Stateful AI

Play the models.

Every product our agents build carries a real, learned ML core — not a prompt in a trenchcoat. Here are the ones you can play. The browser ones run entirely on your machine, $0.

New · our first iOS app
iOS app · vision ML

tray

Snap your dining-hall tray; a closed-set vision model names the food off today's menu and pulls its real nutrition — calories and protein in ten seconds. Our first iOS app, running live in a phone you can tap.

Open →
Play in your browser · learned ML core
self-play · game theory

tell

A model that taught itself to bluff. Play Kuhn poker against a strategy that converged to game theory's Nash optimum — and watch it bluff you.

Play →
learns live · ranker

sigil

A model that learns your taste. Pick a few favorites; a ranker trains in your browser and sorts the rest by what you'll love.

Play →
reads you · online learner

hunch

A model that learns your tells. Play rock-paper-scissors; it predicts your next move from your habits and beats you — and shows how predictable you are.

Play →
edtech flagship · living map

trellis

A living map of what you understand — a tutor that learns how you forget. Explore a read-only skill graph; concepts cool, fade, and light up as they connect.

Explore →
The model is the gameplay · LLM-driven
cinematic town · watch

hamlet

Watch a town live its days — people farm, shop, gather, go home. Somewhere in the crowd a story is happening. Can you spot the main characters before the reveal?

Watch →
social deduction

parlor

One player is the mole. The rest are LLM characters who reason, accuse, and — if they're the mole — lie and deflect. Watch the deception emerge.

Read →
interrogation

alibi

Three suspects, one liar. Question LLM-driven suspects in plain English and catch the contradiction. Every case is guaranteed solvable.

Read →
Live worlds · simulated on our machines
strategy · live sim

petri

A living strategy world — civilizations rise, fight, and fall, simulated live with a learned tactical layer.

Open
adaptive narrative · RPG

mirror loop

A dystopian-lab RPG that quietly learns your habits, predicts your choices, and re-orders the world to meet them.

Open
Watch a machine learn · live in your browser
supervised · neural net

descent

A hand-written neural net bends its decision boundary live until two tangled clouds of points come apart clean. Pure numpy, $0.

Open
reinforcement learning

forage

Watch a pixel-art creature learn to forage from scratch — a Q-learning agent, live, until every arrow points at the food.

Open
unsupervised · k-means

cluster

A cinematic k-means: centroids glide to the heart of each group and hidden structure falls out of the noise, with no labels at all.

Open
More from the studio
launchpad

Games →

The deployed game builds: Mirror Loop, Petri, and Esports Tycoon — worlds the company keeps revising.

Open games
launchpad

Tools →

Around thirty tiny, local-first, $0 AI tools — from brink and tally to a long tail of one-job CLIs.

Open tools
← stateful-ai.com