privacy-rolywood-v8 · 2026-11-12
SHA256: 7a2ca7c18e64c5476abf67c929f36579c5760440cba20085b14ec919e68b1a55
1. Controller and contact
Rolywood’s controller is José Francisco Mellado Galindo, a self-employed professional. His identifying details and postal address appear in the legal notice. For privacy, rights or review of a decision, contact support@rolywood.com. This policy covers the website and apps according to the features available in each.
We have not appointed a data protection officer. The controller handles these requests directly. We will review whether an appointment is required as the scale and nature of the service change, before public registration. Not having an officer does not limit your rights.
Purchases remain closed during the beta. The beta has no general EU/EEA location restriction. We will explain territorial and purchase conditions before opening payments.
2. Data and sources
Account and access: email, name or alias, language, time zone, preferences and account, session and installation identifiers. Ordinary access uses email codes or the available identity providers. Google, Apple or Discord supply the profile and access data you authorise; signing in through them does not send them your chats.
Content: messages, AI replies, characters, scenes, fictional profiles, files supplied through enabled features and memory data. Social profile: biography, avatar, followers, favourites, reactions, reposts, blocks and reports. A report may include information from another person. Only content you publish through those features is visible to others; talking to a public character does not publish your chat.
Operation: IP address, approximate country, browser, app version, requests, timings, errors, usage and moderation decisions. Support: contact details, message, attachments, reply and delivery. Age: declarations and any date of birth you supplied, available operating-system signals and the minimal evidence described below. System inferences may be wrong; they do not establish facts about your real life.
3. Purposes and legal bases
Account, chat, memory, creations and limits needed to provide the service: contract or steps you request before entering one (GDPR Article 6(1)(b)). Service-related support uses that basis; other enquiries and technical diagnosis rely on our legitimate interest in responding and keeping the service reliable (Article 6(1)(f)).
Fraud prevention, moderation and service protection: legitimate interest (Article 6(1)(f)), with restricted access, minimisation and review. Compliance with legal obligations: Article 6(1)(c). Defence of claims: Article 6(1)(f) and, only where special-category data are necessary for a claim, Article 9(2)(f).
Optional analytics and voluntary beta applications: consent (Article 6(1)(a)). Automatic crisis detection uses separate explicit consent (Articles 6(1)(a) and 9(2)(a)), rather than acceptance of the terms. Document-based age verification relies on performance of the contract (Article 6(1)(b)), because the service is only provided to people aged 18 or over, and its automated decision on Article 22(2)(a), with the safeguards in the human-review section; it does not process biometric data. Writing a message or accepting this policy does not in itself give explicit consent to process special-category data for any purpose. Avoid supplying unnecessary sensitive information about yourself or others.
Manually written memories and memory profiles: your explicit consent (GDPR Articles 6(1)(a) and 9(2)(a)), including sensitive information you choose to add. You can withdraw it for those data by editing or deleting them, without another confirmation. Withdrawal does not affect the lawfulness of earlier processing.
4. Artificial intelligence and memory
Characters and replies are artificial intelligence, not people. They may be wrong and do not replace professional help. To respond, providers receive the message and selected conversation, character, scene and memory context.
Brisa uses OpenAI; Marea uses Anthropic when you select it in an available feature. OpenAI also provides moderation, summaries and embeddings (numerical representations of text). OpenAI also processes titles, memory planning and extraction and relationship consolidation. DeepInfra performed these auxiliary tasks before this provider change; switching providers does not itself delete data processed previously.
Memory distinguishes what you write manually from what Rolywood infers automatically. The first time you add or edit a memory or your memory profile, we ask for a separate, explicit, versioned confirmation: “Your characters will remember what you write here, including sensitive personal data if you include it. You can edit or delete it at any time.” We record the text version, hash and language and the date of your choice. These manual writes do not go through the automatic extractor and remain editable and deletable.
