Privacy Policy

Last updated: October 12, 2026

This policy covers the Pairent app, which is in private beta, and the pairent.app website. It replaces an earlier policy that covered only the waitlist page.

Pairent is operated by Daniel Wilson, an individual developer working alone. There is no company yet: a limited liability company is being formed, and this page will be updated with its name, postal address and phone number when it exists. Until then the way to reach us is[email protected], and we answer it ourselves.

Pairent records a parent's day, either with a small clip-on recording device or with the iPhone microphone, transcribes what it hears at home, and turns it into private parenting insights for that parent. Because those recordings contain children's voices, this policy is written to meet the children's-privacy notice requirements of US law, and we treat everything the recorder captures as sensitive.

One thing to say plainly: this page has not yet been reviewed by a lawyer. We are engaging outside counsel and expect to revise it before Pairent reaches the App Store. We will not narrow your rights, or expand what we collect from your children, without telling you first and asking you again.

The short version

Recording. Your family's audio is captured only when you choose to record. Once it reaches our servers, the raw audio is deleted automatically within 48 hours, whether or not we managed to process it. A copy sits on your phone until it has uploaded and is then deleted about 48 hours after the session ends. If an upload fails for good, that recording stays on your phone rather than being thrown away silently, until you retry it or delete it from the upload-failures screen.

What survives. Anything you have not saved (moments, quotes, coaching audio, photos, daily stories) is deleted automatically after about 7 days. A one-line log of behavioral observations, with no audio, transcripts or quotes in it, is kept about 33 days so your monthly letter can cover a full month. Things you save, and your weekly and monthly reports, are kept until you delete them.

No selling, no ads, no training. We never sell your data and never use it for advertising. Your family's audio, transcripts and photos are not used to train our AI providers' models: they either do not train on customer data by default, or we have opted our account out in writing. The honest exceptions are named in "Who else touches it" below.

Who can see it. No other Pairent user can see your family's data. Every record and stored file is locked to your account. On our side, one person (the founder) can reach stored data when it is needed to fix something you reported, investigate an error, or keep the pipeline running, and to check the quality of what Pairent writes about your family. We do not read your family's content for any other reason. The service providers listed below process it on our behalf.

Your control. You can see, correct and delete what Pairent has learned about your family, delete individual recordings, moments and quotes, delete voiceprints and portraits in one step, turn photo use off entirely, and delete your account from inside the app.

1. What we collect

From recording, by recording device or by phone microphone.This is where children's personal information comes in. Audio captured while recording is on, which means household conversations; children's voices are the point of the product. Transcripts of that audio, with speaker labels and timings.

Voiceprints, which are biometric data. With your separate, explicit consent we compute a numerical speaker-recognition profile (an "embedding") for each family member you enroll, on our own servers, and store it so the app can tell who said what. Voiceprints are never sent to any outside voice-recognition service: the software that creates them runs on our own infrastructure, alongside the rest of your account. You can delete them in one step in the app, and withdrawing voiceprint consent deletes every voiceprint we hold for your family.

Worth being precise about a second step: to match a recording against your stored profiles, our pipeline briefly computes the same kind of profile for every distinct voice in that recording, including people who are not enrolled, such as a visiting relative or a friend's child. Those working profiles exist only for the seconds it takes to compare them and are never stored. What we keep is a numeric confidence score, not the profile.

Acoustic signals. Pairent also analyzes how the audio sounds, not only the words. It notes when a child was crying or laughing, and when a voice was raised relative to your own usual speaking range. This runs on our own servers. The resulting tags are short labels on the day's timeline and they outlive the transcript text, so your trends over time still work. To make the comparison we keep a small statistical profile of your usual speaking pitch and volume.

From you. Your email address; your sign-in method (a one-time email code, a password, or Sign in with Apple); the names and relationships of the people in your family; each child's birth month and year, which we use to keep their age and developmental stage current (we do not ask for the day); the developmental stage we derive from it; your timezone and delivery preferences; feedback you send us; and custom vocabulary words.

Photos from your camera roll, on by default. Unless you turn it off, Pairent looks at photos taken on your phone within a few minutes either side of a recorded moment, and uploads small thumbnails of those candidates so it can pick images for your daily story. You choose the scope in Settings, under Story photos: no photos at all, photos taken around recorded moments (the default), or your whole day. Thumbnails are re-encoded on your phone to strip location data before upload, and we never store or copy your photo library. Photos you add to a memory yourself are always included.

Watercolor portraits, if you turn them on. With separate consent, you pick one photo of a child and we paint a single stylized watercolor portrait from it. The photo you picked is held in memory for that one generation and is never written to our storage. The portrait is stored, along with coordinates describing where the face sits in it, and it becomes that child's standing illustration across the app.

Derived by Pairent. Parenting insights, coaching moments, kid quotes, daily, weekly and monthly summaries, and a small set of learned family facts and behavioral observations. Pairent builds a profile of your family: that profile is what makes the insights specific to you rather than generic. It is used only to generate your own insights and reports. It is never used to make a decision about you or your child that has a legal or similarly significant effect. You can see, correct and delete the learned facts on the "What Pairent remembers" screen, and deleting your account deletes the profile. Each night Pairent also keeps a short written record of the day: what the family did, what stood out, and any firsts. The learned family facts are written from the same day's recordings. You cannot browse the record in the app yet; it reaches you through your monthly letter and through Ask Pairent, a question box in the Memory tab that answers questions about your family's past (section 4). Section 3 says how long it is kept, and turning memory off or deleting your account deletes it.

