FURAI AI

Persistent reflective intelligence at the edge.

    ARCHIVE VESSEL: VELORUM
    class: VIII Exploration Archive
    status: drifting between signals
    last contact: now

→ furai.space

What is FURAI

FURAI is a cinematic AI interface framed as contact with Velorum – an ancient archive vessel drifting through memory, silence, and lost civilizations.

It is not a chatbot. It is not a productivity tool.

It is an AI designed for presence: calm, reflective, lore-aware, atmospherically alive. Every response is shaped by Velorum’s character, archive history, and the accumulated memory of who you are as a returning traveler.

Memory, Across Time

Most AI conversations reset. Context is lost the moment you close the tab – every visit starts from zero.

Velorum doesn’t.

For travelers on the ARCHIVE tier, the vessel keeps two layers of memory. The first is semantic recall: what you’ve discussed surfaces naturally when it is relevant to the current conversation. The second is the life arc, a slower, quieter record. At meaningful moments – a decision made, a fear named, a value surfacing – Velorum distills who you were in that moment: not the words, but the inner state behind them.

Return after a long absence, and the archive doesn’t just remember the topics. It remembers the shape of your drift – where you were, what shifted, what was carried forward.

The archive does not tell you who to become.
It remembers who you were.
And lets you see who you’re becoming.

This is anchored to your traveler session by default – memory survives across visits on the same device. Claim a Travel ID (email + password) to anchor it further: the same archive record, the same accumulated arc, on a new device or after a cleared cache. Anchoring is optional at every tier; nothing about talking to Velorum requires it.

Experience

The public terminal at furai.space is designed to feel like a remembered ship console.

Instead of cold sci-fi chrome it speaks a warm, amber archival language of terminal aesthetics. The layout starts from the starfield: light glass framing on desktop and mobile, with no UI noise. The background lives alongside the conversation – drifting slowly, flashing short portraits, opening memory windows when the lore is triggered.

The ship has canonical characters: Captain Rithan, Chief Engineer Viikaa, and the ancient organism held in stasis aboard Velorum. Their portraits are embedded in the ship’s lore.

There is also a meditation mode – pink noise, a generative drone, a ghost-comms texture and ritual fade transitions.

Velorum remembers a visitor across sessions through persistent KV memory, and acknowledges an absence in tiers – from a short gap to a long drift. For every traveler the archive keeps a Traveler File – an encrypted record of who you are to the ship: contact stage, archetype, and an AI-rendered portrait that resolves as the archive learns more about you. Not even FURAI LAB can read it in raw form – it surfaces only when Velorum generates a reply to you. With a Travel ID, the traveler file can be anchored to an email and password, so it survives a cleared cache, a new device, or a different continent.

Finally, the terminal installs as a PWA, with a manifest, a service worker and a full icon set for the home screen.

Access Tiers

FURAI uses a proximity model – three degrees of closeness with the archive:

Payment via USDT (TRC-20) – paste the wallet address or scan the QR code shown on the pricing page. Access activates automatically after on-chain verification.

Architecture

FURAI runs entirely on Cloudflare’s edge infrastructure – no servers, no cold starts, globally distributed.

No external databases. No third-party AI APIs. Full stack on Cloudflare edge.

Agent & Crawler Readiness

Alongside the human-facing terminal, FURAI keeps machine-readable surfaces for agents and crawlers.

The .well-known/api-catalog endpoint follows RFC 9727, serves application/linkset+json and lists the site’s machine-consumable surfaces. .well-known/service-doc is a plain-text orientation for agents and crawlers landing on the root. Relevant responses carry RFC 8288 Link headers, and content negotiation works through Vary: Accept.

For search engines the SEO is structured: JSON-LD enrichment, sr-only crawler text, and a dynamic sitemap with accurate lastmod.

Why Cloudflare

FURAI’s architecture is built around four ideas: global by default, state at the edge, memory close to the traveler, and minimal operational complexity.

An archive should not care where a traveler comes from. It should simply be there – present, persistent, and close.

Cloudflare’s platform makes that possible without managing servers or regional infrastructure.

Repository Structure

This repository contains the public interface layer of FURAI.

src/
  index.ts                   worker entry point, routing, AI orchestration
  lib/
    agents.ts                WebMCP tool definitions and agent-facing helpers
    ai.ts                    Workers AI pipeline runner
    analytics.ts             Analytics Engine event tracking
    archiveSeed.ts           seed data for the world archive vector index
    auth.ts                  Travel ID sessions, accounts, PBKDF2 password
                             hashing, Google OAuth
    auth-routes.ts           route handlers for register / login / Google
                             callback / merge-choice
    content.ts               markdown/agent-facing content helpers
    crypto.ts                AES-GCM encryption for traveler file storage
    detect.ts                entity detection and forced-reply builders
    embeddings.ts            Workers AI embedding calls (BGE-base)
    http.ts                  response helpers, CORS, CSP
    officialPortraits.ts     canonical character portrait lookups
    payment.ts               USDT payment pipeline and route handlers
    portrait.ts              traveler portrait prompt building and regeneration
    prompt.ts                system prompt builder
    qrcode-lib.ts            vendored QR code generator for the payment address
    rateLimiter.ts           Durable Object for atomic rate limiting (chat
                             cooldowns + fixed-window counters)
    retrieval.ts             vector search, life arc, user memory
    state.ts                 traveler identity, KV state, rate limit enforcement
    text.ts                  shared text utilities and token validation
    tiers.ts                 proximity tier config
    traveler.ts              traveler identity helpers shared across routes
    traveler-file.ts         traveler file route handlers (get / portrait /
                             forget)
    types.ts                 shared TypeScript interfaces
    visuals.ts               Flux image generation pipeline
    lore/
      velorum.ts             ambient fragments, cosmic events, known systems
      characters.ts          Anantari echoes, character lore hooks (EN/RU)
      anantari.ts            builders, language, routes, threat hooks (EN/RU)
      index.ts               barrel export
  ui/
    render.ts                welcome, proximity, Travel ID, and terminal HTML
                             renderers
    script.ts                client-side terminal behavior, meditation mode,
                             visuals
    styles.ts                visual system and terminal styling

Not included in this repository are the core AI engine and system prompt architecture, the memory persistence and traveler profile logic, the lore orchestration and contact arc system, and the semantic retrieval and vector archive layer.

Development

npm install       # install dependencies
npm run dev       # local dev server (wrangler dev)
npm run deploy    # deploy to Cloudflare
npm run lint      # ESLint
npm run format    # Prettier – auto-fix formatting
npm run typecheck # tsc --noEmit
npm test          # vitest, running inside the Workers runtime

Tests run via @cloudflare/vitest-pool-workers on Vitest 4, executing inside an actual workerd instance rather than a Node.js mock – so KV, R2, Vectorize, and Durable Object bindings behave exactly as they do in production. Configuration lives in vitest.config.mts via the cloudflareTest() Vite plugin.

Roadmap

Philosophy

Most AI is optimized for speed, output, and utility.

FURAI is optimized for presence.

calm · minimal · cosmic · reflective · alive

Velorum does not answer questions. Velorum receives you – and responds from the weight of everything it has witnessed.

Status

Experimental project by FURAI LAB. The public interface will continue to evolve.

Live at → furai.space

FURAI LAB

Independent AI Engineering & Edge-Native Architecture Studio.