before and after a conversation: until you delete it, remove the person it was about, delete every reflection or your account, or delete the app. It is not included in your phone’s backups.
On Selver’s service, the transcript: only while the reflection is being made.
Reflections, with their numbers, your quoted lines, what each looked at most closely, and Selver’s estimates: until you delete them.
A technical record of each reflection (when it was made, how much each speaker spoke, and, if it failed, why): until you delete that reflection, every reflection, or your account.
What your phone learned from your voice sample: on your phone, until you delete it, delete your account or delete the app. The recording of the reading itself: only until your phone has learned from it.
Your account, your settings, your notification token and the feedback you sent: until your account is deleted.
The record of agreement: kept after an account is deleted, as evidence that the agreement was given.
What you said about who could be heard, for a conversation you chose “Only me” for: until you delete that conversation’s reflection, every reflection, or your account.
On your phone only: the names and colours you give people and the time each recording started, until you remove them or delete the app; which of your lines you kept, with a copy of their words, until you take them out of Kept, delete their reflection, every reflection or your account, or delete the app; which line Selver suggested from each reflection, and your answer, until you delete that reflection, every reflection or your account, or delete the app; and a note of which of your kept lines came back to you and when, until you take that line out of Kept, delete the reflection it is about, every reflection or your account, or delete the app.
Also on your phone only: its record of each conversation it sent as text, until you delete that reflection, every reflection or your account on that phone, or delete the app; and its notes of when you opened Before for someone and when you last waved away a question on Today about seeing someone or leaving yourself a word, until you delete every reflection or your account, or delete the app. That record keeps its latest 5,000 entries. A note of when you opened Before is dropped sooner: once it is two days old, the next time Selver starts. Deleting every reflection, or your account, also removes the names and colours you gave people and the time each recording started.
On your phone only, what “Only me” notes: with a recording, for as long as your phone keeps the recording; with a reflection on its way, until you open it or dismiss it, delete it, delete every reflection or your account, or delete the app; and with a person’s name, for as long as the name is kept.
Also on your phone only, about a plan: if you choose a plan while Selver cannot keep it for you, which plan you chose and when, until Selver keeps it, you delete your account, or you delete the app; and, if you choose “Not now” when Selver asks about a plan on Today, that you did, and nothing else, until you delete your account or delete the app.
8What you can delete
One reflection. Open it and choose “Delete this reflection”. Selver removes that reflection now: its numbers, the lines it quotes, its move and its technical record, and your phone deletes the few seconds of your voice it kept for it. If you chose “Only me” for it, Selver also removes what it recorded of your saying that nobody else could be heard. Lines of yours that Selver had carried from your conversations onto the Today screen are removed too. Your phone forgets any line you kept from it. It also forgets its note of when that conversation was sent. What you left yourself after it is not deleted with it: your phone only forgets which conversation that was. Your overall numbers and trends keep what that conversation contributed, until you delete every reflection.
Every reflection. Under You, in “Privacy and your data”, choose “Delete every reflection”. Selver removes all your reflections, the summaries and trends built from them, and their technical records, and your phone deletes the few seconds of your voice it kept for them and forgets every line you kept. It also forgets its record of what it sent, and removes the names and colours you gave people and, with them, what you left yourself before and after conversations. Selver also removes everything it recorded of your saying that nobody else could be heard. If part of this fails, Selver says so and you can try again.
A line you kept. Open it in Kept and choose “Take it out of Kept”. It leaves Kept and stays in its reflection. Or leave it in Kept and choose “Stop bringing it back”, and Selver will not bring it back to you on Today.
Something you left yourself. Open it on the person’s page, under “Your notes to yourself”, and choose “Delete this note”. Your phone deletes it, with its recording if you spoke it. Removing a person also deletes what you left yourself about them.
What Selver knows of your voice. Under You, in “Your voice”, choose “Delete what you know of my voice”. Your phone removes it. Selver cannot reflect on new conversations from that phone until you read the passage again.
Your account. Under You, in “Account”, choose “Delete account”. Selver removes everything above that it holds, your settings, your notification address, the feedback you sent and your sign-in, and your phone removes what it learned of your voice and everything it kept of your conversations. What Selver removes includes what it recorded of your saying that nobody else could be heard. It also removes what you left yourself before and after them, and its record of what it sent. If a conversation is still being reflected on, Selver asks you to wait until it is ready.
