AMENTI.AI

Two trends are colliding. Frontier AI has become cheap enough to give anyone a personal, conversational tutor — and education is racing to adopt it, yet recoiling from chatbots that hallucinate and cite nothing.

The Amenti Interface resolves that tension. History has left us a vast treasure trove of content — the words, ideas, and lives of its greatest minds — and almost all of it sits dormant: scattered, intimidating, unread. The Amenti Interface mines that dormant resource and serves it back to people in a way that has never been attempted — awakening those minds and letting anyone hold a real conversation with them — Plato, Confucius, Frederick Douglass — anchored to the verified, public-domain words they actually wrote.

We have already built the hard part: a curated corpus of ~1,100 historical figures, primary-source "reading rooms," original portrait artwork, an in-character chat layer, the Alpha Helix timeline, Personal and Private Counsel, and Treasure Chess — the whole Amenti Interface, live today. We are raising $500K to convert a working product into a multi-channel business — consumer subscriptions, institutional licensing, and offline distribution (in-flight, cruise, museum) — the last of which monetizes our content at near-100% margin with no AI cost at all.

• Primary sources are inaccessible. The actual writings of history's great minds are intimidating, scattered, and unread — even as demand for authentic, non-AI-slop content rises.
• AI chat is ungrounded. Consumer "character" chatbots invent quotes and facts; institutions won't adopt tools they can't trust.
• Education wants engagement and rigor. Schools, libraries, and lifelong learners want the pull of conversation with the credibility of cited sources. Nothing on the market cleanly delivers both.

The Amenti Interface is a library of legends — and a great deal more. Around the catalog of ~1,100 figures it layers distinct ways to explore, converse with, and play against history:

• Browse & discover — ~1,100 figures across 12 taxonomies, each with original portrait artwork and a curated profile card.
• The Alpha Helix — a one-of-a-kind spiral timeline built as a double helix: one strand of sovereigns (the individuals), one of events, coiling along the years so history reads as cyclical rather than linear. Any node opens straight into a figure — or a conversation. Unlike anything else on the web, it is the gateway into the Amenti Interface at large.
• Reading rooms — verified, public-domain primary-source passages (Plato's Republic, Seneca's On the Shortness of Life, the Analects, Douglass's 1845 Narrative), extracted and provenance-checked.
• Personal Counsel — open any figure's dossier and consult that legend one-on-one, grounded in their own documented words and voice.
• Private Counsel — pose a single question to several legends at once. The Counsel of Amenti convenes a panel, each figure answers in their own voice, and the interface returns a synthesized judgment. A boardroom of history, on demand.

Beyond the library sits the game room — Treasure Chess, a three-dimensional naval chess game in which every piece is a historical figure, rendered as a ship under constellation-and-alchemy sails, with a flagship in the Amenti colors. It is real-time and two-player — a lightweight Cloudflare referee relays moves between players — and clicking any vessel opens that figure's biography, so the board and the library draw on a single roster. Several mechanics are its own: a bounty system that puts a price on enemy commanders; an optional Money Mode that tracks each side's winnings; a full checkers variant on the same fleet; and a persistent trophy room where every captured legend is collected — a personal museum of the minds you've taken. Themed scenarios and a records log round it out. It is the engagement surface that turns a reference library into a place people return to.

Discovery itself is gamified. Each legend carries a quiz; a clean run posts a streak, unlocks the figure's biography, and mints Star Tokens — a soft in-app currency that opens bonus content and threads the player from the question straight into the reading room and the shop. Alongside it, an AI-reading audio player hosts ~30-minute in-voice sessions — Lincoln's fireside, Caesar at the Rubicon, Hannibal across the Alps — drawn from each figure's own writings. These are rendered once and streamed many times, so they carry the offline economics of the static library, not the live-AI cost.

A recurring editorial spine, Atlantica, publishes a fresh "Today's Dispatch" from the world of the legends — serialized, deep-linked back into the catalog, and built to bring readers back on a schedule. Together with the weekly newsletter, it gives the Amenti Interface a heartbeat of new content without new production cost.

• Inference is cheap and falling. A grounded conversation now costs ~1–6¢ depending on model tier — a price that did not exist two years ago.
• AI-in-education is at an inflection. Roughly two-thirds of ed-tech vendors are integrating AI; institutions are budgeting for it now.
• The "verified source" wedge is wide open. As hallucination fatigue grows, provenance becomes a premium feature. Our corpus is the moat.
• Falling LEO-satellite connectivity costs are expanding the in-flight and maritime content markets we can license into.

Figures per industry research (Fortune Business Insights, Grand View, Mordor Intelligence, 2026); ranges reflect differing methodologies and should be source-cited in the final deck.

TAM → SAM → SOM. The global EdTech TAM is ~$200B+. Our serviceable market — AI-assisted humanities and primary-source learning, institutional licensing, and curated offline distribution — is a focused slice where our verified-source corpus is differentiated. Initial obtainable market: English-language consumers, early library/school pilots, and one or two IFE/cruise content deals.

Diversified, stackable revenue streams — consumer to institutional:

• Consumer subscription (freemium). Free tier (ad-supported, capped, lower-cost model) funnels to Pro at $9/mo. The brand and funnel.
• Usage-based / BYO-key. Power users buy metered credits (marked up over cost) or bring their own API key (zero cost to us).
• Institutional licensing — the margin engine. Annual site licenses to libraries, K-12 districts, universities, and museums. Predictable; usage bounded; high margin.
• Offline distribution licensing. The static library needs no network — ideal for in-flight, cruise, hotel, and museum installs, licensed per fleet/venue with no per-use AI cost.
• Advertising & affiliate. Native ads on the free tier plus affiliate links to the full works. Defrays the no-AI-cost browse traffic.
• Commerce — collectibles & books. Limited "Legend Cards" as digital collectibles, plus curated physical books (e.g. a $26 hardcover biography) sold or affiliated from the very page a quiz or reading drops the user onto. Storefront margin on top of near-zero content cost.
• Audio & newsletter. The 30-minute readings seed a free weekly email — one legend, one voice, one half-hour, plus a human contributor essay — that doubles as the cheapest acquisition channel and a sponsorship surface.

