Peithra Privacy Policy
How Peithra, a free early-beta debate platform, handles the personal data of people who sign in and debate.
Last updated: November 13, 2026
About this policy
Peithra is a browser-based platform for structured 1-on-1 debates. You sign in, optionally enter matchmaking, get paired with another user, receive a topic, and debate in a private room using text chat and real-time voice.
Peithra is free to play and in early development (beta). There are no ads. Peithra may offer a way to make a voluntary payment to support the project; where it does, paying is optional and buys no advantage in any debate. This policy explains what personal data we collect, why we collect it, who we share it with, and the choices you have.
Effective date: November 13, 2026.
The short version
The essentials at a glance. The sections below carry the full detail.
- Ranked matches are recorded from your own microphone so the judge can score what you said. Agreeing to that is part of the Terms you accept when you create your account, because a judged debate cannot be delivered without it. Custom and social rooms are not recorded at all.
- Your recorded turns are transcribed by Deepgram, summarised by DeepInfra, and the debate is judged by Anthropic's Claude. The judge reads only your spoken words, never room chat.
- After a debate, both debaters can open its replay. For debates played from October 1, 2026, it includes a transcript of what each of you said.
- We never sell your data. No ads, no advertising cookies.
- Using debate content to improve the AI judge is optional and needs explicit consent from BOTH debaters. Text only, never audio. You can turn it off anytime.
- You can download your data or permanently delete your account yourself, at any time.
- Sometimes the law makes us keep, or a legal claim needs, a sealed copy of some of an account's data, even after the account is deleted: when we report suspected sexual abuse of a child, when the police or a court requires it, or for a legal claim. It can include what other people said to that account. It is locked away, used for nothing else, and deleted when that need ends.
How you sign in
You sign in with Google (Google OAuth, using Auth.js / NextAuth). There is no password for you to manage. Where Discord sign-in is enabled, you can also sign in with Discord.
From your sign-in provider we receive and store the following, so we can create and identify your account:
- Your name
- Your email address
- Your profile image
Keeping you signed in
To keep you logged in, we store standard sign-in records:
- An account record, including OAuth tokens issued by your sign-in provider
- Session records (a session token and its expiry)
- A session cookie set in your browser so you stay signed in
What else we store
Alongside your sign-in details, we keep information tied to your profile and your debates so the platform can run matchmaking, track results, and show debate history:
- An optional username (a display handle) if you choose one, and your date of birth, which is used to enforce age rules and, without being shown to anyone, to count minors in our record of any data breach. It is also kept in a sealed copy after a report of suspected child sexual abuse (see below)
- The country you chose, if you chose one. It is used to offer prompts about that country's politics, and it is kept in a sealed copy after a report of suspected child sexual abuse, or when a police request or an order names it (see below)
- Profile and gameplay stats: your role, an ELO skill rating, wins/losses/draws, total debates, and account created/updated timestamps
- Text-chat messages you send in a debate room: the message text, the room, a timestamp, and a snapshot of your display name, role, and rating at the moment you sent it
- Competitive match records: that a match happened, who was in it, the assigned topic, which side you argued, the match lifecycle, and timestamps
- Audio recordings of your own microphone during your recorded turns, and the text transcripts produced from them. The audio is deleted after 30 days, and the transcripts stay with your debate record until you delete your account
- Web server logs: your IP address, the page or file your browser asked for, your browser type and the time. They help us keep the service running and secure, and they are deleted after about two weeks, unless a sealed copy must keep some of them (see below)
- Crash reports, if a page fails in your browser: the type of error, where in our code it happened, which kind of page it was (for example a debate room, but not which one) and your browser type. They are sent without your sign-in cookie and never include anything you typed
Voice in debate rooms
Voice in a debate room is real-time between the two debaters. It is sent directly between you where your networks allow it. Where a direct connection is not possible, the audio is relayed through our relay provider (Cloudflare) to reach the other debater.
Ranked matches are recorded. Your own microphone is recorded during your turns so the judge can read a transcript of what you said. Agreeing to this is part of the Terms you accept when you create your account: judging is voice-only, so a recorded turn is how a judged debate is delivered, and we rely on the contract between us rather than on consent for it. You are told this before your account is created, and if you would rather not be recorded you can decline there, in which case we delete the sign-in details we already hold.