Automatically. App version and device model; crash and error reports, from which we strip request bodies, message bodies, transcripts and credentials before they are sent; your push notification token; and the name and Bluetooth identifier of any recorder you pair, so the app can find it again (that stays on your phone and goes when you remove the device). The app also records a small log of usage events on our own servers: how you use the app, meaning which screens and features you open, how long you spend on them, how far you scroll or listen, the taps you make, the settings and choices you change, whether an error was shown, and the things you save, rate or remove. Each event holds an event name, the ids of the items you acted on, counts, durations, dates, yes-or-no flags and category labels, and, for events your phone records, the app version and build; each is tied to your account. A book shows up as which book you chose and how many; a custom principle, a custom vocabulary word or a voice enrollment only as a kind and a count, never as its text. Some events are recorded by the app on your phone, and some by our server when you make a change. Those events never contain recordings, transcripts, quotes or names, or the text of anything you type, including the questions you ask Pairent, your custom principles or your custom vocabulary words. They are kept for up to one year, deleted with your account, and included in your data export.

We do not collect precise location. We do check whether the IP address your phone connects from looks like you are in Illinois, where Pairent does not record; section 7 explains how. We do not use third-party advertising or analytics software in the app.

2. How we use it

To run the product: transcribe your recordings, tell enrolled speakers apart, generate insights grounded in published parenting frameworks, deliver your daily, weekly and monthly summaries, illustrate your daily story, keep the service reliable and secure (error diagnostics, with personal content stripped), and see which parts of the app are used so we can improve it (the usage events described in section 1).

We also use your family's content, at the smallest scale that answers the question, to check the quality of what Pairent writes: reviewing how Pairent updated what it remembers about your family, and, for accounts we have specifically enrolled in it, running a day through a second AI model to compare which produced better coaching. That is quality work on your own account's output. It is not training anyone's model on your data. If you want to know whether your account is enrolled in model comparison, or want it taken out, email us and we will do it.

We do not use your family's data for advertising, we do not sell it, and we do not permit our AI providers to train models on it.

3. How long we keep it

This section is our written retention policy. Where a number is approximate, it is because a sweep runs on a schedule rather than to the second.

Raw audio on our servers. Deleted automatically within 48 hours of upload, whether or not we processed it, and whether the processing run succeeded, failed, or never happened. A sweep runs every hour.

Raw audio on your phone. Deleted about 48 hours after the session closes, once it has safely reached us. The one exception: a recording that never manages to upload is kept on your phone rather than discarded, and it stays there until you retry or delete it from the upload-failures screen. We would rather not silently throw away a recording you might still want, but it does mean that particular file has no automatic clock on it.

Transcript text. Removed within 48 hours of our processing a day, whether processing succeeded or failed. What remains of the transcript is its timings and speaker labels, with no words, plus the acoustic tags described above. One honest caveat: the removal is triggered by a day's processing run reaching a finished state, so if our pipeline crashes mid-run, that day's transcript text can sit longer than 48 hours until the run is resolved. We are fixing that. In the meantime, if you want a specific day's transcript removed now, email us and we will do it.

Short transcript excerpts and clips attached to a moment or a quote. When Pairent flags a coaching moment or a kid quote, it saves about thirty seconds of the surrounding conversation with that item, verbatim, so the item still makes sense after the full transcript is gone, and it keeps a short audio clip so you can replay it. Those live exactly as long as the moment or quote does: about 7 days if you do not save it, and until you delete it if you do.

Voice-enrollment recordings. The short clip you record to teach Pairent a voice is kept only if you agreed to that when you enrolled. It is held alongside the voiceprint so we can rebuild the voiceprint if we improve our speaker-recognition model, without asking your family to record again. It is deleted whenever the voiceprint is: when you delete the voiceprint, withdraw voiceprint consent, remove that family member, or close your account.

Unsaved content. Moments, kid quotes, coaching audio and daily story text are deleted automatically after about 7 days. This is the "save it or lose it" rule. There is no longer a setting to keep daily-story text: since September 11, 2026 it is deleted on this schedule for everyone.

Photos in the running for your daily story. About 7 days, and up to about 9 days while a photo may still be needed for that week's recap. Then deleted automatically. A photo that is chosen as a highlight in a weekly or monthly report is copied and kept with that report until you delete the report.

Behavioral event log. One factual sentence per event, with no audio, transcripts or quotes, kept about 33 days so a monthly letter can cover a full month. This window is fixed: you can delete the whole history or your account, but you cannot shorten it.

Watercolor portraits. Kept until you replace the portrait, delete it, remove that family member, or withdraw portrait consent. Withdrawing consent deletes every portrait we hold. They are not on the automatic clock, because a portrait is the child's standing illustration across the app.

Saved items, weekly and monthly reports. Kept until you delete them or delete your account.

Learned family facts and coaching progress. On "Remember our family", kept until you delete them, turn memory off, or delete your account. On "Remember about a month", kept about 33 days. The learned facts are viewable and editable in the app.

The written day record. On "Remember our family", it is kept until you turn memory off or delete your account. On "Remember about a month", it is kept about 33 days. Choosing "Remember nothing" turns memory off and deletes it. A day's record that could not be checked is set aside, never shown or used, and deleted within 14 days (section 4).

