Every claim · in writing
Astrael is an AI company. Below is exactly what models we use, what corpus we reference, where your data flows, and how you control it. Every line is auditable.
Astrael-1 is our proprietary AI system. Five layers: server-side divination compute (real ephemeris, solar-term calendar, deck logic, hexagram casts), a retrieval corpus (primary-source canon + modern psychology), a prompt-engineering layer, a dual-engine model layer — the Claude family as our partner backbone plus our own Astrael-1 LoRA fine-tunes, trained on leading open-source foundation models with our bilingual divination corpus — and a router that dispatches each task to whichever engine serves it best. Astrael-1 leads on Chinese cultural-language tasks; Claude leads on long-form synthesis. Both can be active in a single session.
| Task | Astrael-1 mode | Latency |
|---|---|---|
| Long-form reading interpretation | Astrael-1 standard | 1–3 s/token |
| Deep cross-tradition synthesis | Astrael-1 deep | 2–5 s/token |
| Quick draws · daily horoscope · titles | Astrael-1 fast | <200 ms/token |
| Embeddings (memory + recall) | Astrael-1 retrieval | ~50 ms/req |
| Real-time voice readings | Astrael-1 voice | <400 ms first audio |
Our partner LLM provider does not train its general models on API customer content under standard enterprise terms — your inputs are used to answer you, not to train anyone's models. Our own Astrael-1 fine-tunes train only on explicitly-consented data (corpus + our in-house expert annotations — we never silently train on user conversations). A full sub-processor list is available on legal request.
Our system prompts and retrieval corpus reference the following primary sources and modern psychology texts (within fair-use scope):
- ·Waite-Smith Rider tarot deck (1909, public domain)
- ·Aleister Crowley · The Book of Thoth (1944) · scholarly reference
- ·Claudius Ptolemy · Tetrabiblos (~150 CE, public domain)
- ·Xu Ziping (徐子平) · 渊海子平 (~960 CE, public domain)
- ·《滴天髓》《穷通宝鉴》《紫微斗数全书》 · Chinese metaphysics canon, public domain
- ·Wang Bi (王弼) · I Ching annotations (~250 CE, public domain)
- ·Richard Wilhelm · I Ching translation (1923, public domain in most jurisdictions)
- ·Carl Jung · selected essays on archetypes & synchronicity · scholarly reference
- ·Sigmund Freud · selected essays on the unconscious · scholarly reference
These materials are not exclusive training data and are not part of model weights — they appear in system prompts and RAG retrieval corpus.
- ·Astronomical ephemerides — astronomy-engine library, arc-second precision
- ·Chinese solar-term calendar — lunar-typescript, accurate back to 1900 CE
- ·Tarot deck draw — Fisher-Yates shuffle with crypto.randomBytes
- ·I Ching hexagram cast — 50-yarrow-stalk simulation, deterministic from seed
These computations run server-side. The LLM does not compute them — only interprets them. This is what prevents the LLM from hallucinating planetary positions or solar-term dates.
- Access — CA CCPA/CPRA & EU GDPR · download all your data from Settings · Privacy
- Deletion — Settings · Privacy · Delete my account · permanently erased within 30 days
- Do not sell — we never sell your data to anyone. Ever.
- File a request · email privacy@astrael.ai · response within 30 days
Astrael is for users 13 years and older (US COPPA) / 16 and older in the EU (GDPR; some member states require 18). Our service is not a substitute for medical, financial, legal, or psychological advice. If you are in crisis, please reach a local hotline — US 988 (24/7), local listings in our Help center.