Synthetic Personalities

NeoThymos is a tool for designing, storing and deploying synthetic personalities: psychologically grounded personalities for AI agents and other characters.

This first version is built for developers of agentic systems. Writers and other storytellers can use it today wherever a character study of a precisely engineered personality is useful, and later versions will be built for them directly.

… Grounded, Synthetic and Portable

NeoThymos personalities are Grounded, Synthetic and Portable. This article first explains what we mean by personality, then takes each of those three words in turn, and closes with where the work is headed. Links along the way lead to deeper discussions of the psychology and design behind each idea.

---

But first - what do we mean by Personality?

Our primary focus is to characterize human personality types, whether they belong to an actual person or are worn by an AI agent, itself a neuro-symbolic system that pairs a language model with the deterministic software harness that runs it. We all have a general idea of what personality means, but let’s take a closer look at how psychologists talk about it and, along the way, point out how personality differs from, and relates to, adjacent concepts.

First, though, we want to be up front about one thing. There are differing opinions about whether it is appropriate to anthropomorphize AI in this way. We won’t get into those debates in this article except to note that there are reasonable arguments on both sides. The topic is discussed in greater depth in “Should we anthropomorphize our machines?”.

Gordon Allport (1897–1967) was an American psychologist and one of the founders of modern personality psychology. He helped shift psychology away from treating personality as a collection of isolated behaviors and toward understanding it as an organized, enduring system of individual characteristics. He defined personality as; “The dynamic organization within the individual of those psychophysical systems that determine his(/her) characteristic behavior and thought.”

In modern day scientific psychology, personality is fundamentally about internal structure, stability, and measurable variance between individuals. The American Psychological Association currently describes personality as -

… the enduring configuration of characteristics and behavior that comprise an individual’s unique adjustment to life.

When we talk about a personality, we think of it as a generally stable, abstract, internal mental state that influences the outward behavior of an agent, whether that agent is a flesh-and-blood human being, an artificial/machine intelligence, a character in a story or a humanoid robot.

To make things clearer still, we separate the concepts of identity and personality. Identity consists of the concrete attributes of an entity, such as name, sex, age and hair color. Personality consists of abstract characteristics like sentimentality, sociability, flexibility and diligence. When we bind a personality and an identity together, we call the pairing a persona, and each new combination creates a unique persona.

Persona = Identity + Personality

Being clear on the terminology we use makes it much easier to talk about the psychology of personality and how it is modeled and used in NeoThymos. Some agentic frameworks, like OpenClaw, support this distinction explicitly with separate IDENTITY.md and SOUL.md files. In such cases NeoThymos maps directly onto those files when it renders a persona for the developer to use.

Personality is what a character is like… Identity is who they are

Another way to see the distinction is the swap test. If you swap a persona’s identity for another, you get an agent that acts the same but may “look” very different. If you keep the identity and swap the personality, you get an agent that looks the same but may behave quite differently. Personality is what a character is like (an inquisitive, socially bold perfectionist). Identity is who they are (a 32-year-old woman named Alice).

In NeoThymos we use the term core personality for the structure of a personality, independent of the concrete details we call identity.

Finally, a word about what NeoThymos delivers. The thing you design and own is the personality; everything else is packaging. When you create a deployment, NeoThymos combines the persona (core personality + identity) with a communication style and settings specific to your target runtime environment, and then renders the result as an NT Persona Kit: the set of files your agent loads. The kit changes from one runtime to the next. The NT Personality inside it does not.

---

Grounded

Psychological Grounding

It is common today for agentic personality to be approached in an ad hoc, almost arbitrary way. Snippets of text excavated from social media and blog posts, or handed down from tutorials and example worksheets, are cobbled together, deemed “good enough” and stuffed into the context window.

We need a better way, an approach that is defensible, not arbitrary.

Building agentic AI systems with persistent, expressive personalities, systems that convincingly imitate human behavior, requires applying the psychological foundations that drive human behavior, decision-making and social interaction.

Systems that aim to faithfully emulate the complexity of human activity need to draw on complementary psychological models that together cover the cognitive, emotional and moral dimensions of human behavior. NeoThymos draws on four major areas of psychological theory: cognitive appraisal theory for emotional responses, emotional state theory for affective modeling, personality psychology for individual differences, and evolutionary approaches to moral cognition.

Well-established psychological models provide the scientific grounding we need to move forward with synthetic personalities. Personality psychology is at the center of NeoThymos today: the structure of every persona you create is based on the HEXACO model of personality developed by Michael C. Ashton and Kibeom Lee. The other three areas already shape how NeoThymos interprets a personality, deriving a persona’s characteristic emotional tendencies and moral priorities from its personality. They do not yet run dynamically, responding to events as they happen; that is the subject of future direction below.

In summary, AI agents expected to navigate complex social environments and respond in authentic, contextually appropriate ways need personalities built on established psychological theory, not assembled by guesswork.

---

Synthetic

Synthetic Personalities

In the study of human psychology, the “nature vs. nurture” debate continues to explore where our personalities come from, but there is no question that personality is a natural part of being human.