Custom and social rooms are not recorded. Recording never starts automatically anywhere: it starts from the visible recording control, and a recording indicator stays visible the whole time.
You are only ever recorded from your own microphone, and you never record anyone else. A recording indicator is visible while recording.
Audio recordings are stored on our own server and deleted after 30 days. Their transcripts stay with your debate record, and are erased when you delete your account.
Transcription and AI judging
Debates with recorded turns are transcribed and judged automatically:
- Your recorded voice is sent to Deepgram to be converted into text.
- The resulting text is sent to DeepInfra, which summarises each turn into the argument map you see on the canvas.
- At the end of the debate the transcript is sent to Anthropic (Claude), which produces the verdict. In ranked matches that verdict changes your rating automatically. Casual debates with recorded voice are judged too, but never affect anyone's rating.
What the AI judge can and cannot see
The judge scores the voice transcript only. It never reads room chat, and a debate with no recorded voice is refused, not guessed at.
Helping improve the AI judge
Peithra's AI judge is being developed, and debate content can help train it. This is strictly optional. A debate is only ever used for training when ALL of the following are true:
- Both debaters have opted in to judge training (Settings, Privacy & safety, 'Help improve the judge').
- Both debaters were adults (18 or over) when the debate ended.
- The debate was voice-sourced, meaning it was judged from real recorded speech.
What training uses, and what it never uses
When a debate is used, only text is used: the spoken transcript and the judge's verdict. Audio recordings are never used for training. We never perform voice recognition or speaker identification.
Training data is never sold or shared with third parties. You can withdraw your consent at any time in Settings, which stops any future use of your debates. Deleting your account makes your debates ineligible for training use and removes your personal data as described below.
This training consent is a genuine choice, and it is the only one of its kind here. It is not the same thing as the recording described above: recording a ranked match is a term of using Peithra, because a judged debate cannot happen without it, while letting an already-judged debate help improve the judge is optional, is never required to play, and can be turned off at any time.
Where your data lives and who we share it with
Your data is stored in our own PostgreSQL database on our own server / VPS.
We do not sell your data, and we do not use advertising cookies. We share data only as follows:
- Google, and Discord where you sign in with it, for sign-in
- Deepgram, for speech-to-text (receives your recorded voice)
- DeepInfra, for argument summaries (receives the text of your speech)
- Anthropic, for judging (receives the text of your speech)
- Cloudflare, for relaying live voice when a direct connection is not possible, for optional backup storage of recordings, and for storing our database backups and the sealed copies described below, both encrypted so that Cloudflare cannot read them
- Stripe, for processing any payment you choose to make (receives your payment details directly; we never see or store your card number), and for its fraud checks while a payment form is open (see Cookies)
- Hostinger, which hosts our email: it carries the messages we send you and the ones you send us
- Our hosting and infrastructure providers
- Others, only if we are legally required to disclose it
International transfers
The providers above process your data in the United States, which may be outside your home country. Where your data is transferred, we rely on each provider's own data-processing terms, and we are putting standard contractual clauses in place with counsel.
What other users can see
Some of your information is visible to other users:
- Your display name is visible to people you debate
- Your rank tier is shown on your public profile once you have completed your placement matches
- Your text-chat messages are visible to others in the same room
- Your live voice is heard by the other participant(s) in the room
- In a ranked debate, live captions of what you say are shown to the other debater and to anyone watching, and each of your turns is summarised into short argument cards that they can also see
- After a debate, both debaters can open its replay. It shows the argument cards and the judge's verdict, and for debates played from October 1, 2026, a transcript of what each of you said during your turns
Cookies
When you open a payment form, Stripe's payment script loads. Until you close or reload the tab (moving to another Peithra page does not stop it), it sends Stripe details of your device and how you use our pages, and sets cookies, to detect fraud; it can also load hCaptcha, a check against automated payments. Stripe does not use these for advertising. Apart from that, we use only the essential sign-in/session cookie, which keeps you logged in, and no advertising cookies.
Children
Peithra is not directed to children under 13, and we refuse sign-ups under 13. Content from under-18s is never used for judge training, which is limited to debates between adults as described above.