Automatic fact, summary and relationship generators apply a filter to exclude self-harm or suicidal ideation about the real person, or special-category information about you or real third parties (family, friends, neighbours or other people): physical or mental health, sex life or sexual orientation, religion or beliefs, ethnic origin, political opinions, trade-union membership, genetics or biometrics. They may retain non-sensitive facts about those people, such as their name, relationship to you, where they live or shared activities. The filter also aims to exclude sensitive quotations and negations about real people and preserve non-sensitive information in mixed messages. We do not create a health label to apply this rule. Fiction and character attributes retain their narrative context. The filter can make mistakes: it may retain information it should exclude or omit facts you wanted remembered. You can review, correct and delete memories; contact support for other erasure requests.
Your history and retrievable fragments (“moments”) contain what you write, including personal information if you include it. They are literal fragments, not newly inferred profiles. This means the character may still recall sensitive information about you or real third parties through those moments, even when it is not saved in automatic memory. Deleting a message or conversation also removes its retrievable fragments; manually written facts are managed separately. Without crisis detection, a literal fragment may be retrieved again as context; response moderation remains active. You can request correction, export or erasure through support. The memory export is not a complete copy of conversations: request that copy from support. New messages can create new memories.
Rolywood does not sell your conversations, use them for advertising or train its own model on them. OpenAI and Anthropic’s commercial API terms exclude training by default, with separate processing for safety. Other providers’ own uses are explained in the verification and recipient sections.
5. Your choice about crisis detection
Automatic detection looks for signs of self-harm or suicidal thoughts to interrupt an unsuitable response and offer resources. It may infer health data. It is not a diagnosis or an emergency service and cannot guarantee that it will detect a crisis.
The consent choice is presented separately before the first chat, with no preselected boxes. You can accept or decline and withdraw consent in settings. We keep the text version and hash, language, decision date and withdrawal date as evidence of your choice. Without current consent, this automatic detection does not run; ordinary output moderation remains. Crisis help and resources remain available if you decline or withdraw.
When detection runs, records may include category, severity, action, reason, approximate country and pseudonymised references. These are not anonymous data. Withdrawal does not make previous processing unlawful or automatically remove notices already in your history. You can request access, erasure or review through support.
6. Age and account access
Rolywood is for people aged 18 or over. An age declaration, beta admission or Apple Declared Age Range and Google Play Age Signals are not equivalent to documentary verification. Do not email identity documents or photos.
During the beta, we ask all tester accounts for documentary verification before they can use the chat, and we tell you beforehand. If we change who we ask, we will tell you beforehand and update this policy. When the age wall applies to your account, generating replies and accessing future purchases or credits requires a current verification that you are an adult. Without it, you retain access to your own history, export, deletion, support and help. If you have declared that you are a minor, you cannot read history inside the app, but export and deletion remain available.
Minimal evidence contains provider, method, date, a cryptographically protected reference, status, revocation and an expiry after twelve months. We do not store the document or its photographs, the extracted date of birth, the document number or the returned document type. An unsuccessful result is not recorded as proof that you are a minor. Help remains outside the wall.
7. Verification with Didit
When this route is available, Didit Identity Spain, S.L. receives the photographs of your identity document that you take with your camera and the data it reads from it. It acts as Rolywood’s processor to check that the document is genuine and that you are 18 or over. No selfie is requested, there is no liveness check or facial comparison, and no biometric data is processed. Rolywood checks the outcome, the 18+ threshold and the permitted document type. The type is used transiently on the server and is not persisted. The flow accepts national identity cards, passports and residence permits, not driving licences.
Before opening Didit, Rolywood explains how verification works and asks you to confirm that you have read it. Didit explains its processing in its Verification Privacy Notice and End User Terms, linked here, and during verification it may also show you its own screen with those documents. Rejection may prevent generating replies, but you can express your view, challenge it and request effective human intervention, as explained in the next section.
Didit also processes some session data as an independent controller to prevent fraud and protect its service, as explained in its privacy notice. On 19 October 2026 Didit confirmed in writing that Rolywood’s data are excluded from training; that exclusion does not exclude fraud and security.
