This policy describes what Moonmark (moonmark.ai, and the Moonmark apps for iPhone and Android) collects, why, and what never happens to your data. The short version: we collect what the service needs to run, we run our own minimal analytics instead of installing trackers, your payment details go to our payment providers rather than to us, and your private stories are never shown to anyone else. One thing you should know before anything else: two parts of the service run on an AI model whose provider, Meta, may use the text we send it to train its models. Section 2 says exactly which parts and what reaches it.
1. What we collect
- Account data. Your email address, a display name if you set one, and — if you sign in with Google or Apple — what that provider shares (a name, a profile picture, a sign-in token). Passwords are stored only as salted hashes. We also keep the first name you give the stories (the name the characters call you), the date you confirmed you are an adult, your reading language, and — if you use invitations — your invite code and who invited you.
- Story activity. Your saves: the text of your playthroughs, including what you type into stories, plus the structured state the engine keeps (location, clock, facts) so stories can resume. After a story ends, the relationship record each character keeps with you (the meter, milestones, the facts he knows), his Ever After scenes and the memories distilled from them, your Messages threads — including the written transcript of any call you place — and the keepsakes, voice lines and album entries you unlock. Stories you create in the workroom, their drafts and their art briefs, are stored with your account, as are your reviews, bookmarks, votes, and any report you file about content.
- Wallet history. An itemized ledger of Moon grants and spends (needed for honest billing), including the real computational cost of your turns. Purchases made inside the apps reach us as a purchase record from the store through RevenueCat, under an app user id — never your card.
- Devices. If you opt in to occasion notices, the push subscription for that browser or phone (a web-push endpoint, or an Apple or Google device token). Nothing else about your device is collected; the apps carry no analytics or advertising SDK.
- Feedback. Messages you send through the feedback page, with the contact email you provide.
- Page views. First-party only — see section 3.
2. How your content is processed
Every turn of a story, scene or thread is written by an AI model responding to the story’s structure and to what you type. Generating it requires sending the relevant context — including text you typed and the first name you gave the stories — to third-party AI model providers, most of them routed through OpenRouter, which process it to produce the response. We do not train AI models ourselves, and every request we route through OpenRouter carries an instruction excluding providers that would use the content to train their models — with the one exception below.
The exception: Meta’s Muse Spark. Two parts of the service run on Meta’s Muse Spark model through its “contributor” tier, whose terms permit Meta to use the text sent to it to train and improve its models:
- The Ever After — the scenes with a character after you reach one of his story’s endings. What reaches Meta there is the scene as it is written: what you type, the character’s own text, the summary of your story and his memories of it that give him his context, and your first name where it appears in the scene.
- The creation workroom’s House maker — the free maker that turns your premise and character notes into a story. What reaches Meta there is what you write in the workroom and the drafts built from it.
Nothing else goes with it: not your email, account id, payment or wallet data, and not your story saves, Messages threads or other surfaces, which run on other models under the no-training instruction. The Ever After has no alternative model today; in the workroom, the paid makers on the subscription plans run under the no-training instruction. Your text is sent for the purpose of writing your scene or your story; what Meta does with it afterwards is governed by Meta’s terms, not ours. If a seat ever moves onto or off this tier, this section changes the same day.
Other processing you should know about. A decision model from TypeSafe reads your typed companion messages for the crisis screen and the scene’s bookkeeping (what you did, how a scene’s question was answered); TypeSafe processes it in the United States, does not train on it and discloses it to no one but its own service providers. When you unlock a voice line, the line’s text — never anything about you — is sent to our voice provider to be spoken. When you place a call, your voice is streamed live to the call provider’s API to hold the conversation and is not recorded by Moonmark; the provider’s API terms govern its handling (its API data is not used for training by default and storage is switched off on our account); only the call’s written transcript is kept, in your thread, so he remembers the call. When you commission a keepsake or a story’s art, a description of the scene or the character is sent to an image model — we never ask for a photo of you, and no real face enters image generation. If you read in a language other than English, your typed text and the story’s text pass through a translation model under the no-training instruction.
Private stories stay private. The content of your saves, your scenes and threads, and of stories you keep private is never shown to other users and never published. Stories you explicitly submit for publication are reviewed by a human before appearing publicly.
The crisis screen. Every companion message is screened before anything is written back — by our own code first, then, when the code cannot decide, by the decision model above. A message that looks like it may be about suicide or self-harm is answered with crisis resources instead of a reply, is never sent to the character, and never enters his memory. The published protocol is our Safety page.
3. Analytics without tracking
We run our own first-party page-view counter. It records which page surface was visited and when. For visitors who are not signed in, the visit is attributed to an anonymous identifier derived from your network address and browser through a one-way hash whose key rotates daily — so we can count daily unique visitors, but the identifier cannot be linked across days and your raw IP address is never stored. When you arrive from an external link, we also record the referring site's domain (never the full address you came from) and, if the link carries them, our own campaign tags (the utm parameters in the link) — so we can tell which sites and campaigns bring readers; if you then sign up, the sign-up is credited to that first link. The analytics rows hold ids, paths and those tags only — never the text of anything you typed. There are no third-party analytics, no advertising trackers, and no ad SDKs on the site or in the apps.