Your choices and deletion
You control your own data. From Settings → Account you can download a copy of your data ('Download my data') or permanently delete your account at any time. No request or email needed. What our backups keep, and what restoring one can bring back, is set out below. So is when we keep a sealed copy of an account's data.
If your account is deleted, your profile is removed, and what you said in your debates, the argument cards made from it and the judge's written notes on those debates are removed too; the scores and results stay. Some debate records and messages you sent may be kept so debate history stays intact.
Our database backups keep a copy for up to seven weeks, after which it is gone from them too. If we keep a sealed copy of an account that has been deleted, we take it from a backup before that backup is gone.
We keep a note of each deletion request, cancellation, deactivation, reactivation and deletion, and of each time a moderator stops an account or lifts that stop. It holds an internal number, the kind, the date and who acted: you, a moderator, or Peithra. We keep it while a deletion request is pending and for about 12 weeks after, so that restoring a backup does not bring back a deleted account or undo those choices, so that Download my data can show them, and so that an account is deleted only when its request is on record.
When we restore a backup, we keep your blocks, privacy settings you made more private, friends you removed, room bans and mutes, a moderator's removal of a picture, reports, and any withdrawal of consent to judge training, except a change involving an account the incident itself deleted and the backup brought back. Other changes made after the backup can be lost, such as new friends, posts and debates, and settings you made more public. Restored debates may be sent to our providers again to be transcribed, summarised or judged. We email every account we still have an address for.
If we lose the server or its database, or it is compromised, we restore an earlier copy: normally from the day before, and up to 35 days old after a compromise. Every change since that copy is lost, including the ones listed above. An account deleted since then comes back. If it had asked to be deleted before the copy was made, we delete it again within about 14 days, and until then the people it debated, their friends and the members of social rooms it posted in can see what it left there. If it asked after the copy was made, we pause it where a later copy shows that, and otherwise it is back as it was until it asks again. We email it either way. After a compromise we keep the affected server's disk as evidence until the investigation ends, six months at most, and then destroy it.
Records of our moderation of what you posted, and the reports you filed, are kept for abuse review after your account is deleted; their words are removed once 180 days have passed since a moderator reviewed them; until a moderator reviews them, they are kept.
If you have ever made a payment, a record of it is also kept, holding the amount, the currency, the date, whether and when it was refunded, whether a dispute holds it, and the processor's reference, but not your name. We keep it because financial records have to be retained. Stripe keeps its own record of the payment under its own policy.
When we keep a sealed copy
Sometimes the law makes us keep a copy of some of an account's data, or a legal claim needs one, even after the account is deleted. We call it a sealed copy. It is encrypted and locked away. We use it for nothing else, we share it only with those the law allows, and we delete it when the need ends.
A sealed copy about someone else can include what you said to them or about them, like a message in a debate or a report you made, and it can be given to the police, a court, or the United States organisation that receives reports of child abuse.
If the account has already been deleted, we take the copy from our backups. We keep a sealed copy only in these cases:
- When we report suspected sexual abuse of a child, as United States law requires us to. While a report is being prepared we keep a copy for up to 16 days, and if no report is made, that copy is deleted.
- When the police or a court requires us to keep data, under a law that binds us.
- When we need the data to establish, exercise or defend a legal claim.
What a sealed copy can hold
A sealed copy holds only what its case needs. It never holds the side a player asked to argue or the subjects they turned off. It can hold some data about other people: the other debater and anyone watching in a debate the case concerns, the account's friends, accounts it blocked or that blocked it, people who reported it, other people using the same internet connection, and a child shown in material reported to us.
Depending on the case, a sealed copy can hold:
- the account's identity and sign-in details, such as its handle and display name and the name, email address and picture from its sign-in provider
- the country it chose and its date of birth
- records of any payment it made
- its chat messages, its debates, and its recordings with their transcripts
- reports about it, including reports emailed to us, and our moderation records
- material reported to us, such as a picture, which is sealed so that we never open it
- every picture the account uploaded, not only a reported one
- its friends, the accounts it blocked and the accounts that blocked it, with the dates
- from a debate the case concerns: its topic, which side each debater argued, what the other debater and anyone watching said, the judge's notes and scores, the account's own private notes for it, and, where the case concerns them, the other debater's recorded voice
- web server log lines from an internet address the police or a court name, which can include other people using that address
Who can receive a sealed copy
Data in a sealed copy goes only to:
- Cloudflare, which stores the encrypted copies for us and cannot read them. Cloudflare is covered by the EU-US Data Privacy Framework and its UK extension.