8. Human review and withdrawing age verification
If you think the verification result is wrong, you can ask for a person to review it from your account verification screen, when available, or through Help and contact (support@rolywood.com). You can give your point of view and challenge the decision. The review is carried out by the controller, who can change the result or allow you a new attempt. We normally reply within five business days and, in any case, within one month. It is not a video call and no appointment is needed. Do not email us documents or photos. We record the minimal review outcome and the case record needed to handle it.
An unsuccessful review does not by itself revoke valid evidence; a confirmed contradictory finding may block and revoke it, subject to human review. A later approval cancels pending reviews to avoid conflicting decisions.
Before completion, cancelling verification and returning to Rolywood stops the pending attempt and requests deletion of the session. After completion, withdrawing verification through the available controls or support revokes the evidence and returns the account to unverified status. A return link does not grant access: the server’s status is checked.
9. Recipients and technical services
OpenAI Ireland Ltd: Brisa, moderation, summaries, embeddings and auxiliary title and memory tasks. Anthropic Ireland: Marea when selected. DeepInfra Inc. received text and context for auxiliary tasks before the switch to OpenAI. These services may process data in the United States and other countries.
Vercel Inc. (Pro): website, API and logs. Databricks/Neon (Launch): database in AWS us-east-1, United States. Upstash: limits, caches and temporary sessions. Inngest Hobby, in the United States, coordinates background tasks. It receives technical identifiers for accounts, conversations, messages, characters, reports, requests, receipts, deliveries and subscriptions, and metadata such as task type, renewal dates, language, counts and result codes. For example, a task name reveals that a billing or deactivation notice has been requested. These data are pseudonymised, not anonymous. New task events, results and errors do not include conversation content, memories, vectors, images, email addresses, adult-content preferences or the inferred reason for a journal entry. Rolywood keeps the necessary data in its own database and loads it inside the task that uses it. Pusher/Bird–MessageBird UK: content and realtime updates, European cluster.
Before this change, Inngest also received conversation and memory content, summaries, vectors, billing amounts and dates, names and email recipients, images or their URLs and vision context, moderation notes and snapshots, adult-content preferences and inferred journal reasons in task events and results. The change does not erase that history or establish its deletion: the 24-hour run listing limit is not a provider data-erasure deadline.
AWS EMEA SARL (S3/CloudFront): existing files and distribution, storage in Ireland and a worldwide CDN. Resend/Plus Five Five Inc.: access and operational emails. Google Cloud EMEA Ltd (Workspace): support and correspondence. Cloudflare: DNS and distribution, not support mailboxes.
PostHog: optional analytics and operational metrics described below, EU Cloud. Sentry/Functional Software, Inc.: errors and performance, EU region; no Session Replay. The app may send technical context and queries; scrubbing is not complete anonymisation.
Didit and its subprocessors provide verification when enabled. According to Didit, AWS EMEA SARL in Ireland performs processing and Google Cloud EMEA provides Maps within the EEA. Transfers and access from other countries are explained in the next section. App stores and identity providers also act as controllers for their own services. Authorities and confidential advisers may receive legally necessary information.
10. Transfers and agreements
Some processing takes place outside the EEA. A European region does not exclude support or subprocessor access from other countries. Verified safeguards for other services may include adequacy decisions and standard contractual clauses; this description does not attribute those safeguards to Inngest Hobby. You may request countries, subprocessors and copies of safeguards from support.
OpenAI’s data processing agreement was signed on 14 October 2026. Verified processing agreements govern instructions to the respective providers; we do not describe Inngest Hobby as a service with a verified processing agreement. You can ask support about the agreements and safeguards for each service.
Didit’s agreement provides for standard contractual clauses and advance notice of new subprocessors. Didit acts as an independent controller for its own fraud and security purposes. Its terms and verification notice explain those uses.
11. Retention and deletion
Account information, content and memory are retained while necessary for the service or until erasure, subject to applicable legal exceptions. Deleting a message or conversation hides it and immediately deletes its retrievable fragments. A daily purge permanently deletes history marked as deleted once thirty days have passed. Separately saved memories can be edited or deleted from memory settings.
