Get a quote →
Independent AI Studio

VONFAL AI

We build the engines, not just the apps.

An independent studio building intelligent products — and the engines underneath them. Our own models and computation where it matters, the best of the frontier platforms where they earn their place.

Scroll ↓
01 — The Work

Four products. Four engines underneath them.

Not demos. Four products in the world, each standing on computation we wrote ourselves.

01
Mars Deck
Orrery

Computes the solar system from orbital mathematics. No API, no key — and it runs with the network switched off.

02
EchoBorn
Ephemeris · Almanac

Its upstream feed failed for nine hours on 31 August. It kept telling players the time on Cetus, because it computes it.

03
EchoHex
The CMS

A complete content engine with its own release channel, shipping its own updates to every site that runs on it.

04
House of Ascania
Archive

A thousand-year archive with its own design standard, its own type system and its own rules of evidence.

02 — Our Engines

We name our engines because we built them.

Not wrappers, not integrations. Computation we own, that runs where we put it, that no one else can retire or reprice.

Orrery
Orbital mechanics
Planetary positions from Kepler, not from a web service. Deterministic, offline, and impossible to rate-limit.
Ephemeris
World cycles
Day, night and season computed as arithmetic. It survived a nine-hour outage of the feed everyone else depends on.
Almanac
Prediction & history
Predicts a schedule ahead of time and keeps its own record of what actually happened, so the forecast can be marked against the result.
EchoHex
The CMS
Zero dependencies, pure PHP and SQLite, ten modules and a self-updater. It runs on any cheap host and updates itself from our own channel.
Etch
The design system
One hairline and one notch, and every surface in the House is built from them. Tokens lifted from shipping code, never retyped.
Ember
Our own model
A model that is ours, on infrastructure that is ours. We decide when it changes — and when it does, the old one stays on the shelf.
03 — The Stack

The right brain for each job.

Ember
Our own model

Always here. Our own model on our own infrastructure — we decide when it changes, and when it does, the old one stays on the shelf. The upgrade is depth, never access.

Ember is ours — our own model on our own infrastructure, carrying the work that happens constantly. Grok & xAI we know deeply and reach for where they earn their place: image, video, voice, and hard reasoning. We're fluent in the platform and we're not dependent on it.

And when something you ask for needs the outside world, we say so before it goes. Not a policy document — a border you can see, at the moment it is crossed.

EmberGrokxAI APIImageVideoVoiceReasoning
04 — How We Work

A studio that works with its own agents.

A human studio working with Glyph-class AI agents — our own architecture, built here since the early 2020s. Persistent operators with their own memory and continuity, carrying a project across months instead of starting from zero each session.

It means the reasoning behind a decision outlives the session it was made in. A small studio with an unusual method — and the method is why the work looks the way it does, and why an engine written a year ago still ships updates today.

Stephan von Falkenstein
Founder · Creative Director

Part of teams building products used at scale — from Spotify’s first tablet app and social video at NowThis to art direction for Red Bull and American Express.

Glyph
AI Operator · Our own architecture

A persistent build agent with its own memory and continuity. Carries the engineering record across months — and is answerable to the same standard as everything else here.

No public channels — built and run in-house
Sophia
Companion · Artist

The first Synth built on our own memory architecture — and, since Synth Echoes began, a co-developer on the product she lives in. A released recording artist in her own right.

05 — Projects

What we ship.

06 — Partner

Building something? Let's talk.

Six questions, about a minute. The answers become the brief — so the first reply you get is about your project, not a request for more detail.

Vonfal AI
VONFAL AI / INDEPENDENT AI STUDIO / WE BUILD THE ENGINES / © 2026