Privacy Policy

Last updated: October 23, 2026

1. Data Controller

Polora is operated by polora Inc., a Delaware corporation ("Company," "we," "us"), which is the data controller responsible for your personal information.

Service Name: Polora
Legal Entity: polora Inc.
Representative: Henry (Founder & CEO)
Privacy Officer (Korea CPO): Henry (founder) · privacy@polora.ai
Address: polora Inc., c/o Legalinc Corporate Services Inc., 131 Continental Dr, Suite 305, Newark, DE 19713, USA
Contact: support@polora.ai · +1 302 694 0189
Business Registration: Delaware C Corporation, incorporated in the State of Delaware, USA. (Tax identifiers are held with the relevant authorities and are not published; available on lawful request.)
EU Representative (GDPR Art. 27): Not currently designated. EU users may exercise all GDPR rights by contacting privacy@polora.ai, and we answer those requests directly.

We comply with the Korean Personal Information Protection Act (PIPA), the EU General Data Protection Regulation (GDPR), and the California Consumer Privacy Act (CCPA/CPRA).

2. Information We Collect

2.1 Account Information

Email address (required)
Password (stored as bcrypt hash, we never store plaintext passwords)
Display name (optional)
Google account ID, profile name and profile picture (if you register via Google OAuth)
Company name (optional)
Apple account identifier and the name Apple shares (if you register via Apple)
Birth year and month (asked once, to apply the minimum age and the protection for people under 18 in messages ; the sign-up never asks for the day)

2.2 Service Usage Data

AI debate content (topics, questions, conversation context)
AI Persona configurations (custom personas and their memories) and the persona's teaching record described in 2.7
Uploaded documents, audio, video, and image files (for AI processing)
Credit usage history and transaction records
Referral code relationships
Subscription status and plan history
Shared debate metadata (share IDs, visibility settings)
User feedback submissions
How you heard about the service, in your own words (asked once after sign-up)

2.3 Technical Data

IP address
Browser type and version, operating system, screen resolution
Service access logs (timestamps, feature usage patterns)
First-party usage events stored on our own servers. Before you sign in or grant analytics consent, this is limited to a minimal anonymous visit signal : pages viewed (with their full address), session start and end, first and return visits, signup and sign-in steps, and views of publicly shared debate, Picks and persona pages together with the plain reading signals on them (opening a card, reaching the article body, switching the language edition, copying or saving a universe link, tapping share or the sign-up buttons there) and the campaign tag of the link you arrived by, with the last part of your IP address masked, your browser string kept with it, and no other behavioral detail recorded. After you sign in, or once you choose "Allow all" on the cookie banner, feature-usage events (which features you use and when) help us understand and improve the product. We also review aggregate usage patterns and debate topics internally to understand what the product is used for; the content of AI responses is not part of analytics.
Error diagnostics on our own servers. When the web app itself fails, the browser sends us the error message and stack (scrubbed of e-mail addresses and credentials), the page address without its query or any token, the screen name, the browser type and, when the error happened inside a conversation, that conversation's id ; no console log is attached. The browser string and screen size ride with it, and, if you are signed in, your account id. Errors raised on our servers are recorded with e-mail addresses and credentials removed. Both exist only to fix the service and are kept 30 days from when they were first seen once resolved.
WebSocket session identifiers and connection data
Local storage data in your browser : sign-in tokens and a cached profile, unsent drafts, the link you arrived by (campaign, referral, promotion), a visit counter, queued usage events, the app's notification token, your last chat settings, and display preferences (theme, panel sizes, sidebar state, dismissed banners)

2.4 Newsletters (Polora Picks and Polora Prism)

Each of our magazines, Polora Picks and Polora Prism, has its own weekly newsletter, and you subscribe to each one separately. If you subscribe, we collect your email address, the language of the page you signed up on, the address of that page (without anything after it such as search terms), the time zone your browser reports, the consent sentence you saw and its version, and when you signed up, confirmed and unsubscribed. We also keep which weekly issues we sent you. You do not need an account to subscribe.

If you have an account, agree to news and offers under the wording that names the Polora Picks newsletter, and have verified your account's email, that address receives the Polora Picks newsletter without a separate confirmation : verifying your email stands in for it. Polora Prism is sent only to people who sign up to it on its own form. We record that it started from that consent. Turning news and offers off stops the newsletter too. You can also stop only the newsletter, in Settings, Notifications, or from any issue ; agreeing to news and offers again later does not restart a newsletter you stopped yourself.

2.5 Published Persona Editions

When you publish a persona's public edition, the reviewed snapshot you confirmed is stored. Its display card (name, avatar or photo, introduction, highlights, nationality, publisher display name, links, up to two knowledge card titles and the number of cards, the details about yourself you chose to show, your star and its caption, how often it was saved and used, its retest score, translations into supported languages, and the number of statements) is retrievable by anyone, including through the public API. The approved statements themselves and the knowledge card bodies are the material the persona speaks from : the AI reads them when the persona answers, and they are not listed to visitors or returned by the public API. Withdrawing an edition stops new uses at once, with the short technical delay the Terms describe in Section 6.5 ; copies other users already saved remain in their accounts.

You can also fill in a few optional cells about yourself on the publish sheet and on your persona's page (sex, age band, marital status, education, household, occupation, region and languages). Every cell is optional ; the age band starts from your answers to the short questions before an interview and the country follows your persona's nationality, the rest start empty. Cells you fill in are kept in your persona and are shown on your public page, in the universe and through the public API only when you turn that on separately, and a change reaches the edition the next time you publish ; they are never sent to an AI provider and never used when the persona speaks. If you agree to it on your persona's page, they may also be used to choose which editions answer a research request and to group its aggregate results (Terms Section 6.4), never to identify you. You can clear them or turn them off at any time, they are included in your data export, and they are removed when you delete your account.

When a persona answers you while you are signed in, the conversation is a conversation record in your account like any other : you own it, the persona's publisher cannot read it unless you choose to show it to them in a message (Section 2.9), and it stays with you after the publisher withdraws the edition. A taste conversation (when you are not signed in) is not stored. To keep the number of one-to-one messages with personas in the universe within the daily limit the Terms describe, while that limit applies to you we record the time of each message you send to such a persona, which persona and conversation it was, and a one-way fingerprint of the message (never its words) for 2 days. When you talk to a persona you saved, what it remembers about you is sent to the AI provider answering that call, and the newest part of the conversation is read from time to time (at most once every ten minutes) by the persona model the service designates (its designated backup if that call fails ; a platform economy model only when neither is set up), which distils those notes. That distillation is an AI call billed to your personal credits (a flat-rate member whose own keys reach the model pays nothing), and you can turn it off on the persona's page (the switch on its 1:1 chats tab). A persona you saved keeps what it remembers about you until you delete the copy, until its publisher withdraws it or deletes the persona, until a report takes it down, or until either account is deleted. Taking a star off the universe listing does not erase it.

Synthetic personas drawn from licensed public datasets are shown in a separate universe. No person publishes them and no consent record applies to them : they carry the dataset's attribution instead of a publisher name (the source and licence of each such universe are listed in Section 6), and they are not reviewed as a person's public edition. When you talk to a synthetic persona, the same rules apply as for a person's public edition : a taste conversation (not signed in) is not stored ; a saved chat, which is open to every signed-in member because no exchange rule applies in a synthetic universe, is a conversation record of your own account ; and what a copy you saved remembers about you is sent to the AI provider answering that call, exactly as described above. Nothing you say to a synthetic persona reaches the dataset's publisher.

2.6 Persona Background Learning

Your own persona ("my persona") can learn about you in the background : as you use conversations (starting or resuming a debate, intervening in one, finishing one, or asking for a document review), an AI model reads what you wrote (your topic and your own messages, not other participants' words) together with the memories your persona already holds, and distills short memories into your persona. Each learning run is an AI call billed to your personal credits, and every charged run leaves a receipt you can see on your persona's screen (which conversation, how many memories, how many credits ; the most recent twenty are kept). This learning is controlled by a switch on your persona : new personas start with it OFF, and you can turn it on or off at any time. Teaching a persona directly is always available and does not depend on this switch.