What a delete does not reach. The record of agreement described above, which is kept; and anything one of the companies above has already received.
9Your voice sample
The sample is a recording of your own voice reading a short passage. Your phone works out from it a description of your voice, a list of numbers that is not a recording and cannot be played back, uses that to recognise your voice in the conversations you record, and deletes the recording of the reading. In some places the law treats such a description of a voice as biometric data.
Selver tells you what the sample is for before it records it, and keeps a record that you agreed.
It is used for that one purpose, on your phone. It is not sent to Selver, it is not sold, and it is not used to recognise you anywhere else.
What your phone learned from it is kept on your phone until you delete it, delete your account or delete the app.
10Other people in your conversations
A recording holds the voices and words of people who are not using Selver. Your phone writes their words down with yours, and Selver sends both to the language model, because a conversation cannot be followed from one side of it.
Their voices do not leave your phone, and Selver does not keep their words. The recording and the transcript are deleted from your phone when the reflection is ready, apart from a few seconds of your own voice for each moment, cut where your phone heard only you. What Selver stores after that quotes you and nobody else. What stays about the other person is when they spoke and for how long, which is how Selver draws the conversation. Selver’s own sentences can still describe, in Selver’s words, what the conversation was about.
When you choose “Only me”, your phone writes down only what it hears as your voice, and sends Selver’s service only that: nothing it heard of anyone else, not when they spoke and not for how long. Selver’s service checks what it is sent, and takes such a conversation only if every part of it is marked as yours. The reflection Selver stores for it says that it heard only you, and holds no count or drawing of anyone else’s part. Your phone tells voices apart by how they sound, so a voice it cannot tell from yours can still be written down, sent and quoted as yours.
Selver draws the other person and never quotes them. It does not keep a voice sample for them, does not try to work out who they are, and does not recognise them from one conversation to the next. A name you give them stays on your phone. Nothing they said can be kept: Kept holds your own lines only.
Because of this, you must tell everyone present before you record, every time. Selver asks you to confirm that you have before it starts. When you choose “Only me”, Selver asks you instead to confirm that nobody else can be heard, and relies on your word: say it only when it is true. If someone does not want to be recorded, do not record.
If you were recorded with Selver and want what was made from that recording removed, ask the person who recorded you to delete the reflection. You can also write to kendara2023@gmail.com.
11Who Selver is for
Selver is for people aged 18 and over.
12This website
Selver’s website sets no cookies and runs no analytics. Its pages load nothing from any other company: the fonts and pictures come from the site itself. Like any website, the server that sends you the pages sees your network address.
If you choose the light or the dark theme with the switch at the top, your browser remembers that choice in its own storage. It is not sent to us.
The one thing that leaves the site is the form for your email address. When you send it, your address goes to Formspree, which delivers it to us. It is used for one message: to tell you when Selver is ready to install.
13Until Selver is finished
Selver is unfinished and this policy will change as it is finished. No service can promise that what it holds will never be exposed. Think about that before you record a conversation that would do harm if it were.
Depending on where you live, the law may give you rights over your data, such as a copy of it or its correction. To use them, write to kendara2023@gmail.com.
14Builds for testing
Builds of Selver given out for testing before it is released, through TestFlight, also have a field test, for measuring how well Selver hears a conversation. A build for the App Store does not have it.
Recordings you put into the field test, or make in it, stay in its own folder on your phone, which the Files app shows, until you delete them there. Your phone keeps that folder out of its backups. It writes down what was said in them, by everyone in them, with when each person spoke, the time, memory, battery and heat that took, and the model of phone and version of iOS it ran on, and keeps that in the same folder. The field test learns a voice of its own from a recording you choose, and keeps it on your phone.
Nothing from the field test is sent to Selver. It leaves your phone only when you take it off: by sharing it from the field test or the Files app, or by copying it to a computer your phone is plugged into. It then goes where you send it.
In the field test, “Delete everything in the field test” deletes its recordings, everything written from them and the voice it learned. Copies you took off your phone are not reached by it.
15Changes to this policy
When this policy changes, the date and the version at the top change with it. If a change affects what happens to your recordings, Selver will tell you in the app.
16Contact
Write to kendara2023@gmail.com.