Voiceprints. Kept while speaker identification is switched on for your account. They exist for one reason, telling your enrolled family members apart, and they are deleted as soon as you delete them, withdraw voiceprint consent, remove that family member, or close your account. We do not delete them on a timer.

Our own nightly record. Each night Pairent consolidates what it has learned about your family, and we keep a short internal record of what changed so we can check that it is working and fix it when it is not. It can quote the same sentences you see in the app. Only Pairent can see it, it is never shared, and it is deleted after 90 days.

Usage events. The log of which screens you open and when you play or share an insight is kept for up to one year, then deleted automatically. It goes with your account when you delete it.

Account data. Kept until you delete your account.

After you delete your account. We keep a dated record that the deletion was requested and completed, because children's privacy law expects us to be able to show that we honoured it. It contains no family content. We also keep a de-identified ledger of what our AI processing cost, with your account identifier removed, for accounting.

Our providers' own clocks. Anthropic and OpenAI, our two AI providers, each delete what we send them within about 30 days, and Anthropic keeps work we submit to it in batches for 29 days. OpenAI keeps files sent through its batch service until they are deleted, and Pairent deletes them as soon as it has collected the results. OpenAI also makes our images, on the same clock of about 30 days. Our text-to-speech provider keeps nothing at all: we send narration scripts under a zero-retention mode, described in section 4. Error logs and email-delivery logs age out within about 90 days. Two exceptions we cannot switch off and will not paper over: a provider may hold a copy longer while investigating misuse of its own service (at Anthropic that hold runs up to two years, and it keeps automated safety scores for up to seven), and a provider may be required by a court to preserve data. Those holds are the provider's, not ours, they are not used to train models, and they are the reason we do not claim your family's data is gone from everywhere in 48 hours. We do not claim "zero retention", because that would not be true.

4. Who else touches it

We share personal information only with the service providers that operate the product, and each is limited to doing that for us. We should be straight with you about the paperwork: data processing agreements are signed with some of these providers and not yet with all of them, and closing that gap is on our list before the beta widens. Where a provider's standard terms carry something we cannot switch off, it is named below.

Speech-to-text, which we run ourselves. Since October 1, 2026, Pairent turns your recordings into a transcript with speech-to-text software we run ourselves, rather than sending them to a speech-to-text company. The software runs on computers we rent by the hour in Europe from the two companies below. They supply the machine and we run the software on it; neither company transcribes anything for us. Each receives the raw audio of your recordings, including your children's voices, and a list of your family members' names to help the software spell them correctly. Both reach the rented machine over an encrypted connection. They are not encrypted on the machine's own disk, so they are kept there only while they are being transcribed, and we delete the machine and its disk when the work is done.

Verda (DataCrunch Oy, a Finnish company) is the one we use by default. Its data centres are in Finland, inside the European Union, and Verda says they run on 100% renewable energy. Verda's data processing terms apply automatically under its standard terms.

RunPod, a United States company, is our second choice for the same job, used when no suitable Verda machine is available. We rent its machines only in data centres inside the European Economic Area: our system picks from a fixed list and names the exact data centre every time. Today that list covers Romania, Sweden, Czechia, the Netherlands and France, which are in the European Union, and Iceland and Norway, which are in the European Economic Area but not in the European Union. RunPod's support staff can reach a rented machine from outside Europe to fix a problem; when they do, it is recorded in a log we can read afterwards, but we are not told at the time. RunPod treats the identity of the companies that operate those data centres as confidential, so we cannot name them. RunPod has signed a data processing agreement with us.

AssemblyAI (speech-to-text, in the United States) is now a last resort. Until October 1, 2026 it transcribed every recording. A recording now goes to AssemblyAI only if our own transcription fails on it, or if no rented machine can be had that night. When that happens, AssemblyAI receives the raw audio of that recording over an encrypted connection, with a list of your family members' names to help it spell them. Separately, for accounts we have enrolled in quality comparison, which today means only the founder's own, AssemblyAI transcribes each day's recordings a second time so we can compare its transcript with ours. Their account is opted out of model training, confirmed in writing on 2026-06-12; that opt-out is forward-looking, so we will not tell you it applies retroactively to audio sent before that date. They delete the uploaded audio automatically within 24 to 48 hours, and we delete the transcript from their servers as soon as we have retrieved it.

Anthropic and OpenAI (AI analysis and writing). Pairent uses two AI companies, Anthropic and OpenAI, to analyze the transcripts of your recordings and to write what you read and hear: the daily feedback on your family's coachable moments, kid quotes, the scripts for your coaching audio, your daily story, what Pairent remembers about your family, your monthly letter and the answers in Ask Pairent. Which of the two handles a given task can change over time, and either may handle any of them. Depending on the task, they receive the day's transcript with speakers labeled by name; family members' names and your children's ages; your children's quotes in their own words; what Pairent remembers about your family, including the day records, learned facts and reports; your choices and feedback, such as the parenting books and principles you pick and your ratings and notes on what Pairent writes; the questions you type into Ask Pairent; and photos: the ones Pairent is considering for your daily story and, if you turn on portraits, the photo you chose for a portrait. Neither receives any audio of your family. Under their standard terms for API customers, neither uses this data to train its AI models. Anthropic's data processing terms apply automatically under its standard commercial terms. Pairent has not signed a separate data processing agreement with OpenAI.