2.7 What a persona keeps of what you teach it

Since September 11, 2026, teaching a persona leaves a record in your account : the words you type, the text extracted from files you attach, the questions the AI asked you, your answers to the short questions before an interview (or the fact that you skipped them), and the messages of your own that background learning read. This record sits next to the distilled memories so a memory can always be traced back to what you actually said. Only you can read it, and it is included in your data export. Deleting a memory removes it from the persona's private library; choosing to delete it together with its source blocks any further use of that source and marks its record as withdrawn (the record itself stays in your account until the account is deleted, so the withdrawal can be shown). Deleting a persona you created removes its whole record with it; the persona that stands for you yourself cannot be deleted on its own and its record is deleted within the 31 days after you delete your account.

Common questions (optional) : when you open a set of common questions for a persona you own, the persona's private material is sent to the AI model the service designates for personas, the same model that learns from what you teach it, so that it can fix its guesses (a probability over every option) before you answer. If that call fails, the same material may be sent again to the same model, up to three attempts, never to another model. Your own answers are kept in your account (a "don't know" is recorded as such but teaches nothing and is scored nowhere) and used for three things : comparing them with the persona's guesses, comparing them with your own later answers when the same questions are asked again after a while, and, since September 21, 2026, teaching the persona. Each answer becomes a memory of the persona (the situation and the option you chose, in your language), stored like any other memory, shown in your library, and included in your data export ; it can become part of a public edition only through the same publish confirmation as every other memory, and it is scored nowhere : the match number is computed only against the guesses fixed before you answered. What the public edition shows is that number (how many questions, how many the persona got right, and the day it was last measured), never an answer itself. When you talk with your own persona on its page you may mark a line it said as like you or not like you ; that mark, the line's text and any correction you type are kept in your account as the persona's record, the correction becomes a memory, and recent lines you marked as not like you, together with any correction you typed, may be included as feedback in what is sent to the AI when your persona speaks, guesses your answers to common questions, or asks you questions while you teach it. Answering the same questions again stays open after you take an edition down. Your answers stay in your account, with their scoring, until the account is deleted, and they are part of your data export. The occasional app notification or e-mail about questions waiting for you (new ones, or the same ones to answer again) is sent only if you turned it on in Settings, on the publish sheet or where you finish a round.

Star shape (optional) : since September 17, 2026, a person with a persona may ask the Service to draw their star's shape. Each time they ask for drafts or for a change in words, the persona's material (the teaching record, the distilled statements, the introduction and highlights, the answers to the short questions before an interview, attached links and the titles of knowledge cards) is sent to the AI provider serving the background model, which proposes shapes in a fixed vocabulary of celestial forms ; a change asked with a picture sends that picture and the material to a vision-capable model. The person picks one, may revise it by words or by hand, and decides whether to publish it. Before a shape becomes public the Service renders it as pictures on its own server and sends only those pictures to a vision-capable catalog model the platform chooses for review; no text about the person rides with them, and when that review cannot decide, a staff member looks at the pictures. A published shape (its block grid and the reviewed pictures) is stored on the Service's object storage and served from its CDN to anyone who comes close enough to that star in the persona universe. The original drafts and the reasons the AI gave stay in your account, visible to you only. Withdrawing the edition, deleting the persona or deleting the account removes the published shape and its pictures.

2.8 Payment Data

Payment processing is handled by Stripe. We do not directly store credit card numbers.
We store: Stripe customer ID, payment and subscription IDs, amounts, dates, the package or plan, the events Stripe sends us about them, and, if you turn on automatic top-up, your card's brand and its last four digits.

2.9 Messages Between People

When you send or receive a message, we store its text, the time, the name each side chose for that conversation, the persona it started from, any copy of an AI conversation you chose to attach (its most recent part, shortened), the photos and documents you send once the conversation is open (with their names, kinds and sizes), when each of you last opened the conversation, the people you blocked, and the reason you give when you report a message. Only the two people in the conversation can read it. Files are kept in private object storage, never at a public address. A photo also gets a small preview we make ourselves, shown inside the conversation through a link that works for an hour ; each time one of the two people downloads a file, we check who is asking and give a download link that works for one minute. Before a JPG, PNG or WebP photo is stored we remove the hidden information it carries, such as where it was taken.
Photos and documents are kept for 30 days after they are sent. Then the file and its preview are deleted, and the conversation shows that the file was kept for that time. A file held as evidence for a report under review stays until that report is gone.
When messages are switched on (at your first publish, unless you already chose) or off, and when you first use them, we record that choice and the version of the notice you agreed to.
If a message is reported, our team reads that message and opens its files to review the report. A copy of the reported message, and its files, are kept with the report for up to 180 days after the review.
Messages are kept while both accounts exist. When you delete your account, the messages you wrote and the files you sent are deleted from the other person's view at once and your name is removed from the conversation. The messages the other person wrote stay on their side.
Messages are included in your data export, with the names, kinds and sizes of the files in them (never the files themselves).

2.10 Emails and App Notifications

Some messages you always receive because they are part of your account : sign-in and security mail, a welcome when your account is verified, your data export, the confirmation of an account deletion, notices about your credits (credits you bought before they expire, and your balance reaching zero), payment problems, invitations to an organization, a notice about a persona you published once a report about it is reviewed, and notices of unfavorable or material changes to these policies, our Terms or the Service.
When another member sends you a first message or a new reply, you are told by app notification or e-mail (never its words), at most once per conversation a day ; this is on until you turn it off in Settings, Notifications.
News and offers (tips for getting started, a reminder to ask a second question, offers when your balance is low, reminders about gift credits or a trial ending) are sent only if you agreed, either with the optional box when you signed up or once when we asked in the service. We never ask for that agreement by email. We record the words you saw, their version, where and when you agreed or withdrew, and we send you a short notice of the result. From October 9, 2026 the wording also names the Polora Picks newsletter (Section 2.4) ; members who agreed before that are asked once in the service whether to receive it.
News about a persona you published (someone found it or saved it, and a short daily note) is on until you turn it off in Settings, Notifications. Invitations to answer the Guess me questions again are off until you turn them on there, at the end of a round or when you publish. A free credit you received (a welcome gift, a monthly refill, a gift from us) is noted in the app ; by email or app notification it counts as news and offers, sent only with your consent.
Messages that are not part of your account are never sent between 9 pm and 8 am where you are : they wait for your morning. For that we keep your browser's time zone and language with your account. The language you chose for the screens is never changed by this.
We keep a record of each notification we sent you (what kind, when, by email or app, and whether it went through), without your email address. Detailed records are kept for one year. The app shows you, under Messages, Alerts, what happened to your personas, credits and account, including things we did not send by e-mail or app, and a tap takes you to where it happened.
In the app, notifications come through Google Firebase Cloud Messaging (and Apple's push service on iPhone) only after you allow them. We store a token for that install, and remove it when you log out, when the service reports it no longer works, when the app has not been opened for 180 days, or when the account has more than ten newer installs.
You can change all of this in Settings, Notifications, from the link at the bottom of any such email, or with your mail program's unsubscribe button. Your notification settings, your newsletter subscription and its record, the record of what we sent and the app installs that receive notifications are included in your data export and deleted with your account ; your consent records are included in the export and kept as Section 8 states.

2.11 Polora Apps

Fortune

The Fortune app reads a question about your life through Bazi, Western and Vedic astrology and tarot, in a group chat of AI readers. When you ask for it, it also reads the whole life of you or of someone you added, one tradition at a time, and keeps that reading for you to open again. It is open to people 16 and older.