On collectibles: the live build issues Legend Cards on-chain. Crypto framing energizes some investors and deters others — the same mechanic works equally as plain digital collectibles or print-on-demand, so emphasis can flex to the room.

Current Claude API rates (per million tokens): Haiku 4.5 $1/$5, Sonnet 4.6 $3/$15, Opus 4.8 $5/$25. Prompt caching cuts repeated input 90%; batch is 50% off.

Margin levers we already control: prompt caching (−90% repeated input), model tiering (free users on Haiku), output caps (output is 5× input), batch (−50%), answer-caching for popular figures, and per-user rate limits. Together these cut effective cost 50–80%.

Model tiering keeps blended AI cost well below the all-Sonnet worst case. Each institutional or IFE/cruise license is near-100% margin (capped or offline usage); heavy-user risk is contained by caps and metered overage. The consumer line is the funnel and brand — licensing and distribution are the profit engine.

A live proof point — the Amenti Studios audio archive

The cost discipline this model assumes is not theoretical. The audio-archiving build already running on the stack — full detail in the companion document Amenti Studios — demonstrates the same dynamic at production scale: because rendered content is archived once and replayed free forever, marginal cost is driven by the volume of distinct content produced, not by audience size or traffic. A daily, voiced, AI-narrated dispatch costs on the order of $0.21 per episode to produce, and a full month of daily episodes plus infrastructure runs roughly $12 — a cost floor that barely moves as the audience scales.

Adapted from Amenti Studios, §Cost projections. Production cost falls from roughly 43% of revenue at the Seedling stage to about 1% at Established — the same asymmetry the consumer/AI model above depends on, already observed in a shipping feature.

Moat: a proprietary, provenance-checked corpus and original artwork that took real curatorial effort; the verified-source trust position; institutional contracts once seeded; and an offline-licensable asset that chat-only competitors cannot easily replicate.

Phased roadmap

• Phase 0 — Harden economics. Auth, caps, caching, tiering, metering. Protects margin from day one.
• Phase 1 — Consumer launch (0–50K). Freemium + ads/affiliate; apply for non-dilutive grants (NEH, Mellon, IMLS, Knight).
• Phase 2 — Institutional (50–100K). Library/school/university pilots → paid licenses; metered credits for power users.
• Phase 3 — Distribution & scale (100K–1M). In-flight/cruise/museum licensing as recurring backbone; enterprise/white-label; international.

The ask & use of funds

Raising $500K on a post-money SAFE to reach metering and distribution live, the first three to five institutional / in-flight licenses, and meaningful consumer traction — proof the licensing backbone converts. With no founder salary and no planned hires through this milestone, burn is concentrated in infrastructure, content, and partnerships, so $500K funds roughly 18 months of runway.

Non-dilutive grants from humanities, education, and library foundations extend runway and lend credibility for the institutional channel, reducing dilution.

[To be completed — founder background, relevant domain/technical experience, advisors. Investors weight this heavily; lead with why this team builds this product.]

The Amenti Interface aims to become the definitive living library — the trusted place where anyone, anywhere, can converse with history grounded in truth.

This product is singularly unique. Nothing else has attempted what it does: awakening the greatest minds in history and deploying them across use cases that are at once educational, entertaining, and utilitarian — a tutor, a companion, a counsel, a game. History's treasure is already paid for and out of copyright; the Amenti Interface is the mechanism that brings it back to life and puts it to work.

The long-term asset is a provenance-checked corpus, a trusted brand, recurring institutional revenue, and a distribution footprint across screens, classrooms, airplanes, and ships. Strategic acquirers and partners — ed-tech platforms, educational publishers, digital-library providers, and in-flight/cruise content aggregators — all value curated content, recurring B2B revenue, and a defensible verified-source position.

A · Per-conversation token model

A typical session: a ~1,500-token persona prompt, 5 user turns (~40 tokens each), 5 assistant replies (~250 tokens each), with history resent each turn → ~10,600 input + 1,250 output tokens.

B · Monthly AI cost at scale

Assumption: 5 conversations / active user / month. Browse-only traffic is ~$0.

C · Sensitivity

Cost scales linearly with engagement: light (2 conv/user/mo) ≈ ×0.4; heavy (15 conv/user/mo) ≈ ×3. The average user is safe; the uncapped heavy user and the anonymous bot are the risk — controlled by caps and auth, not revenue.

D · Cost-control levers

• Prompt caching — −90% on repeated input (the persona prompt + injected sources).
• Model tiering — free users on Haiku (~⅓ the cost of Sonnet).
• Output caps — output is 5× input; a concise mode lowers cost materially.
• Answer-caching — serve pre-generated responses to common questions with no live call.
• Batch API — −50% for non-interactive generation (dispatches, FAQs).

E · Technical architecture

Requests flow through a Cloudflare Worker proxy, which is the natural enforcement point for per-user metering, quotas, model-tier routing, and prompt-cache headers. Static library content (reading rooms, cards, portraits) is served directly and carries no inference cost — the basis of the offline-licensable module. See the product diagram in §III.