The Framework
A computational model of identity. Where behaviour is derived from psychology, not guessed.
Part 1
What you see in the library are not random characters with trait lists. Each character has an internal structure: a psychology that determines how they react, what they protect, and where they break. You can read them once and understand them. Push them in any direction and predict how they will respond. Not because you memorised their backstory, but because you understand how they think.
That is what makes them worth remembering.
Part 2
Every character is built through the same pipeline:
What shaped them. Their nature (temperament they were born with), their culture (what their world rewarded and punished), and their key experiences (the moments that changed them).
Derived from formation. We use Schwartz's 10 universal human values - the same framework used in cross-cultural psychology research. Each character has a weighted value profile. Not invented. Calculated from who they are.
Emerge from values. Their anchor (the value they protect above all else), their limits (lines they will not cross), their variance (how predictable they are), and their defences (how they cope when threatened). Based on Vaillant's hierarchy of defence mechanisms.
Follows from all of the above. When you know what shaped someone, what they value, and how they protect themselves - their reactions are not surprising. They are inevitable.
The Science Underneath
Part 3
This is where we are going.
Today, the framework guides AI generation. We give the model a character's formation and ask it to derive values, properties, and behaviour. It follows the rules. Usually well. But it is still generation - probabilistic, variable. We are building something different: a computational model where values are calculated from formation, where properties are derived through formulas, where decisions are computed before any language model speaks.
Same identity + same situation = same response type. Every time. Not because we hope the AI follows instructions. Because the math guarantees it.
We are continuously refining the model - deepening the psychology, expanding the dimensions, improving the precision. When complete, the AI does not decide what the character does. The computation decides. The AI only finds the words.
Part 4
This is not about NPCs. Characters are the proof of concept. The first application. The way we demonstrate that the architecture works. But the pattern is universal:
Entity + Values + Situation → Computed Decision → Generated Expression
This applies to AI companions that stay in character across thousands of interactions, game NPCs that respond consistently without scripting every possibility, autonomous agents whose decisions you can trace and trust, and any AI system where you need to know WHY it did what it did.
The problem with current AI is opacity. Values are trained into a black box. You hope it learned correctly. When it misbehaves, you cannot explain why. We are building toward AI where values are explicit, decisions are computed, and every output traces back to the structure that produced it. Transparent. Auditable. Trustworthy.
Not because we told it to behave. Because the architecture guarantees it.
Status
Live Now
In Development
Looking For
Identity Engine. Characters worth remembering.