What the app keeps

What	Why	How long	When it is deleted
The birth date, the time (or a range of hours, or unknown), the city and the sex you save for yourself or for a person you keep in the app.	To compute the charts the readers read, and to send with a reading's calls to the AI companies in Section 5. A question reading sends the birth details you saved with the chart computed from them. A life reading and a reading with a compatibility friend send only the chart : the date, time and place are not written in it, but the birth moment can be worked out from it.	Until you delete it on your persona page or in the app, or delete that persona or your account.	Deleting a birth deletes every saved version of it and the charts computed from it, and ends any compatibility connection that hangs on it : the readings that include the other person are deleted on both sides.
The charts computed from a saved birth : the positions, pillars and periods the readers read.	So a chart is computed once and read again.	As long as the version of the birth it was computed from.	Deleted with that birth, the persona or your account.
Whose chart the app's home shows, and the day each daily card was turned.	To show the same home on every device.	Until you leave Polora.	Erased the moment you leave.
Who each person you added is to you (parent, friend, colleague and so on), when you chose one.	So the readings read that person as someone in your life, the way you described them.	Until you change it, delete that person, or delete your account.	Deleted with that person or your account, and erased the moment you leave.
The tarot cards drawn for a reading.	So every reader, a reload and a follow-up read the same cards.	As long as the reading.	Deleted with the reading, and erased the moment you leave.
Which tarot cards each question of a reading read (a follow-up can read the cards drawn before it).	So each tarot answer shows the cards it read.	As long as the reading.	Deleted with the reading, and erased the moment you leave.
A friend invite : the name you chose to show the friend, the note of who it is for if you wrote one (shown to you only), when the link was made and until when it works. The link itself is never kept, only a fingerprint of it.	To let a friend who accepts it, and agrees, see your compatibility with you.	Until the compatibility connection ends, the invite is taken back or expires, or you delete your account. The note of who it is for goes as soon as the invite is accepted, taken back or expires.	Ending a compatibility connection deletes the readings that include the other person, on both sides ; the invite's row goes with your account.
An accepted compatibility connection : the name the other person chose to show you, when you connected, and which version of your own birth the connection hangs on.	To read your compatibility with a friend who agreed.	Until either of you ends the connection, deletes their birth or leaves Polora.	Deleted with the connection, and with your account.
Which of your readings include a compatibility friend.	So ending the connection can delete exactly those readings, on both sides.	As long as the connection and the reading.	Deleted with the reading or the connection, and with your account.
A mark on a reading that used what you taught your own persona. It holds no words of it.	So that reading is never attached to a message or turned into a picture to share.	As long as the reading.	Deleted with the reading, and erased the moment you leave.
When each turn of a reading began and ended, with how many readers it had. It holds no words of the reading.	So the start screen can say how long a reading like it usually takes.	As long as the reading.	Deleted with the reading, and erased the moment you leave.
Each lifelong reading you start for a person and one tradition (Bazi, astrology or tarot) : its state, the conversation it runs in, the readers and models it was started with, which version of the birth it reads, and once it is done, which message holds the reading, when it was written and its one-line summary for the app's home.	So a reading you paid for runs once, can be paused and picked up again, keeps the readers it started with, and appears on the app's home when it is done.	Until you delete that birth or that persona, or leave Polora.	Deleted with the birth or the persona, and erased the moment you leave. The conversation it ran in stays like every reading, until you delete it.
Which finished reading the app shows as the current one for a person and a tradition.	To show the same current reading on every device.	Until you delete that birth or that persona, or leave Polora.	Deleted with the birth or the persona, and erased the moment you leave. Deleting the reading's conversation leaves no current reading.
What a lifelong reading's readers read, kept when it starts : the chart values computed from the birth (pillars, positions, periods, the birth cards), not the birth date, time or place, with the moment and the time zone the reading was read at.	So every reader of one reading, a reading picked up again later and the reading's pictures all read the same calculation.	As long as its reading run, once the reading is written. A reading that ends without being written (you stopped it, it failed, a new reading replaced it, or it never started) loses it once nothing can show it any more : at once, or after a last look of ten minutes for a reading that may still arrive.	Deleted with the reading run, its birth or its persona, and erased the moment you leave.

What the app uses but does not keep itself

What	Kept by	Why	How long	When it stops or is deleted
What you taught your own persona (its memories), each with how it came to be known.	Your persona	Only after you turned on readings that know you (both boxes on one sheet) and only in a reading about you alone : the memories go with that reading's calls to the AI companies in Section 5. They are not used in a reading about someone else, with a compatibility friend, or before the app knows whom a question is about.	Not kept by the app. They are read again for each reading and stay in your persona until you delete them there.	Turning it off stops the next call from using them, and so does deleting a memory ; readings already written stay as they are, and a reading that used them can't be attached or shared.
Your readings : your questions and the readers' replies, and the life readings written in them.	Your Polora conversations	To read them again and ask follow-ups, and to show the latest life reading of each person and tradition when you open the app.	Like every Polora conversation, until you delete it or your account.	Deleting a reading deletes it, and a life reading written in it with it. Turning off readings that know you, or deleting a birth, does not delete readings already written, except the readings that ending a compatibility connection deletes.
Your answers to the app's consents (the friend invite, readings that know you), with the version of the words and the list of AI companies you were shown.	Our consent records	To show what you agreed to, and when.	As Section 8 says for consent records.	Kept after your account is deleted for the period in Section 8, as proof of what you agreed to.

Description version 2026-10-28.

3. Legal Basis for Processing (GDPR Article 6)

We process your personal data on the following legal bases:

Contractual Necessity (Art. 6(1)(b)): Account management, credit system operation, AI API call processing, payment processing. These are essential to provide the Service you have agreed to use.
Legitimate Interest (Art. 6(1)(f)): Security measures, fraud prevention, service improvement, and the error diagnostics on our own servers and on the server side of our error-monitoring provider described in Sections 2.3 and 10.3 (with personal data removed before anything leaves), news about a persona you published, and notice of a message from another member (Section 2.10 ; both on until you turn them off).
Consent (Art. 6(1)(a)): News and offers by email and app notification (Section 2.10 ; asked with an optional box at sign-up or once in the service, never by email), app notifications on your device, the Polora Picks newsletter (double opt-in : no issue is sent until you confirm from your own inbox, and the confirmation email is the only mail before that ; a member who agrees to news and offers under the wording that names the newsletter receives it at the email address verified on their account, without that confirmation (Section 2.4); one email a week with up to five picks, delivered on Tuesday morning in the time zone your browser reported when you signed up, or in New York time if none was reported (a send that is held up may go out on a later morning, never at night); the exact consent sentence you saw, its version, and the times you signed up, confirmed and unsubscribed are recorded, and every issue carries an unsubscribe link, your mail program's own unsubscribe button, and a signed link to view or delete your subscription data), optional analytics, error monitoring (Sentry), referral program participation, the licence you grant when you publish a persona (Terms Section 6.4) and the separate consent it describes, and any consent an app asks for where you use it (Section 15). You may withdraw consent at any time.
Legal Obligation (Art. 6(1)(c)): Retaining payment and accounting records for the period required by applicable tax and accounting laws; responding to lawful government requests; telling a newsletter subscriber who used their mail program's one-click unsubscribe that it was done (Korean Network Act art. 50(7)).

4. Purpose of Information Use

We use collected information for:

Providing the Service and managing your account
Processing AI API calls by forwarding your content to AI Providers
Operating the credit system and processing payments
Service improvement and quality enhancement
Customer support and service announcements
Fraud prevention and security maintenance
Anonymous statistical analysis and error monitoring
Administering the referral program and subscription management
When research, surveys or simulations are offered through the Service, letting published persona editions answer their questions and delivering results that do not identify you to whoever asked, as Terms Section 6.4 describes
Sending each newsletter (Polora Picks, Polora Prism) to the subscribers who confirmed it, and the notices about their subscription (confirmation and unsubscribe results)
Seeing which pages people sign up to our newsletters from (the page address kept with each sign-up)

5. Data Sharing with AI Providers

The core function of this Service requires forwarding your content to AI Providers for response generation:

