Canonical, not generic
Every answer is grounded in the PÚNYCODEX canonical sources — LSJ, Beekes, Faulkner, KTU, the Avesta, the Kojiki, and 170+ scholarly references — not scraped forum posts or Wikipedia summaries.
A specialized intelligence for mythological names, Unicode restorations, original scripts, pronunciation, etymology, punycode, and homograph safety — grounded in the PÚNYCODEX canon, not the generic web.
The Difference
General-purpose models treat names as strings. The Oracle treats them as living philological objects: every acute accent, every macron, every hieroglyph, every punycode mutation, every citation. It knows when Apóllōn is not apollon, and it knows why that distinction matters to scholars, security teams, linguists, and creators.
Capabilities
General-purpose models know names as words. The Oracle knows them as restorations — with stress, length, script, source, and safety context.
Every answer is grounded in the PÚNYCODEX canonical sources — LSJ, Beekes, Faulkner, KTU, the Avesta, the Kojiki, and 170+ scholarly references — not scraped forum posts or Wikipedia summaries.
The Oracle reads, writes, and converts between Unicode restorations and their punycode domain forms. It understands why Apóllōn is not the same as apollon.
It detects mixed-script deception, normalization tricks, invisible characters, and lookalike domains — and refuses to help with spoofing, impersonation, or trademark abuse.
Ask for IPA, phoneme breakdowns, or the original hieroglyphic, cuneiform, runic, Greek, Devanagari, or CJK form — with provenance for every step.
It maps planetary, elemental, alchemical, and tarot correspondences, and draws careful analogies to physics, biology, neuroscience, and philosophy — always labeling confidence.
It can trace how chaos echoes the Big Bang, how Apollo and Ra speak to light, and how biblical figures connect to older mythic patterns — without flattening their differences.
Live Preview
Tap a question to see how the Oracle reasons across Unicode, script, etymology, and safety.
Select a question above to consult the Oracle.
Training Data
The model is being built from the same canonical sources that power the website, API, browser extension, and mobile app.
Corpus Categories
Use Cases
"What is the scholarly Unicode form of apollon, and why is it dual-tier?"
"Is this punycode domain a canonical PÚNYCODEX name or a spoof?"
"How do I pronounce Hádēs in Attic Greek, and what does each sound signify?"
"Show me the hieroglyphic or cuneiform original for this name."
"How does the Babylonian Tiamat relate to the Greek Kháos and modern cosmology?"
"What are the planetary, elemental, and alchemical associations of Venus/Aphrodítē?"
Roadmap
Each phase is gated by data quality, safety review, and scholarly sign-off.
894 lexicon entries, original-script provenance, source citations, and etymological threads structured for model training.
Dialogue, mythology synthesis, scientific analogies, pronunciation, and homograph-safety examples generated deterministically.
University reviewers validate outputs per pantheon, flag hallucinations, and approve canonical answer templates.
Refusal training for impersonation, trademark abuse, and harmful homograph generation; adversarial red-teaming.
Full fine-tuning on dedicated hardware, followed by API access, browser integration, and a waitlist release.
Trust
The Oracle is designed to deepen understanding, not to enable abuse. Every output carries provenance and safety context.
Every claim is traceable to LSJ, Beekes, Faulkner, KTU, the Avesta, the Kojiki, or another authoritative source.
The Oracle refuses requests that would help spoof brands, deceive users, or generate infringing domain registrations.
Output is framed as scholarly restoration and education, not commercial endorsement of any third-party trademark.
Flagship answers and high-risk topics are reviewed by the scholarly board before becoming model training targets.
We are preparing the model for training on dedicated hardware. If you want early access, research collaboration, or API notifications, reach out.