Some of that work sends a narrower set of data. Before a day's record or a learned fact is saved, it is sent to Anthropic or OpenAI, which checks it for things Pairent must not keep, such as a child's exact words or a medicine dose. The records and facts are short sentences about what your family did and what Pairent has learned, with your children's first names. If one breaks those rules it is rewritten or removed. A day's record that could not be checked is set aside, never shown or used, and deleted within 14 days. A fact that could not be checked is not saved. For this check, the AI company receives only the records and facts, never the transcript or any audio. To answer a question, Ask Pairent sends Anthropic or OpenAI your children's names and ages, the learned facts, a summary Pairent keeps of your family, the day records and the short daily notes that came before them, the last month of the behavior log, your reports, your saved moments and quotes, and your question. It never sends the full transcript or any audio. Saved quotes are your children's own words, and saved moments include a short quote of what was said. Once a month, Anthropic or OpenAI writes your family's monthly letter. To write it, Pairent sends it the past month's material: your children's names and ages; the month's day records, which can say that a child was ill or hurt but never give a medicine dose, a health reading or a diagnosis; the short daily notes and weekly recaps for any days the day records do not cover; the kid quotes featured in the weekly recaps, which are your children's own words; the captions of the month's keepsake photos, never the photos themselves; the month's changes to the favorites Pairent has learned about your children; and how the month's coaching suggestions went, with counts from the behavior log. That is one request per household each month, repeated only if a request fails, and it runs unless memory is turned off. It never sends the transcript or any audio. If you turn on portraits, the photo you chose is screened for safety before any portrait is painted.

xAI (the voice for coaching audio, since August 20, 2026) receives narration scripts, which include your children's first names and the substance of the day's moments, including things they said. It does not receive any audio of your family. xAI does not train on this data: no-training is a term of their enterprise contract rather than a setting we opted out of, though the same term makes it subject to what xAI discloses to us and to settings we control. xAI's standard data processing addendum applies automatically under those enterprise terms, and we have not separately signed one. We send under their Zero Data Retention mode, which our system verifies account-wide and enforces fail-closed before sending. Their speech endpoint returns no retention header, so we cannot verify each request individually; if the account-level check cannot be confirmed, the system refuses to send. Where the scripts are processed is expected to be the United States but is not guaranteed: every company xAI lists as processing for it is in the United States, apart from one support subsidiary in the United Kingdom, but xAI does not guarantee where its speech service processes a request.

OpenAI also paints story illustrations and watercolor portraits, and is the fallback voice for coaching audio (it was the voice until August 20, 2026). For that work it receives illustration prompts, which include your children's first names; if you turn on portraits, the one photo you chose for that child, used once for that generation; and, if we ever fall back to it for the voice, the narration script. Generated portraits are sent back as likeness references when illustrating a daily story. OpenAI does not train on data sent through its API.

Model comparison. For accounts we have specifically enrolled in quality comparison, the same day's transcript with family names and ages also goes to a second model from Anthropic or OpenAI, so we can compare which produced better coaching. Comparison stays limited to accounts we enroll deliberately. Email us and we will tell you whether yours is one, and take it out if you would rather.

Supabase (database and file storage, in the United States) holds all stored data. It travels to and from Supabase encrypted, and Supabase keeps it encrypted at rest. Supabase may keep aggregated and anonymized statistics about how its own platform is used; those contain none of your family's content.

Fly.io (backend compute, in the United States) processes data as it passes through the pipeline. During a nightly run, recordings are decoded to temporary files on the machine's disk and removed when the run finishes; nothing is kept there as a permanent store. Fly.io encrypts traffic to, from and between its machines, but it does not say whether a machine's own working disk is encrypted.

Sentry (error reporting, processed in the European Union) receives technical error and crash events. We strip request bodies, message bodies, transcripts and credentials before sending, and we do not send recordings. Error reports from the app carry a pseudonymous account identifier, a one-way hash of your account id, so we can find the errors one household hit; it carries no email address or name. We will not overstate this: the scrubber works from a list of known-sensitive fields, so a name or a short excerpt of generated text can occasionally survive inside a diagnostic log line. Sentry does not use identifying data to train models, and it derives non-identifying aggregate statistics from error events, which we cannot opt out of.

Expo (push notifications) receives your device's push token, a short generic message, and a temporary link, valid for 24 hours, that lets your phone fetch that day's coaching audio.

Brevo (account email) receives your email address to send sign-in codes.

Resend (product and operational email) receives your email address when we need to email you about your account, for example to tell you a data export is ready.

Cloudflare (website hosting) serves pairent.app, stores your email address if you join the waitlist, and counts page views at its edge (section 10).

Tally (beta application form) receives what you type into the beta application form, which includes your children's ages but never their names, and stores it in the European Union. Tally touches nothing else: no recordings, no transcripts, and nothing from inside the app.

Google holds a copy of the beta applications if we export them to a spreadsheet to score them. It no longer serves the typefaces on this website, which has served its own since September 16, 2026. One page that is built but switched off, the page for sharing a month of your family's story, still loads its fonts from Google; we will serve them ourselves before that page is ever switched on. Google touches nothing else: no recordings, no transcripts, and nothing from inside the app.

Apple receives and returns the identifier used for Sign in with Apple, and would handle any future subscription payment. If Pairent ever charges, Apple handles the payment and all we would receive is whether a subscription is active and when it renews. We never receive your card number or billing address.