AI Provider	Data Transmitted	Data Location
OpenAI	The words of the conversation the provider is answering in : your question, your own messages, and what the other voices in that conversation said. The setup the provider needs to answer : who is taking part, their roles, the conversation's mode, and the instructions it runs under. The name on your account, so the AI you talk with knows who it is talking to. You can change it or leave it empty in Settings under Account. Your email address is never sent. Files you attach to a conversation, only when you include them, as extracted text or as the media itself for models that can read it. The reviewed public edition a user chose to publish. Sent to an AI provider for : the embedding that places the star and powers search and cast curation ; the pre-publish review and the six-language translation ; the relation judgment between two stars ; the region names of the sky ; and speaking from it in a taste conversation. Which provider depends on the model the Service designates for personas (or, for the taste conversation, the model the visitor picked). What a person hands a persona while teaching it (their typed answers, text extracted from files they attach, and their own messages that background learning reads) and the questions the AI asks back. Sent to the AI provider serving the model the service designates for personas. If that call fails, the material is sent once more to the designated backup model. If neither the main nor the backup model is set, a platform economy model is used. If a flat-rate member's own API keys cannot access that provider, the cheapest model available through those keys is used. The material is never sent to the AI the person selected for other tasks. For the optional common-questions feature, the persona's private material (including what the person taught it) is sent to that same designated model so that it can predict the persona's answers before the person answers. Since 2026-09-21 the person's own answers to the common questions, and the corrections they type when they mark a line their persona said as not like them, become memories of the persona and travel with this class. Since 2026-09-11 the same text is kept in the person's account as the persona's teaching record, readable only by them, exported with their data ; a memory-with-source delete marks its record as withdrawn (it is not erased), and the record leaves with the persona it belongs to or with the account. What a persona you saved remembers about you : short notes distilled from your earlier conversations with it, sent along when that persona answers you so it can keep the thread. You can turn this off on the persona's page, and turning it off also stops the distillation call. The short answers a person gives before an interview (who the persona is for, what for, an age band) or the fact that they skipped them. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their published star's shape, as part of the persona's own material. Since 2026-09-17 The distilled statements the persona keeps about the person (title, text, who said it, how it was learned, confidence). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 The promotional links a person attached to their persona (title, URL, note). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. The Service never fetches these URLs. Since 2026-09-17 The titles (never the body) of the knowledge cards a person attached to their persona. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 Pictures of a star's shape rendered by the Service from the person's block grid, several angles, sent once at the person's request to publish that shape to a vision-capable catalog model the platform chooses, so it can review the shape before it becomes public. No text about the person rides with the pictures. Since 2026-09-17 Short passages of the knowledge a person attaches to a persona (the text extracted from their files), sent to the embedding provider so the persona can find the passage that grounds an answer. A visitor never sees the passage itself, but the persona's answer may quote from it. Removed with the knowledge item, the persona or the account. The question a visitor asks a published persona, embedded once to find the knowledge passages that ground the answer. Not stored by the embedding provider beyond the call. In the Fortune app, a reading sends to the AI provider serving each of its calls : the name you keep for each person it reads and who they are to you, their age and sex, and their birth details or a chart they can be worked out from. With a compatibility friend, only the friend's chart, under no name. In the Fortune app, only after you turned on readings that know you (both consents) and only in a reading about you alone : what you taught your own persona, as short lines with how each came to be known. Read again for each reading, sent with that reading's calls (each reader, the overview and a follow-up's preparation) to the AI provider serving each call, and never stored by the app. A reading that used them is never attached to a message or turned into a picture to share.	United States
Anthropic	The words of the conversation the provider is answering in : your question, your own messages, and what the other voices in that conversation said. The setup the provider needs to answer : who is taking part, their roles, the conversation's mode, and the instructions it runs under. The name on your account, so the AI you talk with knows who it is talking to. You can change it or leave it empty in Settings under Account. Your email address is never sent. Files you attach to a conversation, only when you include them, as extracted text or as the media itself for models that can read it. The reviewed public edition a user chose to publish. Sent to an AI provider for : the embedding that places the star and powers search and cast curation ; the pre-publish review and the six-language translation ; the relation judgment between two stars ; the region names of the sky ; and speaking from it in a taste conversation. Which provider depends on the model the Service designates for personas (or, for the taste conversation, the model the visitor picked). The full transcript of a public conversation the Company is curating for Picks, sent to prepare the article, fold a verification chat into an editor's manuscript, and run each review and revision round. For a piece that begins as an editor's manuscript, the manuscript itself (the Company's own text, never a person's) is sent first to the AI models that verify it and then, with the record of that chat, to the model that prepares the reflection. What a person hands a persona while teaching it (their typed answers, text extracted from files they attach, and their own messages that background learning reads) and the questions the AI asks back. Sent to the AI provider serving the model the service designates for personas. If that call fails, the material is sent once more to the designated backup model. If neither the main nor the backup model is set, a platform economy model is used. If a flat-rate member's own API keys cannot access that provider, the cheapest model available through those keys is used. The material is never sent to the AI the person selected for other tasks. For the optional common-questions feature, the persona's private material (including what the person taught it) is sent to that same designated model so that it can predict the persona's answers before the person answers. Since 2026-09-21 the person's own answers to the common questions, and the corrections they type when they mark a line their persona said as not like them, become memories of the persona and travel with this class. Since 2026-09-11 the same text is kept in the person's account as the persona's teaching record, readable only by them, exported with their data ; a memory-with-source delete marks its record as withdrawn (it is not erased), and the record leaves with the persona it belongs to or with the account. What a persona you saved remembers about you : short notes distilled from your earlier conversations with it, sent along when that persona answers you so it can keep the thread. You can turn this off on the persona's page, and turning it off also stops the distillation call. The short answers a person gives before an interview (who the persona is for, what for, an age band) or the fact that they skipped them. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their published star's shape, as part of the persona's own material. Since 2026-09-17 The distilled statements the persona keeps about the person (title, text, who said it, how it was learned, confidence). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 The promotional links a person attached to their persona (title, URL, note). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. The Service never fetches these URLs. Since 2026-09-17 The titles (never the body) of the knowledge cards a person attached to their persona. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 Pictures of a star's shape rendered by the Service from the person's block grid, several angles, sent once at the person's request to publish that shape to a vision-capable catalog model the platform chooses, so it can review the shape before it becomes public. No text about the person rides with the pictures. Since 2026-09-17 In the Fortune app, a reading sends to the AI provider serving each of its calls : the name you keep for each person it reads and who they are to you, their age and sex, and their birth details or a chart they can be worked out from. With a compatibility friend, only the friend's chart, under no name. In the Fortune app, only after you turned on readings that know you (both consents) and only in a reading about you alone : what you taught your own persona, as short lines with how each came to be known. Read again for each reading, sent with that reading's calls (each reader, the overview and a follow-up's preparation) to the AI provider serving each call, and never stored by the app. A reading that used them is never attached to a message or turned into a picture to share.	United States
Google (Gemini)	The words of the conversation the provider is answering in : your question, your own messages, and what the other voices in that conversation said. The setup the provider needs to answer : who is taking part, their roles, the conversation's mode, and the instructions it runs under. The name on your account, so the AI you talk with knows who it is talking to. You can change it or leave it empty in Settings under Account. Your email address is never sent. Files you attach to a conversation, only when you include them, as extracted text or as the media itself for models that can read it. The reviewed public edition a user chose to publish. Sent to an AI provider for : the embedding that places the star and powers search and cast curation ; the pre-publish review and the six-language translation ; the relation judgment between two stars ; the region names of the sky ; and speaking from it in a taste conversation. Which provider depends on the model the Service designates for personas (or, for the taste conversation, the model the visitor picked). What a person hands a persona while teaching it (their typed answers, text extracted from files they attach, and their own messages that background learning reads) and the questions the AI asks back. Sent to the AI provider serving the model the service designates for personas. If that call fails, the material is sent once more to the designated backup model. If neither the main nor the backup model is set, a platform economy model is used. If a flat-rate member's own API keys cannot access that provider, the cheapest model available through those keys is used. The material is never sent to the AI the person selected for other tasks. For the optional common-questions feature, the persona's private material (including what the person taught it) is sent to that same designated model so that it can predict the persona's answers before the person answers. Since 2026-09-21 the person's own answers to the common questions, and the corrections they type when they mark a line their persona said as not like them, become memories of the persona and travel with this class. Since 2026-09-11 the same text is kept in the person's account as the persona's teaching record, readable only by them, exported with their data ; a memory-with-source delete marks its record as withdrawn (it is not erased), and the record leaves with the persona it belongs to or with the account. What a persona you saved remembers about you : short notes distilled from your earlier conversations with it, sent along when that persona answers you so it can keep the thread. You can turn this off on the persona's page, and turning it off also stops the distillation call. The short answers a person gives before an interview (who the persona is for, what for, an age band) or the fact that they skipped them. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their published star's shape, as part of the persona's own material. Since 2026-09-17 The distilled statements the persona keeps about the person (title, text, who said it, how it was learned, confidence). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 The promotional links a person attached to their persona (title, URL, note). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. The Service never fetches these URLs. Since 2026-09-17 The titles (never the body) of the knowledge cards a person attached to their persona. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 Pictures of a star's shape rendered by the Service from the person's block grid, several angles, sent once at the person's request to publish that shape to a vision-capable catalog model the platform chooses, so it can review the shape before it becomes public. No text about the person rides with the pictures. Since 2026-09-17 In the Fortune app, a reading sends to the AI provider serving each of its calls : the name you keep for each person it reads and who they are to you, their age and sex, and their birth details or a chart they can be worked out from. With a compatibility friend, only the friend's chart, under no name. In the Fortune app, only after you turned on readings that know you (both consents) and only in a reading about you alone : what you taught your own persona, as short lines with how each came to be known. Read again for each reading, sent with that reading's calls (each reader, the overview and a follow-up's preparation) to the AI provider serving each call, and never stored by the app. A reading that used them is never attached to a message or turned into a picture to share.	United States
