The Architecture of Living Memory.
Not an archive of the past. A simulation of personality. We preserve a person’s authentic character — not their archived data — and prove every response cryptographically.
Built by independent inventors. Patent priority date August 14, 2025 — three months before the public conversation about this category began.
Every year humanity loses trillions of hours of accumulated wisdom.
When a person passes, their stories, reasoning, humor, and voice disappear with them. We are the first generation in history with the technology to choose otherwise — yet none of today’s tools actually does it.
The market wants a system that can answer a question the person was never asked, in their own way of thinking, with verifiable authenticity. Until now, nothing has.
We made a foundation model speak as one specific person — without training a single new model.
The patent introduces three primitives. Each is a runnable Python package, each is testable in isolation, and each is independently patent-claimable. Together they constitute a new form of digital legacy.
The contrast is stark. The same Claude Opus 4.7 that produces plain answers when called directly produces deeply in-character answers when configured with a Personality Vector. See the side-by-side comparison →
How a question becomes an authentic answer.
Every interaction with a capsule traverses four stages. Each stage is implemented in code, each produces a deterministic hash that is included in the response provenance, and each is independently testable.
The user query is matched against the capsule’s episodic memories using IDF-weighted lexical overlap with tag and recency boosts. Returns top-K.
The Behavioral Prompting Engine assembles a system prompt from the Personality Vector’s structured fields. Memories are scaffolding, not content.
Claude Opus 4.7 (or any compatible model) extrapolates a response. The model has never been trained on the creator’s data — it borrows the cognitive style.
The response is hashed, the manifest is built (PV signature + meta-prompt hash + response hash + memory IDs), signed Ed25519, and returned.
Three independent layers of protection — and a public freedom-to-operate analysis.
We address Microsoft’s 2017 patent on “creating a conversational chatbot of a specific person” explicitly. Our six deltas — Personality Vector as structured intermediate, Behavioral Prompting (extrapolation vs retrieval), Federated Learning, PEFT/LoRA, multi-modal biometric with on-device discard, Authenticity Seal + Digital Will — are each mapped to specific source files in the repository.
Read the design-around analysis →One platform. Four revenue streams. One strategic outcome.
The same patented architecture serves four distinct buyers, each with a different willingness-to-pay, sales motion, and regulatory posture. The strategic outcome is a license to a hyperscaler — Apple Memories, Google Photos, or Meta — when the category matures.
Adults leaving a “living legacy” for grandchildren. Bereaved families reconstructing from existing data via Fast Capture Mode.
Capsules of retiring CTOs and chief engineers. Preserve reasoning before retirement. No grief-tech legal risk.
Living archives of cultural figures, last native speakers of endangered languages, master artisans.
Hospice and grief-counseling partnerships. FDA digital therapeutics track. Year-3 line; gated behind clinical RCT.
We hold the priority date the rest of the field is now chasing.
Claude Opus 4.7 and GPT-4o deliver sustained behavioral consistency at price points under $10/M tokens. Voice cloning is consumer-grade. 3D facial scanning is on every iPhone.
A high-profile public statement in late 2025 catalyzed the category. HereAfter AI scaled. StoryFile signed DARPA. Our PPA was already filed.
Article 50 (in force August 2, 2026) mandates provenance for AI-generated personal content. Privacy-by-design and the Authenticity Seal are not features — they are regulatory necessities.
Two inventors. One mission.
Rodion Sorokin and Serhii Nikolaichuk are independent inventors working at the intersection of AI ethics, cognitive science, the philosophy of technology, and Human-Computer Interaction. AI Time Capsule is one of seven USPTO filings in their portfolio — a coherent research programme, not a series of isolated products.
Read founder biographiesFirst-named inventor on US Application #19/319,614. Cross-domain expertise in AI, cryptography, and digital provenance. Lead author on 11 academic manuscripts founding the field of Computational Personality Science.
s.rodion2009@gmail.comFounder & CEO of The Capital Index (Austin, TX), a deep-tech venture studio. Holds seven USPTO filings across Entropy Protocol, AI Ethical Blackbox, Symbiotic Assistant, SPYNO, MOIRE, Project Agora, and AI Time Capsule.
nikolaichuk.s.f@gmail.com“Rule one: never confuse activity with progress.”
em_003.Talk to Will. Verify his seal. Compare him against vanilla Claude. See for yourself why this is not another chatbot.