← Halmeoni

Privacy

Last updated 2026-10-31

This page describes what Halmeoni collects, why, how long we keep it, and who processes it. In short: we keep as little as we can, and we do not use your data to train AI models.

We process the data below to give you the readings and conversations you ask for (performance of our contract with you; GDPR art. 6(1)(b) where it applies). Nothing here relies on separate consent. Please do not write about sensitive matters such as health, religion or politics in the chat.

What we collect

  • Birth details (name or nickname, date, time, birthplace, and optionally sex at birth) to calculate your chart. If you are signed in, the charts you open are saved to your account so you can come back to them; you can delete each one in the app.
  • Your questions and Halmeoni’s answers, sent to our AI provider to produce each reading. If you are signed in, your conversations are saved with the chart they belong to and deleted with it. Conversations that trigger our crisis response are never saved.
  • Your email address, if you sign in, and whether it was invited to the beta. If you sign in with Kakao and don't agree to share your email, we don't receive it.
  • Your sign-in account details, if you sign in with Google or Kakao: the member ID, name or nickname and profile picture address they send us. Used only to recognise your account, and deleted when you leave.
  • Beta keys: ended on 2026-10-31. Key hashes and the charts, conversations and feedback saved under keys were deleted the same day, and AI usage records left under keys were changed so they can no longer be linked to anyone.
  • Feedback you choose to send: your rating and comment, and for a rated answer, an excerpt of the question and answer (never your birth details). A copy goes to the operator’s Google spreadsheet so it can be read, and comments and thumbs-down ratings are also emailed to the operator.
  • Usage records: for each AI call, the feature, the model, the number of tokens, how long it took and whether it failed. These records hold no question or answer text.
  • Usage analysis (pseudonymised): from saved charts, chats, readings and usage records we make a pseudonymised copy — names, birth dates and times, birthplaces and emails removed or replaced with labels — to count which features and questions are common and decide what to build next. Only the operator sees it; we never try to re-identify anyone, share it, or use it for advertising or AI training. If an AI tool (Anthropic’s Claude) helps with the analysis, it gets only the pseudonymised copy, with training turned off.
  • Crisis-message counts: the time and language each time our crisis message is shown, with no text and nothing that identifies you, to check the safeguard works and to report as the law requires.
  • Which link brought you: the tag at the end of a post’s address (for example ?ref=disquiet). Your browser keeps the first tag in localStorage for 30 days, and it is stored once when you sign up. It only counts which post people came from, holds no birth details or name, and is deleted when you leave.
  • Cookies, all needed for the service. We use no analytics cookies; Google sets ad cookies only when you choose to watch an ad (see “Lucky pouches and ads” below):
    • hm_locale: your language, one year
    • hm_theme: your light or dark choice, one year
    • hm_age_block2: marks a browser that was turned away for age (no age or birth date in it), 24 hours
    • a sign-in session, if you sign in
  • What you typed before signing in (temporary): while you leave to sign in after pressing the button, the name (or nickname), birth date, time, place and gender you entered wait for 10 minutes in your own browser (localStorage) only. They are not sent to us or stored on our side; they are used and erased when the page opens again after sign-in, and ignored after 10 minutes.
  • A one-way, keyed hash of your IP address to limit abuse of feedback sent without signing in. We do not store your IP address itself.
  • Your country and region, as our host estimates them from your IP address, only to apply the minimum age your local law sets. Not stored.
  • Purchases (when they open): the order number, product and the readings you have used. Card details go to our payment reseller and never reach us.
  • Lucky pouches and ads: only when a signed-in user chooses to watch an ad does Google load one. Google then processes your IP address, browser details and ad cookies (to cap how often an ad is shown and to stop fraud). We ask for non-personalized ads only, and never pass your birth details or conversations to ads. We record only when you watched an ad, how many pouches you got and what you spent them on, kept with your account and deleted with it. No ads are shown in the EU, the EEA, the UK or Switzerland.

How long we keep it

  • Saved charts and conversations: until you delete them or your account.
  • Feedback and usage records: for the beta and up to 12 months after it, then deleted or anonymised.
  • Pseudonymised analysis copies: deleted when the analysis is done, at the latest one year after they were made. Totals that identify no one may be kept.
  • Written readings are cached by chart (the eight characters, not your birth date, time, place or name) so the same chart is not written twice.

Who processes it

  • Our AI provider, OpenRouter, Inc. (United States), which passes requests to a model host it selects to run Anthropic Claude models, or OpenAI GPT models as a fallback (hosts such as Anthropic, OpenAI, Amazon Web Services and Microsoft Azure, in the United States and other countries) — generates readings from your chart and questions. We ask that requests go only to hosts that do not store prompts or use them for training.
  • Supabase Inc. (a United States company; data stored in Seoul, South Korea) — stores accounts, saved charts and conversations, balances, feedback and usage records.
  • Vercel Inc. (United States) — hosts the site, keeps access logs and estimates your country from your IP address.
  • Anthropic, PBC (United States) — processes pseudonymised copies during usage analysis, with training turned off.
  • Google LLC (USA) or Kakao Corp. (South Korea) — only if you choose to sign in with them, to confirm it is you and send the account details above.
  • Google (United States) — a rewarded ad, only when you choose to watch one; non-personalized ads only (policies.google.com/technologies/ads).
  • Google (United States) — a copy of feedback in the operator’s spreadsheet, and alert emails to the operator.
  • Lemon Squeezy — our merchant of record for payments, taxes and receipts (when purchases open).

These providers process data outside your country, mainly in the United States, and the operator is in South Korea. Data is sent over encrypted connections (HTTPS) each time you use the service. Their privacy policies: openrouter.ai/privacy, policies.google.com/privacy, supabase.com/privacy, vercel.com/legal/privacy-policy.

What we never do

  • Sell your data or use it for advertising profiles.
  • Use your data, chats or uploads to train AI models, ours or anyone else’s.

Relationship readings (coming soon)

When relationship readings launch: if you upload a conversation, phone numbers, email addresses, street addresses, card numbers and links are masked before anything is sent to the AI. The original conversation is deleted within 24 hours; we keep only the reading and a summary. Please only upload conversations you are part of.

Children

Halmeoni is not for children under 13 (14 for Korean, 14 for Spanish, 15 for French and 16 for German readings), and we do not knowingly collect their data. Where your local law sets a higher age (16 in Ireland or the Netherlands, 18 in Colombia or India, for example), that age applies. When a birth date shows someone is too young, nothing they entered is stored or sent to the AI, and the same browser is turned away for 24 hours. Purchases and relationship readings, such as compatibility, are for adults only.

For users under 18 we do not offer relationship or compatibility readings, the marriage reading, or picking a day for a confession or a date. The AI is told not to give romance advice or talk about sexual content, alcohol, gambling or dieting, and its answers are checked once more before they are shown. Messages about suicide or self-harm get helplines instead of a reading, and the chat says at the start that Halmeoni is an AI, not a person. We follow child-protection and privacy laws, including Korea’s Personal Information Protection Act, the US Children’s Online Privacy Protection Act (COPPA) and Article 8 of the EU GDPR, and our practices reflect our AI providers’ (Anthropic, OpenAI) guidelines for serving minors.

Your rights

You can ask us for a copy of your data, to correct it, or to delete it. You can delete saved charts and conversations yourself in the app, and delete your account with everything saved for it on the Delete my data page. Contact us and we will respond within 30 days.

Contact

Halmeoni is run by 01 Gongbang (01공방), South Korea. Email: 01gongbang@gmail.com. See also the contact page.