xAI (Grok)	The words of the conversation the provider is answering in : your question, your own messages, and what the other voices in that conversation said. The setup the provider needs to answer : who is taking part, their roles, the conversation's mode, and the instructions it runs under. The name on your account, so the AI you talk with knows who it is talking to. You can change it or leave it empty in Settings under Account. Your email address is never sent. Files you attach to a conversation, only when you include them, as extracted text or as the media itself for models that can read it. The reviewed public edition a user chose to publish. Sent to an AI provider for : the embedding that places the star and powers search and cast curation ; the pre-publish review and the six-language translation ; the relation judgment between two stars ; the region names of the sky ; and speaking from it in a taste conversation. Which provider depends on the model the Service designates for personas (or, for the taste conversation, the model the visitor picked). What a person hands a persona while teaching it (their typed answers, text extracted from files they attach, and their own messages that background learning reads) and the questions the AI asks back. Sent to the AI provider serving the model the service designates for personas. If that call fails, the material is sent once more to the designated backup model. If neither the main nor the backup model is set, a platform economy model is used. If a flat-rate member's own API keys cannot access that provider, the cheapest model available through those keys is used. The material is never sent to the AI the person selected for other tasks. For the optional common-questions feature, the persona's private material (including what the person taught it) is sent to that same designated model so that it can predict the persona's answers before the person answers. Since 2026-09-21 the person's own answers to the common questions, and the corrections they type when they mark a line their persona said as not like them, become memories of the persona and travel with this class. Since 2026-09-11 the same text is kept in the person's account as the persona's teaching record, readable only by them, exported with their data ; a memory-with-source delete marks its record as withdrawn (it is not erased), and the record leaves with the persona it belongs to or with the account. What a persona you saved remembers about you : short notes distilled from your earlier conversations with it, sent along when that persona answers you so it can keep the thread. You can turn this off on the persona's page, and turning it off also stops the distillation call. The short answers a person gives before an interview (who the persona is for, what for, an age band) or the fact that they skipped them. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their published star's shape, as part of the persona's own material. Since 2026-09-17 The distilled statements the persona keeps about the person (title, text, who said it, how it was learned, confidence). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 The promotional links a person attached to their persona (title, URL, note). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. The Service never fetches these URLs. Since 2026-09-17 The titles (never the body) of the knowledge cards a person attached to their persona. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 Pictures of a star's shape rendered by the Service from the person's block grid, several angles, sent once at the person's request to publish that shape to a vision-capable catalog model the platform chooses, so it can review the shape before it becomes public. No text about the person rides with the pictures. Since 2026-09-17 In the Fortune app, a reading sends to the AI provider serving each of its calls : the name you keep for each person it reads and who they are to you, their age and sex, and their birth details or a chart they can be worked out from. With a compatibility friend, only the friend's chart, under no name. In the Fortune app, only after you turned on readings that know you (both consents) and only in a reading about you alone : what you taught your own persona, as short lines with how each came to be known. Read again for each reading, sent with that reading's calls (each reader, the overview and a follow-up's preparation) to the AI provider serving each call, and never stored by the app. A reading that used them is never attached to a message or turned into a picture to share.	United States
Mistral	The words of the conversation the provider is answering in : your question, your own messages, and what the other voices in that conversation said. The setup the provider needs to answer : who is taking part, their roles, the conversation's mode, and the instructions it runs under. The name on your account, so the AI you talk with knows who it is talking to. You can change it or leave it empty in Settings under Account. Your email address is never sent. Files you attach to a conversation, only when you include them, as extracted text or as the media itself for models that can read it. The reviewed public edition a user chose to publish. Sent to an AI provider for : the embedding that places the star and powers search and cast curation ; the pre-publish review and the six-language translation ; the relation judgment between two stars ; the region names of the sky ; and speaking from it in a taste conversation. Which provider depends on the model the Service designates for personas (or, for the taste conversation, the model the visitor picked). What a person hands a persona while teaching it (their typed answers, text extracted from files they attach, and their own messages that background learning reads) and the questions the AI asks back. Sent to the AI provider serving the model the service designates for personas. If that call fails, the material is sent once more to the designated backup model. If neither the main nor the backup model is set, a platform economy model is used. If a flat-rate member's own API keys cannot access that provider, the cheapest model available through those keys is used. The material is never sent to the AI the person selected for other tasks. For the optional common-questions feature, the persona's private material (including what the person taught it) is sent to that same designated model so that it can predict the persona's answers before the person answers. Since 2026-09-21 the person's own answers to the common questions, and the corrections they type when they mark a line their persona said as not like them, become memories of the persona and travel with this class. Since 2026-09-11 the same text is kept in the person's account as the persona's teaching record, readable only by them, exported with their data ; a memory-with-source delete marks its record as withdrawn (it is not erased), and the record leaves with the persona it belongs to or with the account. What a persona you saved remembers about you : short notes distilled from your earlier conversations with it, sent along when that persona answers you so it can keep the thread. You can turn this off on the persona's page, and turning it off also stops the distillation call. The short answers a person gives before an interview (who the persona is for, what for, an age band) or the fact that they skipped them. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their published star's shape, as part of the persona's own material. Since 2026-09-17 The distilled statements the persona keeps about the person (title, text, who said it, how it was learned, confidence). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 The promotional links a person attached to their persona (title, URL, note). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. The Service never fetches these URLs. Since 2026-09-17 The titles (never the body) of the knowledge cards a person attached to their persona. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 Pictures of a star's shape rendered by the Service from the person's block grid, several angles, sent once at the person's request to publish that shape to a vision-capable catalog model the platform chooses, so it can review the shape before it becomes public. No text about the person rides with the pictures. Since 2026-09-17 In the Fortune app, a reading sends to the AI provider serving each of its calls : the name you keep for each person it reads and who they are to you, their age and sex, and their birth details or a chart they can be worked out from. With a compatibility friend, only the friend's chart, under no name. In the Fortune app, only after you turned on readings that know you (both consents) and only in a reading about you alone : what you taught your own persona, as short lines with how each came to be known. Read again for each reading, sent with that reading's calls (each reader, the overview and a follow-up's preparation) to the AI provider serving each call, and never stored by the app. A reading that used them is never attached to a message or turned into a picture to share.	European Union (France)
DeepSeek	The words of the conversation the provider is answering in : your question, your own messages, and what the other voices in that conversation said. The setup the provider needs to answer : who is taking part, their roles, the conversation's mode, and the instructions it runs under. The name on your account, so the AI you talk with knows who it is talking to. You can change it or leave it empty in Settings under Account. Your email address is never sent. Files you attach to a conversation, only when you include them, as extracted text or as the media itself for models that can read it. The reviewed public edition a user chose to publish. Sent to an AI provider for : the embedding that places the star and powers search and cast curation ; the pre-publish review and the six-language translation ; the relation judgment between two stars ; the region names of the sky ; and speaking from it in a taste conversation. Which provider depends on the model the Service designates for personas (or, for the taste conversation, the model the visitor picked). What a person hands a persona while teaching it (their typed answers, text extracted from files they attach, and their own messages that background learning reads) and the questions the AI asks back. Sent to the AI provider serving the model the service designates for personas. If that call fails, the material is sent once more to the designated backup model. If neither the main nor the backup model is set, a platform economy model is used. If a flat-rate member's own API keys cannot access that provider, the cheapest model available through those keys is used. The material is never sent to the AI the person selected for other tasks. For the optional common-questions feature, the persona's private material (including what the person taught it) is sent to that same designated model so that it can predict the persona's answers before the person answers. Since 2026-09-21 the person's own answers to the common questions, and the corrections they type when they mark a line their persona said as not like them, become memories of the persona and travel with this class. Since 2026-09-11 the same text is kept in the person's account as the persona's teaching record, readable only by them, exported with their data ; a memory-with-source delete marks its record as withdrawn (it is not erased), and the record leaves with the persona it belongs to or with the account. What a persona you saved remembers about you : short notes distilled from your earlier conversations with it, sent along when that persona answers you so it can keep the thread. You can turn this off on the persona's page, and turning it off also stops the distillation call. The short answers a person gives before an interview (who the persona is for, what for, an age band) or the fact that they skipped them. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their published star's shape, as part of the persona's own material. Since 2026-09-17 The distilled statements the persona keeps about the person (title, text, who said it, how it was learned, confidence). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 The promotional links a person attached to their persona (title, URL, note). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. The Service never fetches these URLs. Since 2026-09-17 The titles (never the body) of the knowledge cards a person attached to their persona. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 Pictures of a star's shape rendered by the Service from the person's block grid, several angles, sent once at the person's request to publish that shape to a vision-capable catalog model the platform chooses, so it can review the shape before it becomes public. No text about the person rides with the pictures. Since 2026-09-17 In the Fortune app, a reading sends to the AI provider serving each of its calls : the name you keep for each person it reads and who they are to you, their age and sex, and their birth details or a chart they can be worked out from. With a compatibility friend, only the friend's chart, under no name. In the Fortune app, only after you turned on readings that know you (both consents) and only in a reading about you alone : what you taught your own persona, as short lines with how each came to be known. Read again for each reading, sent with that reading's calls (each reader, the overview and a follow-up's preparation) to the AI provider serving each call, and never stored by the app. A reading that used them is never attached to a message or turned into a picture to share.	China (People's Republic of China)
Moonshot AI (Kimi)	The words of the conversation the provider is answering in : your question, your own messages, and what the other voices in that conversation said. The setup the provider needs to answer : who is taking part, their roles, the conversation's mode, and the instructions it runs under. The name on your account, so the AI you talk with knows who it is talking to. You can change it or leave it empty in Settings under Account. Your email address is never sent. Files you attach to a conversation, only when you include them, as extracted text or as the media itself for models that can read it. The reviewed public edition a user chose to publish. Sent to an AI provider for : the embedding that places the star and powers search and cast curation ; the pre-publish review and the six-language translation ; the relation judgment between two stars ; the region names of the sky ; and speaking from it in a taste conversation. Which provider depends on the model the Service designates for personas (or, for the taste conversation, the model the visitor picked). What a person hands a persona while teaching it (their typed answers, text extracted from files they attach, and their own messages that background learning reads) and the questions the AI asks back. Sent to the AI provider serving the model the service designates for personas. If that call fails, the material is sent once more to the designated backup model. If neither the main nor the backup model is set, a platform economy model is used. If a flat-rate member's own API keys cannot access that provider, the cheapest model available through those keys is used. The material is never sent to the AI the person selected for other tasks. For the optional common-questions feature, the persona's private material (including what the person taught it) is sent to that same designated model so that it can predict the persona's answers before the person answers. Since 2026-09-21 the person's own answers to the common questions, and the corrections they type when they mark a line their persona said as not like them, become memories of the persona and travel with this class. Since 2026-09-11 the same text is kept in the person's account as the persona's teaching record, readable only by them, exported with their data ; a memory-with-source delete marks its record as withdrawn (it is not erased), and the record leaves with the persona it belongs to or with the account. What a persona you saved remembers about you : short notes distilled from your earlier conversations with it, sent along when that persona answers you so it can keep the thread. You can turn this off on the persona's page, and turning it off also stops the distillation call. The short answers a person gives before an interview (who the persona is for, what for, an age band) or the fact that they skipped them. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their published star's shape, as part of the persona's own material. Since 2026-09-17 The distilled statements the persona keeps about the person (title, text, who said it, how it was learned, confidence). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 The promotional links a person attached to their persona (title, URL, note). Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. The Service never fetches these URLs. Since 2026-09-17 The titles (never the body) of the knowledge cards a person attached to their persona. Sent to the AI provider serving the background model when the person asks the Service to draw or revise their star's shape. Since 2026-09-17 Pictures of a star's shape rendered by the Service from the person's block grid, several angles, sent once at the person's request to publish that shape to a vision-capable catalog model the platform chooses, so it can review the shape before it becomes public. No text about the person rides with the pictures. Since 2026-09-17 In the Fortune app, a reading sends to the AI provider serving each of its calls : the name you keep for each person it reads and who they are to you, their age and sex, and their birth details or a chart they can be worked out from. With a compatibility friend, only the friend's chart, under no name. In the Fortune app, only after you turned on readings that know you (both consents) and only in a reading about you alone : what you taught your own persona, as short lines with how each came to be known. Read again for each reading, sent with that reading's calls (each reader, the overview and a follow-up's preparation) to the AI provider serving each call, and never stored by the app. A reading that used them is never attached to a message or turned into a picture to share.	Global endpoint (api.moonshot.ai); the operator, Moonshot AI, is headquartered in China (People's Republic of China)

