Emotional substrate, licensable

The physics layer for things that feel.

Most AI characters fake emotion with scripted flags or an LLM call every time something happens. Ours runs on real emotional dynamics, oscillators that rise, settle, and drift the way a nervous system actually does. Cheap, consistent, and alive over time.

How it works

Three parts. No prompt engineering required.

You send context in. The engine handles the feeling. You get a state back your character can act on, whether that's a line of dialogue, a stat change, or a facial expression.

01 / input

Something happens

A player betrays an ally, a user shares bad news, a quest goes wrong. The event goes in as plain context, nothing to hand-script.

02 / oscillator core

Emotion does what emotion does

Coupled oscillators carry the feeling, let it pull on related emotions, and let it fade back to baseline on its own clock, not yours.

03 / output

A state you can use

Out comes a clean emotional state your game or chatbot can act on instantly, no extra model call needed to "decide" how it feels.

The problem

Two ways people fake this. Both have a ceiling.

Scripted states

A handful of mood flags swapped by hand. Fast and cheap, but it never feels lived-in. Every player notices the seams eventually.

LLM every tick

Ask a model how the character should feel, every single time. Convincing in short bursts, but slow and the bill adds up fast at any scale.

Interactive Emporium

Runs locally on real emotional physics. Holds a grudge, calms down, gets pulled between feelings, all without calling out to a model or writing it by hand.

Where it fits

Built for two kinds of characters.

Game NPCs

Characters that remember how they feel about you

An NPC that stays cold after a betrayal, warms back up over real time, and reacts differently depending on what it's already feeling. No dialogue tree can fake that consistency over a 40-hour playthrough.

AI companions & chatbots

A mood that carries across the conversation

Instead of resetting to neutral every reply, the companion's emotional state persists and shifts the way a real one would, without burning extra tokens to manage it.

Proof, not promises

Built and stress-tested in a real project before it ever became a product.

88–97%
less compute than calling a model for every emotional beat, verified in sandbox testing.
98/100
engine health score after a full multi-pass hardening sweep across every layer.
34–52
ticks for an emotion to settle back to baseline on its own, tuned per emotion, not a flat timer.

Want to see it react in real time?

Send your email and I'll get a short demo to you, a character actually holding and shifting emotion on screen, no slides needed.