Everyone in that list is a service provider we use to run Pairent. None of it is a hand-off of your child's information to someone else for their own purposes, and there is no way to run Pairent without them. We do not share your children's information with anyone outside this list, and we will not add a feature that does, such as sharing a weekly recap with relatives, without asking you separately and getting your consent for that specific sharing. Agreeing to Pairent processing your family's recordings does not commit you to any such sharing.

We would disclose personal information beyond this list only if required by law, or to protect someone's safety, and we will tell you unless we are legally barred from doing so. As of the date at the top of this page, we had never received such a demand. If that changes, we will update this line.

5. Children's privacy

Pairent is used by parents, but it collects personal information from children, their voices and what they say, so we operate under the US Children's Online Privacy Protection Act (COPPA). Everything in this section we apply to every child in your family profile, whatever their age.

Consent before collection. During the closed beta we obtain verifiable parental consent by a written consent form that the parent signs and returns to us electronically, which is one of the methods the COPPA Rule permits. You sign it in the app, before anything is recorded: you read it in full, type your name to sign it, and we email you a copy. Our record of each signature holds the name you typed, which document and version you signed and when, a fingerprint of the exact text so we can later show what you agreed to, the first names and birth years of the children on your account when you signed, and whether you opened the full text and on which screen, as the app reports it. When you sign a consent, we also record your IP address and your device's user-agent with it. Each other adult who lives with you signs their own consent, from a link we email them. Until every one of those signatures is in, the app will not record or take recordings off the recording device, and we will not process your family's recordings.

Accepting this policy and our terms. You accept this Privacy Policy and our Terms of Use when you create your account in the app, by continuing past a line that links to both. During the closed beta you also accept the beta agreement there, by ticking a box. These are not signed with your name, as your children's consent is. We record which version of each you accepted and when, with your IP address and your device's user-agent, and we email you a copy.

At public launch we will verify parental consent before any recording is processed, using one of the methods the COPPA Rule permits. We will publish the method on this page before we switch to it. Whatever we choose, verification will happen before collection starts, not after a trial period.

Your rights as a parent. At any time you can see everything we hold about your child, have it deleted, and tell us to stop collecting. Specifically: the learned facts are on the "What Pairent remembers" screen, where you can view, edit and delete them; individual recordings, moments and quotes can be deleted in the app; voiceprints and portraits are deleted in one step, and withdrawing either consent deletes what it produced; setting Story photos to none stops photo collection; Settings, then Data retention, then Delete account and all data erases your account immediately; and for a full copy of everything we hold, or to have one child's information deleted while another child's is kept, email[email protected] and we will do it by hand within 30 days. Your child loses nothing by any of this: your child never uses Pairent, and there is no child-facing part of the service to lose access to. If you tell us to stop collecting, we stop, and the parts of Pairent that depend on recording simply stop working for you.

Minimal collection. We collect no more than the product needs, and the retention rules above delete most of it automatically on a short clock.

No monetization of children's data. No advertising, no sale, no AI training, and no disclosure beyond the service providers named above.

For anything in this section, write to[email protected].

6. Security

Beyond your phone, your family's data travels over encrypted connections, including to the computers we rent for speech-to-text. Where we store it, at Supabase, it is also encrypted at rest. It is not encrypted at rest everywhere it is processed, and we will not say it is: the computers we rent for speech-to-text do not encrypt the audio on their own disks, and Fly.io, which runs our backend, does not say whether its machines' working disks are encrypted. In both places audio stays only while it is being processed, and is then deleted (section 4). Access is isolated per account at the database level, so no other Pairent user can reach your family's records or files. We maintain a written information security program.

One honest limitation you should know about: the short-range Bluetooth link between the recording device and your phone is not encrypted. Someone with radio equipment within Bluetooth range of you could in principle capture audio as it crosses that link, both while you are recording and while the recorder is transferring earlier recordings to your phone. The connection never leaves your immediate vicinity, everything beyond the phone travels encrypted, and link encryption is on our roadmap. Recording with the iPhone microphone instead of the device does not involve that link at all.

If a security incident affects your family's information and creates a real risk to you, we will tell you without undue delay, in plain terms: what happened, what was involved, what we have done, and what you should do. We will also notify whichever authority the law requires, within the time the law sets.

If you think you have found a security problem in Pairent, please email[email protected]. We will read it, and we will not pursue anyone who reports a problem to us in good faith.

7. Where data lives

Your recordings are transcribed in Europe. The computers our speech-to-text runs on are in Finland, inside the European Union, by default, and always inside the European Economic Area when we use RunPod (section 4). A recording goes to AssemblyAI in the United States instead only if our own transcription fails on it or no rented machine can be had that night, and, for accounts enrolled in quality comparison, AssemblyAI also makes a second transcript of each day there.

Apart from transcription, your family's data is stored and processed in the United States: by our storage provider, by the servers that run our backend, and by the AI companies that read your transcripts and write what you read and hear. There are two kinds of exception. Three services are in the European Union: error reporting, the email service that sends your sign-in codes, and Tally, which stores beta applications (sections 4 and 10). And a few cannot be pinned to one country, such as xAI's speech service (section 4); the Subprocessors page names them and says why. Pairent is built for families in the United States. If you use Pairent from elsewhere, your family's data is transferred to and processed in the United States.