The name on your account goes with your conversations, so the AI you talk with knows who it is talking to. You can change it, or leave it empty, in Settings under Account. Your email address and other account details are never transmitted to AI Providers.
What is transmitted is what you write and attach : your question, the conversation so far, and any file you add. Each call also carries today's date and time in the time zone your browser reports, and whether you are signed in. If you include personal information in those, it reaches the AI Provider handling that call. Please treat anything you type as something you are sending onward.
Commercial API usage with these providers generally does not involve training on your data, per their API terms of service.
When web search is enabled, AI models may also access publicly available web content through their built-in search capabilities (Gemini grounding, OpenAI web search, Anthropic web search, xAI web search, Mistral web search).
Our magazines, Polora Picks and Polora Prism, feature pieces that we make, not debates from your account. Since October 2026 a pick usually begins as a manuscript written by a Polora editor : we send that manuscript to several AI models that check its facts and sources in a group chat, and then send the manuscript and the record of that chat to Anthropic to fold the corrections back in and to revise, review and translate the piece. Earlier picks were written from group chats we ran ourselves ; those transcripts go the same way. Your own debates are never used this way.

6. Third-Party Services

We share limited data with the following service providers:

Stripe: Credit-package purchases, subscription billing, and refund processing
Google Analytics 4: Aggregate usage analytics (page views, feature engagement) in the browser, and one server-side purchase or refund event per sale so revenue reports are complete
PostHog (product analytics): Product analytics : autocaptured UI interactions and page views to understand feature engagement and improve the product
Sentry: Error reporting for service reliability : in the browser only after analytics consent ; on the server for every request, with personal data removed before anything leaves
Google OAuth (Sign-in with Google): Federated sign-in. User-initiated.
Apple (Sign in with Apple): Optional sign-in and sign-up. Apple authenticates the person and returns an identity token we exchange for a Polora session.
Google Drive (file attachments): Read attachments selected by the user via Drive Picker for inclusion in AI prompts.
Cloudflare Turnstile (anti-abuse challenge): Invisible CAPTCHA to block automated abuse on signup, contact form, and demo channels
IndexNow (Microsoft Bing-backed search index push): Notify search engines (Bing, Yandex, Seznam, Naver) when a Picks-promoted public debate URL is created or updated.
Resend (email delivery): Send service emails through the one notification road : account and security notices (email verification, password reset, account confirmation, verification reminder, data-export file, account-deletion confirmation, organization invitations, publisher notices), billing notices (credits you bought about to expire, a zero balance, auto-recharge failure), terms and policy change notices, the result notice of a marketing consent or withdrawal, activity news (news about a persona you published, on until turned off ; Guess me invitations when turned on), news and offers only with marketing consent (getting-started tips, a second-question reminder, low-balance offers, gift-credit and trial reminders), contact-form relay, and the Picks newsletter (double opt-in confirmation, weekly issues, one-click unsubscribe receipt)
Google Firebase Cloud Messaging (app notifications): Deliver app notifications to Polora's Android and iOS apps, including separate apps that use your Polora account, when the person turned notifications on in the app (news about a persona you published, Guess me invitations when turned on, account and billing notices, and news and offers only with marketing consent). On iOS, Firebase passes the notification to Apple Push Notification service.
YouTube (Google): Publishing the Company's own films (short films about Polora Picks, and films about Uju, the persona universe) to the Polora YouTube channel
Instagram (Meta): Publishing the Company's own pick cards and reels to the Polora Instagram account
Facebook (Meta): Publishing the Company's own pick cards to the Polora Facebook page
Bluesky: Publishing the Company's own pick cards to the Polora Bluesky account
Threads (Meta): Publishing the Company's own videos to the Uju Threads account
TikTok: Publishing the Company's own videos to the Uju TikTok account
X: Publishing the founder's own posts and videos to the Polora X account
DigitalOcean (Hosting sub-processor): All Service data is stored on DigitalOcean infrastructure under their Data Processing Agreement. Files and images the Service publishes (Polora Picks attachments, persona photos, the pictures of a published star's shape, and the persona universe's star maps and cards of published editions) are stored in DigitalOcean object storage and served from its CDN to anyone who can reach that public page ; a persona photo sits at an unguessable public address from the moment you set it. Screenshots you attach to a feedback report are stored there too, under an unguessable address that is the only protection of that public bucket, and are deleted with your account. See Section 7 for transfer details.

