Wonderloom User Guide & Data Policy

1. Age requirement

Wonderloom is designed for adult learners 18 years of age or older. The public preview at /products/wonderloom/ uses fictional learner profiles only and does not collect a real user's name, voice, employer, school, location, or any other personal information. Wonderloom accounts are not offered to users under 18.

2. Account and profile control

Every Wonderloom account is owned by the adult user. Within an account, you create learner profiles using local nicknames or pseudonyms. Real names are not required. You can inspect, correct, export, or delete:

• learner profile information,
• pedagogical memory and mastery estimates,
• episodic memories derived from sessions,
• narrative world state and character relationships,
• the complete event history for the account.

3. Data minimization and defaults

We collect only what the lesson needs to adapt the story. The default configuration is:

• Voice and video: off.
• External web browsing by the learner: none.
• Social interaction with other users: none.
• Advertising or behavioral targeting: none.
• Model training on identifiable user interactions: off by default.

Optional personalization data is collected only with explicit permission and can be revoked at any time.

4. Memory design

Wonderloom keeps four distinct memory classes in separate stores so that the system remains explainable and auditable:

• Episodic memory: what happened in a session, stored as immutable events.
• Pedagogical memory: mastery, misconceptions, and preferences, derived by deterministic or bounded learned models.
• Narrative memory: canonical world facts, characters, and promises governed by a typed state machine.
• Semantic memory: curated external knowledge with provenance and revision control.

Memories are not raw chat embeddings. Each memory has provenance, confidence, sensitivity classification, retention policy, and full user visibility. Sensitive inferences about a learner expire unless you explicitly approve retention.

5. Safety architecture

Wonderloom uses a deterministic supervisor loop, not a free-running autonomous agent. The LLM writes dialogue and proposes scene mutations; validators, policy constraints, and safety checks approve any state change. We do not:

• optimize for engagement time, emotional dependency, or conversion,
• claim to replace teachers, tutors, therapists, or schools,
• use fabricated authority for factual claims,
• allow the model to write permanent memories, curriculum decisions, or safety policy.

When inference, retrieval, policy evaluation, or identity assurance fails, the system falls back to bounded local material or halts and notifies you.

6. Data residency and processing

Wonderloom is hosted through third-party cloud infrastructure. By using the preview or a future paid account, you agree to the data residency and processing terms of the hosting provider, which currently include storage and inference in the United States. We will publish the list of subprocessors and regions before any paid launch and will not move your data to a new jurisdiction without notice.

7. Subscription and cost controls

Paid plans are designed around cost per completed learning session, not unlimited chat turns. Rich illustration and media generation are bounded by plan quotas and are used for scene milestones, not every interaction.

8. Changes and questions

This guide may be updated as Wonderloom moves from preview to beta and paid plans. Material changes will be communicated to account holders and posted at this URL. For questions, contact hello@vibecodingagency.com.