- The National Center for Missing & Exploited Children (NCMEC), a nonprofit organisation that United States law names to receive reports of child sexual abuse. It is not a government authority.
- Police and courts in the United States, under legal process.
- Authorities in the United Kingdom, under a United Kingdom order, and authorities in the European Union, under a European Union order.
- A court, and the other parties to a legal claim, under that claim's own procedure.
- Our email host, Hostinger, when part of a copy is sent by email.
- Neither the European Union nor the United Kingdom has found that NCMEC, or police and courts in the United States, protect data to their standard. Some of what we send them relies on the exceptions the law allows for such transfers, and some relies on no exception, because United States law requires it.
How long a sealed copy is kept
A sealed copy is deleted within two weeks of its need ending. Where a copy is locked for the period the law sets, it is deleted when that lock ends instead, which is at most two weeks after that period, even if the need ended sooner. Web server log lines in a copy are kept the same way. The periods are:
- After a report of suspected child sexual abuse: one year from the report. While a report is being prepared: up to 16 days.
- For a United States police request to preserve data: 90 days, or 180 days if the request is renewed.
- For a European Preservation Order: 60 days, or 90 days if it is extended, or, once the authority confirms that it wants the data, until we have handed it over.
- For a United Kingdom production order: until two weeks after the date the order sets for handing over the data it names.
- For a legal claim: until the claim ends.
The legal basis for a sealed copy
In the United Kingdom, we keep a copy after a report, for a police request or for a European Preservation Order because detecting and preventing crime and keeping children safe are recognised legitimate interests in UK data protection law, and a copy for a United Kingdom order because the law requires it. Where a copy holds information about a crime or other sensitive information, we rely on the conditions the Data Protection Act 2018 sets for preventing or detecting unlawful acts, safeguarding children and legal claims.
In the European Union, we keep the part of a copy about a reported act because of our legitimate interest in reporting a crime we come across. European guidance links that interest to reports made to the police, and our reports go to NCMEC, so this basis can be disputed. The rest of a copy kept after a report, and a copy kept for a United States police request or a United Kingdom order, is kept because that country's law requires it, and we know of no European Union law that does. In those copies, information about a crime and other sensitive information need a European Union or national law that allows them, and we know of none that reaches us. While a report is still being prepared, no duty to keep the rest of the copy exists yet: we keep it only to make the report the law requires, and we know of no European Union law that allows that either. We keep a copy for a European Preservation Order because European Union law requires it.
Everywhere, we keep a copy for a legal claim because of our legitimate interest in establishing, exercising or defending it. A copy kept for a claim holds information about a crime only where the law allows it.
Where a copy rests on our legitimate interest under Article 6(1)(f) of the GDPR (a legal claim everywhere, and in the European Union the part of a copy about a reported act), we have weighed it against the rights of everyone in the copy, and a child's interests generally come first. You can ask us for that weighing at support at peithra dot com.
Your right to object to a sealed copy
Where we keep a sealed copy because of a legitimate interest, you have the right to object at any time, for reasons to do with your own situation, by emailing support at peithra dot com. We will then stop, unless we have compelling reasons that override your interests, rights and freedoms, or we need the copy to establish, exercise or defend a legal claim.
Other requests and complaints about a sealed copy
An objection, or a request to delete your data, does not end a copy we must keep after a report, or for a police request or an order, while that duty lasts. We record what you asked, what we decided and why.
Whatever you ask us about your data, we ask which countries you were in when you used Peithra, as we ask everyone, so we know which privacy laws protect you.
To complain about a sealed copy, email support at peithra dot com. We will acknowledge your complaint within 30 days, keep you told of its progress, and tell you the outcome without undue delay. You can also complain to a data protection authority, such as the Information Commission in the United Kingdom or the authority in your country in the European Union, or go to court.
Contact
You manage your own data directly in the app: from Settings → Account you can download a copy of your data or permanently delete your account at any time. This self-service is the fastest way to exercise your data rights on Peithra.
For anything self-service does not cover, including a question about a payment, email support at peithra dot com.