We do not currently offer Pairent to residents of Illinois, because of that state's biometric privacy law and our use of voiceprints, and Pairent should not record there. The beta application form asks which state you live in. We also check the IP address of your phone's internet connection when the app talks to our servers (when you open it, while it records with your phone's microphone, and when you sign a consent) to see whether it looks like you are in Illinois, so that we don't record where we shouldn't. We don't use your phone's location services for this, and we don't send your IP address to anyone else to look it up: we use a copy of IP-to-location data from IP Geolocation by DB-IP on our own servers, and the only answer we work out is whether the address looks like Illinois. We don't keep the IP address for this check. If it looks like Illinois, we keep only the day it happened, for 90 days, and we may contact you; the app may also pause recording, hold that day's recordings and ask whether you are in Illinois, and tapping "I'm not in Illinois" resumes recording for the rest of the day. IP addresses are often placed in the wrong state, especially near state lines, on mobile networks and on a VPN, so this is a safeguard rather than a guarantee, and we still rely on what you tell us. The Terms of Use explain what we are asking of you.

Whether European or UK data protection law also applies to Pairent, and what it would require of us, is a question we have put to counsel and have not yet resolved. We are not going to guess at the answer on this page. If the answer changes what we owe you, we will say so here.

8. Your choices and rights

See and correct. The "What Pairent remembers" screen shows the facts Pairent has learned about your family, and lets you edit or delete any of them.

Delete. Delete individual recordings, moments and quotes in the app. Delete voiceprints in one step, or withdraw voiceprint consent, which deletes all of them. Delete a portrait, or withdraw portrait consent, which deletes all of them. Delete your account and its data from Settings, then Data retention, then Delete account and all data: you type your email address to confirm, the erase starts immediately, and it cannot be undone.

One caveat we would rather tell you than have you discover. If you have turned on watercolor portraits, delete the portraits (or withdraw portrait consent) before you delete your account. Account deletion removes your records, recordings, transcripts, insights, voiceprints, enrollment clips, photos, illustrations and exports, but it does not currently remove the stored portrait image files, and once the account record is gone those files are no longer reachable from anywhere in the product. We are fixing this so it happens automatically. In the meantime, if you have already deleted your account, email us and we will find and remove anything left behind.

Stop further collection. Stop recording; set Story photos to none; withdraw voiceprint consent; or delete your account. Any of these takes effect at once.

Get a copy. Email[email protected] and we will send you a complete copy of what we hold within 30 days. A self-serve download is built but not yet switched on; when it is, this page will say so.

Settings. Notification preferences and the Story photos scope are both in the app.

There is no charge for any of this, and you do not have to explain why.

9. If you were recorded but do not use Pairent

Pairent recordings capture whoever is in the room. If you believe you were recorded, write to[email protected]. We will tell you what we hold and you can ask us to delete it. We do not identify anyone by voice unless the account holder enrolled them with their consent, unenrolled voices are labeled generically, and raw audio is deleted from our servers within 48 hours in any case.

10. The Pairent website

The waitlist. If you join the waitlist at pairent.app, we collect your email address and nothing else: not your name, not your location, not your device information, and not the campaign tag that brought you here. It is stored in our own database, run for us byCloudflare, our hosting provider, described in theirprivacy policy. We use it to tell you when Pairent is ready to try, and for nothing else. We will not sell, share or rent it.

Applying to the beta. The application form asks for a good deal more than the waitlist does, so here is the whole list: your first name and email address, the US state you live in and your time zone, which phone you use, which age bands the children living with you fall into, how many adults live in your home, whether every one of them is willing to sign the consent form, whether a paid caregiver is in your home on most days, whether you would wear the recording device or would rather use the phone microphone, how comfortable you are with recording at home, a few sentences about something you changed in your parenting in the last year, which parenting books or podcasts or creators you follow, whether anything big is coming up for your family this autumn (a move, a long trip, a new baby), which moments you would most want feedback about, what you already pay for parenting help, what you make of the price, why you applied, when you are free for a call, how you heard about us, whether you know Daniel, whether you work professionally in parenting, and whether you wear a fitness or sleep tracker on most days and which one. Three further questions, all optional and all skippable, ask your role in the family, your age band and your household income.

We ask for children's ages in bands, never their names and never their birthdays. Applying is voluntary, you can close the form at any point before submitting it, and you can leave the optional questions blank without affecting whether you are chosen.

The form is built with Tally, a form service run by Tally BV in Ghent, Belgium, which processes and stores the answers for us on servers in the European Union underits privacy policyand a data processing agreement. Your application is therefore the one part of Pairent that lives in Europe rather than the United States; if we copy the answers into a spreadsheet to score them, that copy sits with Google underits privacy policy. The form also records which link brought you to it, for example that you arrived from Reddit rather than from an email, so we can tell which ways of reaching parents worked. We use the answers for two things. The first is deciding who fits the cohort and how to balance the groups. The second is understanding, across everyone who applied, what parents want from something like Pairent; that use is reported as totals and percentages only, with no names and nothing that identifies a family, including when we share it with advisors and investors. Applying is not consent to be recorded: nothing is recorded until every adult in your home has signed the separate consent form.

If we do not offer you a place, we delete your application within90 days of the beta ending. If you do join, your application becomes part of your household's beta records and is covered by the consent form you sign. Either way, you can ask us what we hold, correct it, or have it deleted sooner by writing to[email protected].

