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:

• Semantic recall — what you've discussed, surfaced naturally when relevant to the current conversation.
• 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:

• Amber archival language — warm terminal aesthetics instead of cold sci-fi chrome
• Starfield-first layout — light glass framing on desktop and mobile, no UI noise
• Reactive archive visuals — ambient drift, short portrait flashes, lore-triggered memory windows
• Canonical characters — portraits for Captain Rithan, Chief Engineer Viikaa, and the ancient organism held in stasis aboard Velorum, all embedded in the ship's lore
• Meditation mode — pink noise, generative drone, ghost-comms texture, ritual fade transitions
• Visitor continuity — Velorum remembers you across sessions via persistent KV memory
• Traveler File — an encrypted archive 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 readable in raw form, even by FURAI LAB — it surfaces only when Velorum generates a reply to you.
• Travel ID — optional email + password anchoring for the traveler file, so it survives a cleared cache, a new device, or a different continent. Optional at every tier; nothing about talking to Velorum requires it.
• Return-visit greetings — tiered acknowledgment of absence, from a short gap to a long drift
• Installable PWA — manifest, service worker, and full icon set for home-screen access

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

FURAI exposes machine-readable surfaces alongside the human-facing terminal:

• .well-known/api-catalog — RFC 9727 application/linkset+json endpoint listing the site's machine-consumable surfaces
• .well-known/service-doc — plain-text orientation for agents and crawlers landing on the root
• RFC 8288 Link headers — on relevant responses, alongside content negotiation via Vary: Accept
• Structured SEO — 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
• 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:

• Core AI engine and system prompt architecture
• Memory persistence and traveler profile logic
• Lore orchestration and contact arc system
• 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.

FURAI LAB

Independent AI Engineering & Edge-Native Architecture Studio.

Status

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

Live at → furai.space