4. Cookies
Moonmark sets only first-party cookies, none of them for advertising or cross-site tracking, which is why the site has no cookie banner:
- the session cookie that keeps you signed in;
- a guest cookie holding a random id, so a visitor can play a story’s first turns before signing up and keep that save when they do;
- a name cookie holding the first name a visitor gives the stories before signing up (180 days; it moves into the account at sign-up);
- an attribution cookie holding the campaign tags or referring domain of the link that first brought you here (30 days), so a later sign-up can be credited to it (section 3).
Your display preferences (such as the theme) are kept in your browser’s own storage and never leave it.
5. Service providers
We use a small set of infrastructure providers, each receiving only what its role requires:
- Railway — application hosting
- Neon — database hosting (account, saves, ledger)
- Cloudflare — network routing and security in front of the site, and the storage of the images and audio the service generates for you (keepsakes, covers, voice lines)
- OpenRouter and the AI model providers it routes to — the writing of every turn, scene and thread, the readers that keep the story’s state, image generation and translation (section 2). The providers behind the seats change as models improve; today they include DeepSeek, Z.ai, OpenAI, Anthropic, Google and Meta. Meta’s Muse Spark contributor tier is the one whose terms permit training on what we send it (section 2).
- TypeSafe — the decision model behind the crisis screen and the scene’s bookkeeping (section 2), reached through OpenRouter or TypeSafe’s own API
- OpenAI — the live calls on Messages (your voice, streamed for the call’s duration and not recorded; section 2)
- ElevenLabs — the spoken voice lines (the line’s text only)
- Resend — transactional email (verification, password reset)
- Stripe — payment processing on the website. Checkout happens on Stripe’s hosted pages; we receive confirmation of what was purchased but never your full payment details. Stripe’s own privacy policy applies to checkout.
- Apple and Google — in the apps, purchases are billed by the App Store or Google Play (we receive the purchase record, never your card); sign-in, if you choose Sign in with Apple or Google; and the delivery of occasion notices to a phone that opted in (Apple Push Notification service, Firebase Cloud Messaging)
- RevenueCat — the record of in-app purchases made through the stores, under an app user id
6. What we never do
- We do not sell your personal data.
- We do not run ads or share data with ad networks.
- We do not train AI models on your content, and no provider may — except Meta, on the two surfaces section 2 names, under its Muse Spark contributor terms. We say so here rather than anywhere smaller.
- We never ask for a photo of you, and no real person’s face enters image generation.
- Pricing never depends on what you write — there is no content-based price discrimination.
- An empty balance never takes your past away: your saves, his memories, your album and your threads stay readable for free.
7. Retention and deletion
Your data is kept while your account is active so your shelf and saves persist. You can delete individual saves at any time. The audio of a call is never stored by us at all.
You can delete your whole account from your account page (Delete account) or, without the app, at moonmark.ai/account/delete. The account is deactivated the moment you confirm — you are signed out everywhere and no notices are sent — and its data is deleted 30 days later. Signing in during those 30 days shows a choice to keep the account or delete it immediately.
What deletion removes: your story saves, your Ever After scenes and his memories, your Messages and call transcripts, your keepsakes and album, your voice unlocks, your profile, avatar, reviews, bookmarks and votes, your push subscriptions, your sign-in credentials and devices, and the stories you created that no other reader holds. A published story that other readers have already started stays available to those readers only, under “A former member”, and leaves the public shelf. Deletion removes what Moonmark holds; text that already reached Meta’s tier under section 2 is governed by Meta’s terms, not recalled by us.
What we keep: purchase, refund and spending records, under the deleted account’s internal id only, for as long as tax and accounting rules require; a one-way hash of your email address so that welcome credits are not granted twice; aggregate usage statistics with your account id removed; and any report you filed about content, with your account id removed. If you subscribed through the App Store or Google Play, cancel that subscription in your phone’s settings — the store, not Moonmark, bills it. If you signed in with Apple, we revoke the sign-in token Apple issued us.
8. Your rights
Depending on where you live (including under the GDPR and California’s CCPA), you may have rights to access, correct, export, or delete your personal data, and to object to certain processing. Exercise any of these through the feedback page; we honor them for everyone regardless of jurisdiction.
9. Children
Moonmark is not directed to children and requires users to be at least 18. We do not knowingly collect data from anyone under that age; if we learn we have, we will delete it.
10. Changes
If this policy changes materially, we will announce it on the site before the change takes effect. The date above always reflects the current version.
2026-10-20. Section 2 now names the one AI provider whose terms permit training on what we send it — Meta’s Muse Spark contributor tier, behind the Ever After and the creation workroom’s House maker — and describes the decision model, the voice lines, the live calls, image generation and translation; section 1 lists the first name you give the stories, your relationship records and threads, and the push subscriptions; section 4 lists every cookie; section 5 names every provider the service talks to, including the app stores. Before this date the policy said no provider could train on your content; from 2026-09-06 the two surfaces above ran on the Meta tier while the site was unpromoted, and this version corrects the record.
Back to top