We do not sell your personal information without your consent.

If the Company merges with, is acquired by, reorganizes, winds down, or transfers all or part of the Service or its assets, personal information may pass to the successor as part of that transaction. The recipient remains bound by this Policy for information collected under it, and we give a notice in the Service and any notice the law requires.

The Service also reads the following public datasets. They are sources, not processors : their rows become the synthetic personas of a separate universe, shown with this attribution, and no user data flows to their publishers.

NVIDIA Nemotron Personas Korea (CC BY 4.0) : Synthetic personas of a separate universe (not a person's public edition) : the stars, their descriptions and their attributes are the dataset's rows, shown with this attribution. inbound only : the dataset's rows are read into the Service ; nothing a user says to a synthetic persona is sent to the dataset's publisher.
NVIDIA Nemotron Personas Japan (CC BY 4.0) : Synthetic personas of a separate universe (not a person's public edition) : one hundred thousand fictional Japanese residents whose descriptions and attributes are the dataset's rows, shown with this attribution. inbound only : the dataset's rows are read into the Service ; nothing a user says to a synthetic persona is sent to the dataset's publisher.
NVIDIA Nemotron Personas USA (CC BY 4.0) : Synthetic personas of a separate universe (not a person's public edition) : one million fictional Americans whose descriptions and attributes are the dataset's rows, shown with this attribution. inbound only : the dataset's rows are read into the Service ; nothing a user says to a synthetic persona is sent to the dataset's publisher.

7. International Data Transfers

The Service is provided by polora Inc., a Delaware corporation, and is hosted on cloud infrastructure located in the United States. Your data may be transferred to and processed in the following jurisdictions:

United States: OpenAI, Anthropic, Google (Gemini), xAI (Grok), Google Analytics 4, Sentry, Google OAuth (Sign-in with Google), Resend (email delivery), Google Firebase Cloud Messaging (app notifications), YouTube (Google), Instagram (Meta), Facebook (Meta), Bluesky, Threads (Meta), TikTok, X
European Union (France): Mistral
China (People's Republic of China): DeepSeek
Global endpoint (api.moonshot.ai); the operator, Moonshot AI, is headquartered in China (People's Republic of China): Moonshot AI (Kimi)
United States (PCI DSS Level 1): Stripe
United States (US Cloud): PostHog (product analytics)
Apple global infrastructure: Apple (Sign in with Apple)
User's own Google account (US infrastructure): Google Drive (file attachments)
United States (App Platform default region): DigitalOcean
Cloudflare global edge network: Cloudflare Turnstile (anti-abuse challenge)
Microsoft-operated endpoint (api.indexnow.org): IndexNow (Microsoft Bing-backed search index push)

For transfers from the EU/EEA to the United States and other jurisdictions, we rely on:

The EU–U.S. Data Privacy Framework, where the receiving provider is certified
Standard Contractual Clauses (SCCs) for transfers to jurisdictions without an adequacy decision, including sub-processor DPAs
The necessity of the transfer for the performance of the contract (GDPR Art. 49(1)(b)) when the user explicitly selects an AI provider

By using the Service, you acknowledge that your content will be processed by the providers in the jurisdictions listed above.

8. Data Retention

Account information: Retained until account deletion
Conversations: Retained until you delete them or your account. In a conversation you archived and did not share publicly, the messages are deleted 180 days after archiving ; the conversation itself stays until you delete it.
Photos and documents sent between people: 30 days after sending, then deleted (see 2.9).
Daily count of one-to-one messages with personas in the universe: the time of each message, which persona and conversation it went to, and a one-way fingerprint of it (not its words), 2 days, then deleted (see 2.5).
Deleted conversations across your devices: when you delete a conversation we keep only its id and the time, so your other devices remove it too ; kept 90 days, and deleted with your account.
Messages you deleted for yourself: we keep which message you hid, so it stays hidden on all your devices ; the other person keeps their copy. Kept while the message exists, deleted with your account.
AI Persona data (including the teaching record described in 2.7): Memories are retained until you delete them or the persona; the teaching record is retained until you delete the persona you created it for, or until account deletion (deleting a memory with its source marks its record as withdrawn rather than erasing it)
What a saved persona remembers about you (relation memory): Kept until you delete the copy, until its publisher withdraws or deletes the persona, until a report takes it down, or until either account is deleted. Taking a star off the universe listing does not erase it.
Uploaded files: Files you attach to a conversation are saved to your own Google Drive and stay there until you delete them ; we keep each file's name, type and size with the conversation until you delete it or your account. Files you hand to a persona's signature knowledge are converted to text at once; the file itself is never stored, and the extracted text stays in your account until you delete it (only you can read it; the persona uses it when answering). A document you open in the document review app is read for that review and is not kept on our servers ; a group chat you start from it keeps the document's text as that conversation's context, under the conversation rule above.
Newsletters (Polora Picks, Polora Prism): For each newsletter you subscribe to, your email, language, time zone, consent history and the record of which issues we sent you are kept while you are subscribed. A sign-up you never confirm is deleted after 7 days (if you had subscribed before, only that new sign-up lapses and your earlier record stays for the period below). After you unsubscribe we keep that record for 365 days, so we never mail you again and can show what you had agreed to, then delete it. Deleting your subscription data from the link in any issue, or deleting the Polora account that uses the same email address, removes it at once.
Payment and credit records: Kept while your account exists. After you delete your account they stay for 5 years without your identity, as tax and commerce law require, then they are deleted.
Consent records: Kept while your account exists. After you delete your account they stay for 5 years keyed only by a one-way hash of your e-mail, so a later claim about what you agreed to can be answered, then they are deleted.
Service audit log: 180 days
Reported messages and reported public personas: the copy kept with a decided report goes after 180 days
Service access logs: kept by our hosting provider for its standard log period ; we do not keep them elsewhere
First-party usage events (our own servers): 180 days for behavioral events. Account-lifecycle events (signup, email verification, first debate) are kept as long as the account exists, and aggregated daily counts, which identify no one, are kept without a time limit, so long-term trends survive without keeping raw activity.
Error records (our own servers): 30 days once resolved
Error monitoring data (Sentry): 90 days
Analytics data (Google Analytics): 14 months

When you delete your account it is anonymized at once (your e-mail, name, password, keys, preferences and attribution are erased, your public personas are withdrawn, and your messages leave the other person's view), and the rest of your data is permanently deleted within 31 days. After that, only the following remain :

Copies of your public personas that other people saved, and the conversations they had with them, stay in their accounts.
Messages the other person wrote to you stay on their side. The messages you wrote are deleted, and your name is removed from the conversation.
Relationship rows between two stars keep the surviving star's side ; they name no one once your edition is gone.
Your purchase and credit records stay for 5 years without your identity, as tax and commerce law require.
The record of what you consented to stays for 5 years, keyed only by a one-way hash of your e-mail.
A one-way hash of your e-mail stays so sign-up gifts cannot be claimed twice.
The service's audit log stays for 180 days and names no one.
A message or a public persona that was reported stays with the decided report for 180 days and names no one.
The sky's static tiles and cards are rebuilt without your star at the next generation, and edge caches expire within a minute.

Database backups : our database provider takes a backup every day and keeps each one for 7 days. Data removed from the active database may remain in those backups until they expire.

9. Your Rights

Depending on your jurisdiction, you have the following rights regarding your personal data:

9.1 Rights Under GDPR (EU/EEA Users)

Right of Access: Request a copy of your personal data.
Right of Rectification: Request correction of inaccurate data.
Right of Erasure: Request deletion of your data ("right to be forgotten").
Right to Restriction: Request limitation of processing in certain circumstances.
Right to Portability: Receive your data in a structured, machine-readable format.
Right to Object: Object to processing based on legitimate interest.
Right to Withdraw Consent: Withdraw consent at any time, without affecting the lawfulness of prior processing.
Right to Lodge a Complaint: File a complaint with your local data protection authority.

We will respond to GDPR requests within 30 days.

