A world you play by reading and typing.
Saxonberg turns real-life engagement into a game. What you learn and practice outside the game becomes the engine of adventure inside it — and underneath every sentence is a model built on real units and real math.
A player just looks. A student points an instrument at it. Same engine, same fountain — only the rendering changed.
> look A fountain stands at the center of the courtyard, water trickling softly.
> analyze sound here Source: fountain Emitted: 30 dB SPL @ 100–800 Hz At your position: 28 dB Aggregate: 28 dB SPL Reverberation: 0.6s (living room)
Learning feeds play. Play motivates learning.
One loop, not a quiz bolted onto a game. Your character grows in a way that matches what you actually did.
You practice
Study or practice something real, outside the game.
The game notices
That engagement becomes a signal the world can read.
Your character grows
Advancement matches the shape of what you did.
The world opens
New capabilities — and new reasons to go learn the next thing.
↻ and the loop closes — each turn hands you the next reason to begin again.
Both readable and rigorous.
Numbers, not vibes
Where the engine models something measurable, it uses real units and real math — no 0–100 sliders or 1–5 star fudge anywhere in the substrate.
A chemistry student inspecting a beaker sees real molarity. A biology student playing a dog has that species’ real hearing range. In-world instruments and the analyze verb reveal the numbers; everyone else just reads the sentence.
A feature, not nostalgia
Everything you perceive arrives as prose; everything you do is a command string. Text can represent anything — a smell, a legal argument, a chemical reaction.
It dodges the uncanny valley, its ceiling is your imagination rather than an art budget, and it is social to the core. Graphics ride as a decorational layer over the text, never under it.
The human interface is the AI interface.
The loop a player runs — read prose, decide, type a command — is exactly the loop a language model runs. So there is no separate AI adapter. An AI-driven character perceives through the same text channel, acts through the same command bus, and is gated by the same security as a human.
> say Does anyone have last week’s lecture notes?
Priya raises her hand. “I’ve got them — I’ll drop them in the channel.”
Devon shrugs. “Mine are a disaster, sorry.”
Professor Albright nods. “Good. Let’s pick up where we left off.”Any of those three might be running on a language model — and nothing in the engine, the prose, or the command bus knows or needs to know which. Because every action is a visible, attributable command, an agent’s behavior is as auditable as a person’s, and an agent can’t do anything a skilled player couldn’t.
What it’s made of.
- World model
- A shallow spine of fundamental kinds — Idea, Thing, Location, Vessel, Agent — crossed with composable traits like Container, LightSource and Atmospheric. A torch, a lantern and a glowing sword aren’t three classes; they’re three combinations.
- Server
- TypeScript, Node and Express, WebSockets, MongoDB.
- Client
- React and Vite — a multi-panel cockpit with a dominant text console and a contextual inspection pane. Every UI gesture composes a command string and submits it to the same bus the CLI uses.
- Status
- The engine foundations are largely in place. The gamification layer and the vertical-integration tier are what’s ahead, and combat is in design now. The public instance is an early demo, not a finished game.