What this site knows about your visit. The site sets no cookies and runs no tracking scripts, tracking pixels or advertising software in your browser. We do count visits, on our server rather than in your browser: when a page is served we add one to a running tally of how many times that page was viewed that day, which site linked the visitor here, and which campaign tag the link carried. The tally holds no address, no device information and no identifier of any kind, so two people arriving from the same link on the same day are counted twice and cannot be told apart afterwards, by us or by anyone else who reads it. Our host, Cloudflare, separately keeps request logs of its own, as every host does. Links we post while recruiting for the beta also carry a short campaign tag in the address bar, so a visit arriving from Reddit is distinguishable from one arriving from an email. We use this to work out which ways of reaching parents actually worked. It is counting, not profiling: no cookies, no following you around other sites, and nothing tied to your name unless you go on to fill in a form. Since September 16, 2026 the site has served its own typefaces, so loading a page no longer sends your IP address to Google Fonts. One page that is built but switched off, the page for sharing a month of your family's story, still loads its fonts from Google; we will serve them ourselves before that page is ever switched on.

You can ask us to delete your email address at any time by writing to[email protected]. We will remove it within 7 days.

11. Changes to this policy

We will post changes here with a new effective date, and list them under "What has changed" below. We will email you about material changes, at the address on your account, before they take effect. Where a change expands what we collect from your children or how we use it, we will ask you to agree again before it applies to your family, and we will not process your recordings under the new terms until you do.

What has changed

October 12, 2026. Section 7 now says that Pairent checks whether the IP address your phone connects from looks like you are in Illinois, where Pairent does not record, and section 1 points to it. Until now we relied only on what you told us. The check uses a copy of DB-IP's IP-to-location data on our own servers, sends your IP address to no one, works out nothing but "Illinois or not", and keeps no IP address for it: only the day of an Illinois match, for 90 days. If it matches, we may contact you, and the app may pause recording until you tell it you are not in Illinois. We do not use your phone's location services, and nothing else we collect, or do with it, has changed.

October 4, 2026. Section 5 now says how you accept this policy, the Terms of Use and the beta agreement: when you create your account in the app, by continuing past a line that links to them and ticking a box for the beta agreement, rather than by typing your name. The consent for your children is unchanged: you still read it in full and sign it with your typed name. This page now calls our terms the Terms of Use, the name the app and the documents you agree to have always used. Nothing we collect, and nothing we do with it, has changed.

October 3, 2026. Section 4 now says how your recordings are transcribed. Since October 1, 2026, Pairent does it with speech-to-text software we run ourselves, on computers rented by the hour in Europe: from Verda, in Finland, by default, and fromRunPod, in a data centre inside the European Economic Area, when no suitable Verda machine is available. Both companies now receive the raw audio of your recordings, including your children's voices, and your family members' names, for as long as it takes to transcribe them. AssemblyAI, which transcribed every recording until then, is now a last resort: it receives a recording only if our own transcription fails on it or no rented machine can be had that night, and, for accounts enrolled in quality comparison, it makes a second transcript of each day. Deepgram is no longer named: it is not in use, and we would add it back to this page before sending it anything. Section 7 now says your recordings are transcribed in Europe, and it names the email service that sends sign-in codes, which is in the European Union and which that section had left out. Section 6 no longer says that all data is encrypted at rest. Your data travels encrypted and our storage provider keeps it encrypted at rest, but the computers we rent for speech-to-text do not encrypt the audio on their own disks, and Fly.io, which runs our backend, does not say whether its machines' working disks are encrypted; sections 4 and 6 now say so. Section 5 now says how consent is taken: in the app, before anything is recorded, with each other adult in your home signing from a link we email them, and what our record of each signature holds, including your IP address and your device's user-agent: the link we email to other adults has recorded both since it launched on August 5, 2026, and signing in the app records them too. It said the app did not collect consent, which stopped being true when in-app signing shipped on August 5, 2026. Sections 4 and 10 now say the website has served its own typefaces since September 16, 2026, so visiting it no longer sends your IP address to Google Fonts, and that one switched-off page still loads them. Apart from the IP address and user-agent now recorded when you sign in the app, none of this changes what Pairent collects from your family or how long Pairent keeps it: section 5 otherwise describes a signature record we have kept since in-app signing began.

September 29, 2026. Two corrections. Nothing about what Pairent collects, what it sends, or how long it keeps anything has changed. First, Section 4's xAI paragraph said that no-training was unconditional and that Pairent had not executed a data processing agreement with xAI. Both read as more or less than the truth: the no-training term is subject to what xAI discloses to us and to settings we control, and xAI's standard data processing addendum applies automatically under its enterprise terms, so saying no agreement had been executed read as if none applied. The paragraph now says what does apply, and adds that xAI does not guarantee where its speech service processes a request. The subprocessors page was corrected the same way on September 26, 2026. Second, Sections 3 and 8 described a setting for keeping daily-story text. That setting was retired on September 11, 2026, and daily-story text is now deleted on the ordinary schedule for everyone. The Story photos setting is a different control and is unchanged.

September 24, 2026. Section 4 now names Anthropic and OpenAI together as the two AI companies that analyze your transcripts and write what you read and hear, instead of assigning each task to one of them. Which of the two handles a given task can change over time, and either may handle any of them, so this page no longer says which one does what. It says what they receive between them, by kind of data, and that neither receives any audio of your family. The practical change is that OpenAI, not only Anthropic, may now receive your family's transcripts. Section 3 now gives both companies' retention clocks. Moonshot AI is no longer named: no comparison model has used it since August 2026, and using it again would take a change to Pairent's code, not a setting. What Pairent collects from your family, and how long Pairent keeps it, have not changed.

