Aurea uses three AI providers to deliver your experience. Each provider handles a specific layer of the game and receives only the data it needs to perform its role.
All story generation, game narration, player choices, scene transitions, and story arc planning are handled by Anthropic's Claude models. These are the three AI layers that make up Aurea's "AI Dungeon Master." Your in-game text input and adventure context are sent to Anthropic's API via our proxy server. Anthropic's privacy policy: anthropic.com/privacy
Origami-style character figurines and hand-drawn adventure maps are generated by image models hosted on OpenAI and Replicate. Your character description and map parameters are sent to these providers only at image-generation time. Generated images are permanently stored in Firebase Storage — no temporary CDN URLs are persisted. OpenAI privacy policy: openai.com/privacy. Replicate privacy policy: replicate.com/privacy
When voice narration is enabled, the Narrator's output text is sent to Inworld's TTS service to be spoken aloud. No player input is sent to Inworld — only the AI-generated story text that will be narrated. Inworld privacy policy: inworld.ai/privacy-policy
Aurea is built from two layers of content — content our team wrote, and content generated live by AI for each player's unique session.
| Content | Origin |
|---|---|
| Story narration, scene descriptions, NPC dialogue, player choices | AI Generated per session by Anthropic (Narrator layer) |
| Story arc planning, dramatic tension structure, plot beats | AI Generated per adventure by Anthropic (Architect layer) |
| Scene transitions, pacing decisions, NPC behavior flags | AI Generated per turn by Anthropic (Director layer) |
| Character portrait images | AI Generated once per character by OpenAI / Replicate |
| Adventure map images | AI Generated once per adventure by OpenAI / Replicate |
| Rules engine (D&D 5E mechanics, dice, ability checks, combat) | Human Hand-authored by the Aurea team; AI cannot override |
| World lore, world definitions, adventure frameworks | Human Authored by Patrick Leung with the assistance of AI writing tools (Claude.ai). The in-app narrator runs against this authored canon at runtime. |
| Character creation options (races, classes, abilities) | Human Defined by the Aurea team per SRD 5.1 |
| App interface, navigation, and feature logic | Human Built by the Aurea engineering team |
The D&D 5E rules engine operates independently of the AI. Combat outcomes, ability checks, and mechanical decisions are computed deterministically — the AI Narrator cannot hallucinate mechanical facts (e.g., inventing a spell that doesn't exist, or changing your hit points without a valid mechanical trigger). This structural separation between rules and narrative prevents a whole class of AI confabulation.
The Narrator layer operates under a 24-section rulebook that our team maintains. This rulebook includes explicit guards against:
Aurea is rated 12+ on the Apple App Store (PEGI 12 / ESRB Teen equivalent on Google Play). The rating reflects the presence of fantasy combat — described in text, not depicted in graphic visuals — consistent with tabletop RPG content of the same age bracket.
Apple's Ask to Buy feature applies to any in-app purchases made on child accounts. Aurea collects a coarse age range at sign-in to provide age-appropriate experiences; it does not collect date of birth. Additional parental controls are enforced at the platform level by Apple and Google.
If you encounter AI-generated content that doesn't meet our standards — content that seems harmful, inappropriate for the age rating, or factually misleading in a real-world way — you can report it in two ways:
Open Settings → Report from inside the app. The report captures the session context automatically so we can investigate the specific output you encountered. This is the fastest path to getting the content reviewed.
If you prefer, or if the in-app flow isn't accessible, email us directly at tsongkapa@gmail.com with a description of what you saw. Include the character name and world if you remember them — it helps us locate the session.
We review all reports and use them to improve the Narrator safety guidelines. Egregious outputs can be escalated to Anthropic's trust and safety team under their existing usage policies.
Here's exactly what user input goes where, and what we store.
Anthropic, OpenAI, Replicate, and Inworld each have their own data retention policies. We send data to these providers under their standard API terms. We do not have a separate data processing agreement that extends beyond their standard API terms. Refer to each provider's privacy policy (linked in the AI Providers section above) for details on how they handle API request data.