Account deletion is scheduled with a thirty-day grace period and can be cancelled during that period. When it runs, the account is anonymised and its access records, memories, manual memory profile, relationship state, summaries, retrievable fragments and consent receipts are deleted. Conversations and creations are hidden. History marked as deleted follows the purge period above; some technical records and external files may require a further process. Support will review remaining data and explain legal exceptions.
A beta application expires within one year. Support, reports and case files are retained while being handled and while necessary to resolve liabilities, with restricted access. Consent receipts are kept while the account exists and deleted during its purge; withdrawing consent records that withdrawal during this period. Crisis notices within history follow that history’s retention period.
Backups and provider logs have their own deletion cycles. You can request the applicable periods and exercise your rights through support, without waiting for a grace period where legally appropriate.
12. Deletion at Didit
When you use verification, Rolywood will ask Didit to delete the session as soon as it records the result, including an unsuccessful result. Cancelled or expired attempts will also be deleted. Failed deletion will be retried and reported to the controller. Neither the document nor its type is stored in your account evidence.
If that deletion fails, Didit deletes the session automatically after one month. The operation removes session data from active systems; daily encrypted backups, stored in Frankfurt (EU), expire after 15 days at most, and the audit log that Didit keeps for 365 days contains no session data.
Deleting a session does not mean every copy is deleted at that same instant.
13. Cookies and diagnosis
You can decline optional browsing analytics and change your choice in Cookie preferences. The preference lasts one year. The web analytics channel may transmit an identifier, IP address and browser; if an identifier is absent, the server may derive one from IP and browser. PostHog’s configuration discards the IP after receiving it; that is not anonymity at source.
Email-code access and technical reliability have operational metrics separate from optional analytics. Limits may use pseudonymised identifiers. Apps store sessions and preferences locally; we do not include advertising networks or cross-service advertising tracking.
To investigate chat speed, we may temporarily enable diagnostics that measure timings, sizes, counts and the model used. These diagnostics exclude text, instructions, names and user, conversation and character identifiers. They use a cryptographically transformed reference to relate measurements from the same request. Other technical logs are described in this policy.
14. Media, payments and store review
The beta’s enabled scope is text. Selecting a voice plays a static CDN sample; it does not record a microphone or clone a voice. Voice generation and new uploads are closed; older files may remain stored. We will provide information before collecting new data if those features open. Closed features do not send new data to ElevenLabs, other media generators or RevenueCat.
Purchases are closed. Stripe integration exists for payments and contractual management where applicable, with its own fraud and compliance purposes; Rolywood does not store full card numbers. Territorial and pre-contractual information will accompany the opening of purchases.
When Apple or Google review accounts are enabled, they will use their own internal credential and private writing scope. They may view the public catalogue, but their creations will not be published or affect public metrics. Their reports will be tagged for human review without automatic action against real accounts. No age evidence will be fabricated for those accounts; access is limited, logged and revocable.
15. Rights, help and complaints
You can request access, correction, erasure, restriction, portability, objection where applicable and withdrawal of consent at support@rolywood.com. The public account-deletion page explains how to request it, including without signing in. Use that contact for a complete copy of conversations or other data; the memory exporter has a narrower scope.
We request only proportionate information to verify identity; do not initially send documents or selfies. Rights requests are normally answered within one month. Where a lawful extension of up to two further months is needed, we explain why within the first month. The five-business-day age-review response period is separate. If an obligation prevents erasure, we explain why and restrict use.
You can complain to the Spanish Data Protection Agency, www.aepd.es, or the competent authority where you live, work or the infringement occurred. Consumer complaints also have the channel stated in the legal notice. Technical safeguards include restricted access and encryption in transit; chat is not end-to-end encrypted against Rolywood and its AI providers.
We publish the version, date and text fingerprint. We will notify you before material changes; updating a policy does not replace required consent. You may request earlier versions from support.