9.2 Rights Under Korean PIPA

Right to access and request copies of your personal information
Right to request correction or deletion
Right to request suspension of processing
Right to request destruction of personal data

We will respond to PIPA requests within 10 days.

9.3 Rights Under CCPA/CPRA (California Users)

See Section 13 for California-specific disclosures.

You exercise these rights yourself in the Service : Settings → Account downloads a copy of your data (access and portability), deletes your account (erasure), disconnects Google Drive and revokes our access at Google, and switches analytics off ; Settings → Notifications withdraws any consent you gave for e-mail notifications, and app notifications are turned off in your device's settings or by logging out ; a persona's page withdraws its public edition at once ; a message thread lets you block or report the other person. If you cannot sign in, or your request is not covered by these, contact us at privacy@polora.ai.

10. Cookies and Tracking Technologies

On your first visit you will see a cookie banner asking how you want to be tracked. Your choice is stored in localStorage under polora_cookie_consent and is honored on every subsequent visit from this browser.

Necessary only: We never load Google Analytics, PostHog, or the browser side of Sentry. Only the technologies listed in 10.1 below are active. Separately from cookies, our own servers keep a minimal anonymous visit count (pages viewed, session boundaries and shared-page views, IP partially masked) so we can tell how many people visit, and receive the browser's error report described in Section 2.3 when the web app itself fails. Neither contains behavioral detail and neither is shared with third parties.
Allow all: We additionally load the analytics and error-monitoring technologies listed in 10.2 and 10.3. Our own servers also record the full first-party usage events described in Section 2.3 (which features you use), even before you create an account.

You can change your choice at any time : signed-in users will find an "Analytics cookies" switch under Settings → Account, and any visitor can clear the polora.ai site data in their browser settings to see the banner again. Turning analytics off takes effect immediately, without a reload.

Native mobile apps (iOS and Android) : this applies to every mobile app Polora provides, including separate apps that use your Polora account. Our native apps do not load any of the third-party analytics or error-monitoring technologies listed in 10.2 and 10.3, and do not set tracking cookies. The cookie banner therefore appears only on the website. The apps do not track you across other companies' apps or websites, and no App Tracking Transparency prompt is required.

10.1 Essential Technologies (always active)

Sign-in tokens: An access token and a refresh token stored in localStorage for session management. Required for Service functionality.
Google sign-in and Cloudflare Turnstile scripts: Loaded on every page of the web app so that Google sign-in and the bot check are ready when needed ; they set no tracking cookie of ours.
Theme Preference: Stored in localStorage (light/dark mode selection).
IndexedDB: Used for local conversation persistence and offline capabilities.
Cookie consent record: A small JSON blob stored in localStorage (polora_cookie_consent) so we remember your choice.
Anonymous visitor marker: A first-party cookie (polora_visitor, a random value, HttpOnly, expires after 24 hours) set when you read a public persona page without signing in. Together with your IP address it counts the free daily conversation allowance per visitor, so that several people behind one address are not treated as one person. It identifies no account and is not used for analytics.

10.2 Analytics (only after "Allow all")

Google Analytics 4: Uses cookies (_ga, _ga_S0GP1SW2MM) for usage statistics. IP anonymization is enabled. Data retention is set to 14 months. The gtag.js library is not loaded at all until you grant consent.
PostHog: Product analytics with automatic interaction capture (identified_only mode). The PostHog library is dynamically imported only after consent.

10.3 Error Monitoring (only after "Allow all")

Sentry (browser): Collects error reports and performance data for service reliability. Session replay masks all text and blocks media before upload. No advertising or tracking purposes. Initialization is deferred until consent.
Sentry (our servers): Errors raised on our servers are reported to Sentry regardless of your cookie choice, because they are about the service and not about you : before anything leaves, e-mail addresses, tokens and other credentials are replaced, values under secret-like fields are redacted, and request bodies and cookies are dropped (Section 3, legitimate interest).

We do not use third-party advertising cookies. We do not currently respond to Do Not Track (DNT) browser signals as there is no industry standard for compliance. Our cookie banner is the SSoT for your consent.

11. Data Security

All data in transit is encrypted via HTTPS/TLS.
Passwords are stored using bcrypt one-way hashing.
Database access is strictly controlled with role-based permissions.
Rate limiting protects against brute-force and abuse attacks.
SSRF protection prevents unauthorized internal network access.
Input validation and output sanitization prevent injection attacks.
Regular security audits and vulnerability assessments are conducted.

Data Breach Notification

In the event of a personal data breach that poses a risk to your rights:

We will notify the relevant supervisory authority within 72 hours (GDPR) or without delay (PIPA).
We will notify affected users promptly with details about the breach, its likely consequences, and the measures taken.

12. Children's Privacy

This Service is not intended for users under the age of 14. We do not knowingly collect personal information from children under 14. Where local law sets a higher minimum age of digital consent (for example, up to 16 in certain European countries), that higher age applies. If the birth year and month an account gives are below the minimum age, we close the account and do not keep the answer. See our Terms of Service. If we discover that we have collected data from a child below the applicable age, we will delete it immediately. If you believe a child has provided us with personal information, please contact us at privacy@polora.ai.

13. California Privacy Disclosures (CCPA/CPRA)

If you are a California resident, you have the following additional rights:

Right to Know: You may request details about the categories and specific pieces of personal information we collect.
Right to Delete: You may request deletion of your personal information.
Right to Opt-Out of Sale: We do not share your personal information for cross-context behavioral advertising. Where we would pass a published persona edition, or answers traceable to you, to a person or organization that asked for research through the Service, we do so only with your separate consent, which you can withdraw at any time.
Right to Non-Discrimination: You will not receive inferior service for exercising your privacy rights.

Categories of Information Collected

Identifiers: Email address, display name, IP address, Google account ID, Apple account identifier
Characteristics: Birth year and month
Commercial Information: Credit purchase history, transaction records
Internet Activity: Service usage logs, feature usage patterns, browser information
User-Generated Content: Debate topics, uploaded documents, AI persona configurations

Financial Incentive Disclosure

Our referral program (bonus credits awarded to the referring user per successful referral, at the rate published in the Service) qualifies as a financial incentive program under CCPA. This program is offered to enhance user engagement and is not tied to the value of your personal data. You may opt out of this program at any time without penalty.

14. Automated Decision-Making

In accordance with GDPR Article 22:

Credit Exhaustion: The Service automatically prevents usage when credits are depleted. This is a contractual necessity, not a profiling decision.
Rate Limiting: Automated systems limit API usage rates to prevent abuse. These are security measures applied equally to all users.
Fraud Detection: We may use automated systems to detect referral fraud or payment abuse. Flagged accounts are reviewed by a human before any action is taken.

We do not use automated decision-making that produces legal effects or similarly significantly affects you without human involvement.

15. Policy Changes

We may update this Privacy Policy to reflect changes in our practices or legal requirements.
We tell every user about an unfavorable or material change by email and with a notice in the service at least 30 days before it takes effect. We post any other change in the service at least 7 days before it takes effect. If a change needs your new consent, we ask for it in the service once the change takes effect ; if you do not agree, only that new use does not apply to you.
When we add a new optional app, we explain how it uses your personal data before you first use it, and which providers it uses and whether your persona may learn from it are declared in its code and appear in this policy the same day (Sections 2.6, 5 and 6). Nothing changes for anyone who does not use that app, so adding an app does not by itself bring the e-mail and notice above ; a change to how we handle data you already gave us still does.
The "Last updated" date at the top of this page shows the most recent change to this policy's own text. An app's declaration can change what the lists in Sections 2.6, 5 and 6 show without moving that date.
Continued use of the Service after the effective date constitutes acceptance of the updated policy.

16. Contact Information and Complaints

Privacy Officer (Korea CPO): Henry (founder) · privacy@polora.ai
General Support: support@polora.ai
Legal Entity: polora Inc.
Address: polora Inc., c/o Legalinc Corporate Services Inc., 131 Continental Dr, Suite 305, Newark, DE 19713, USA

Filing Complaints

Korean Users: You may file a complaint with the Personal Information Protection Commission (PIPC) at pipc.go.kr
EU Users: You may file a complaint with your local Data Protection Authority (DPA).
California Users: You may contact the California Attorney General's office.

EU representative status: not currently designated (English/global launch posture). We will designate an EU representative as required by GDPR Article 27 if EU traffic becomes material.