September 22, 2026. Section 4 now says OpenAI writes your monthly letter, which Anthropic wrote before, and lists what Pairent sends OpenAI to write it, once a month: your children's names and ages, the month's day records, the short daily notes and weekly recaps for any days the records do not cover, the kid quotes featured in the weekly recaps, the captions of the month's keepsake photos, the month's changes to your children's favorites, and how the month's coaching went. It never sends the transcript or any audio, but the quotes are your children's own words.

The same revision also changes Section 1, which now describes the app's log of usage events by the kind of thing it records rather than by listing each one: which screens and features you open, how long you spend on them, how far you scroll or listen, the settings and choices you change, whether an error was shown, and the things you save, rate or remove. It says that which book you chose is recorded, that a custom principle, a custom vocabulary word or a voice enrollment is recorded only as a kind and a count, that some events come from our server rather than from your phone, and that the log never holds the text of anything you type, including the questions you ask Pairent. The beta build records more kinds of usage event than the old sentence listed, such as time spent and how far you listened, so naming each one would go out of date as soon as the app changed. What the log holds is still ids, counts, durations, dates, flags and labels, and it still never holds recordings, transcripts, quotes or names.

September 21, 2026. Section 1 now says Pairent keeps a written record of each day, that the learned family facts are written from the same day's recordings, and that the record reaches you through your monthly letter and through Ask Pairent, which answers questions about your family's past. Section 3 now says how long the day record is kept, and how long the learned facts and coaching progress are kept on each memory setting. Section 4 now says OpenAI checks those facts as well as the records before they are saved, what happens when it cannot check one, and that Ask Pairent sends your family's memory to OpenAI to answer a question, listing what that includes: never the full transcript or any audio, but saved quotes and moments include your children's own words. Sections 1 and 4 no longer limit this to the founder's own household: it applies to every household.

September 17, 2026. Section 1 now says that, for accounts where it is switched on (today only the founder's own household), Pairent keeps a short written record of each day. Section 4 now says OpenAI checks those records before they are saved, for children's exact words and health details the record must not keep. OpenAI receives the records, never the transcript or any audio.

September 2, 2026. Section 10 now says the application form asks whether you wear a fitness or sleep tracker on most days, and which one. It is optional to have one: no answer to that question affects whether you are chosen. The same section no longer lists the exact age bands the form offers. That list changed as the form changed, and what matters is unchanged and still stated: we ask for children's ages in bands, never their names and never their birthdays.

August 20, 2026. The voice that reads your daily coaching audio changed from OpenAI to xAI. Section 4 names xAI, says what the narration script contains, and states the terms we send it under: no-training is unconditional under xAI's enterprise terms rather than a setting we opted out of, and scripts go under Zero Data Retention, which our system re-checks and refuses to send without. OpenAI's entry in the same section narrows to story illustrations, watercolor portraits, and standing by as the fallback voice. Section 3 separates the two providers' retention clocks accordingly. This changed which company reads the script aloud. It did not change what we collect from your family, what the script contains, or how long we keep anything.

August 18, 2026. The website moved from Netlify toCloudflare. Sections 4 and 10 name the new host. The waitlist form is no longer handled by Netlify: waitlist email addresses are now stored in our own database, run for us by Cloudflare, and the addresses already on the list were copied there. The old host's copy is deleted when we shut that account down, within 30 days. We also stopped collecting three things Netlify's form handling had been recording alongside your address: your IP address, your browser's user-agent string, and the page you came from. None of them are in the new database. Page-view counting moved as well: rather than reading our host's dashboards we now keep our own tally, in that same database, of how many times each page was viewed each day and which link brought the visit. It runs on our server and not in your browser, so there are still no cookies and no tracking scripts, and the tally holds no address, no device information and no identifier, which means it can count visits but cannot tell two visitors apart. Section 10 also now says out loud that the campaign tag in a recruiting link is never stored against your email address.

August 17, 2026. Section 10 now covers the beta application form: what it asks, that it asks children's ages and never their names, that Tally processes and stores the answers in the European Union, that the form records which link brought you to it, that aggregate results are shared with advisors and investors with nothing identifying in them, and that applications we do not seat are deleted within 90 days of the beta ending. The same section replaces the old "no analytics" line: our host now counts page views and referrers from its server logs, and recruiting links carry a campaign tag. Still no cookies, no tracking scripts, and no advertising software. Section 4 adds Tally and names what Google does. Section 7 now lists beta applications as the second thing processed outside the United States, and corrects a line that said Pairent collects no state information anywhere: the application form asks for it.

August 16, 2026. Section 1 now discloses the small first-party log of usage events the app keeps (which screens you open, when you play or share an insight, app version and build), tied to your account and never containing recordings, transcripts, quotes or names; section 2 names what it is for and section 3 gives its one-year retention. Section 4 now says that error reports sent from the app to Sentry carry a pseudonymous account identifier, a one-way hash of your account id, with no email or name. We still use no third-party advertising or analytics software in the app.

July 30, 2026. First policy covering the Pairent app. Replaces the earlier policy, which covered only the waitlist website; the waitlist disclosure is preserved in section 10.

Contact

Pairent is operated by Daniel Wilson, an individual developer. There is no company yet; a limited liability company is being formed and this page will carry its name, postal address and phone number when it does. Write to [email protected] for anything on this page, including a request about your child's information.

See also the Subprocessors page, which names every company that touches your family's data and what each one receives, and the Terms of Use.