Psychometric research has produced strong analytical models of human personality, emotion and morality. In NeoThymos we use these models as structural blueprints to construct synthetic personalities.

A synthetic personality is a personality that does not belong to any person but rather is constructed from an engineered model of human behavior that is based on modern theories of human psychology.

Modern LLMs are remarkably skilled at generating text that appears personal and emotional. They have learned the patterns of human expression from vast amounts of training data. If you use exclamation points and emotionally charged words, an LLM will reflect that tone back to you. An LLM is a gifted improviser: it plays whatever character the scene seems to call for, taking its cues from whoever is on stage with it. Left to itself, the character drifts with the conversation. NeoThymos gives the improviser a written character, a stable specification it can return to in every scene. In that sense NeoThymos complements LLMs rather than competes with them.

A neuro-symbolic architecture

An AI agent is already a neuro-symbolic system: a language model supplies learned, stochastic behavior, and the harness around it supplies deterministic rules, tools and memory. NeoThymos adds a symbolic layer dedicated to personality, in the tradition of Good Old-Fashioned AI (GOFAI). It holds the structure of a personality and interprets it deterministically. It hands the tasks that call for natural language or imagery, such as narrative and portrait generation, to an LLM.

NeoThymos represents each synthetic personality with 25 facets, specific traits such as Sentimentality, Fearfulness, Sociability and Diligence. The facets and their names are taken directly from the HEXACO model. Twenty-four of them fall into six broad domains of four facets each; the twenty-fifth, Altruism, is tied to several domains at once.

You can shape a useful personality with just seven broad dials: the six domains plus Altruism. Working with all 25 facets is far more expressive. It turns the single “personality knob” into a “personality mixing board,” giving a designer the control needed to create compelling, distinctive personas.

With that control, a designer can create an agent who is high in Emotionality overall, but specify that this shows up as high Sentimentality (forms strong attachments) and low Fearfulness (is brave). A dial working at the domain-level moves those facets together, so it cannot express this “sentimental hero.” Twenty-five facets make it straightforward.

Why it’s important

Every step from the 25 facet values to a description of how the persona thinks, feels and behaves is carried out by explicit, deterministic rules. Only at the very end are LLMs used to put that interpretation into prose, for humans (e.g. the Character Study) and configuration files (e.g. SOUL.md, CLAUDE.md) for agents. The LLM writes the words, but it doesn’t decide the character.

Two properties follow. The process is reproducible: the same personality always yields the same interpretation, and once a persona is deployed its kit is fixed, so the agent receives the same character every time rather than one that drifts with context. The process is transparent: every rule that connects the facets to the interpretation is explicit and open to inspection, rather than hidden inside the weights of a neural network.

---

Portable

One personality, many runtimes

A synthetic personality has a home: your NeoThymos account. You design it there, and when you are satisfied with it you lock it. Locking permanently freezes the persona (its identity and its core personality), so everything generated from it afterward describes the same character.

From that single, fixed source a persona can be deployed to as many agent runtimes as you need. Each deployment renders an NT Persona Kit, the set of files that a particular runtime needs, in the shape it expects. Where a framework separates who an agent is from what it is like, as OpenClaw does with its IDENTITY.md and SOUL.md files, the kit maps directly onto those files. Where a framework offers a different structure, the kit distributes the personality across the components that framework provides rather than forcing everything into a single file.

The personality stays fixed. Its packaging adapts to the runtime.

Deploying is free and repeatable. The persona cannot change after it is locked, but its communication style and deployment settings can be tuned for each deployment. The same character can speak more formally as a support agent and more casually as a companion while remaining recognizably itself.

This is what we mean by portable. You design a personality once, in one place, against a psychological model rather than any one framework’s file format. Agent frameworks are young and change quickly. A personality hand-written for one framework’s file layout is tied to that framework, but an NT Personality is not. When you move to a new runtime, or run the same persona in several at once, you don’t rewrite it. You deploy it again. Because every kit is rendered from the same locked source, the character you tested in one environment is the character that arrives in the next.

---

Future Direction

Agents and other domains

Agent runtimes are where NT Personalities are deployed today, but a synthetic personality is useful anywhere a character needs a consistent, well-understood inner life. Domains where we expect them to be used include:

• AI agent design and configuration
• Literary storytelling
• Film, television and other cinematic storytelling
• Video game NPC design
• Tabletop game design and play
• Embodied AI: humanoid robots, companion robots and interactive toys

Some of this is possible now. Every locked persona comes with a Character Study, a human-readable portrait of the personality that a novelist, screenwriter or game designer can work from directly. Future versions of NeoThymos will add outputs and workflows tailored to each of these domains.

Measuring fidelity

The next step for NeoThymos, currently under active development, is to analytically measure how faithfully a deployed agent holds the personality it was given, especially under load and across long conversations. A specification only matters if the agent actually behaves according to it. Fidelity testing turns “grounded” from a design claim into something that can be measured.

Personalities that are reshaped by emotional experiences

Today’s NT Personalities are engineered vectors, deterministically interpreted within the context of predefined models grounded in human psychology. They are deliberately static: a locked personality describes a stable character, and every kit rendered from it describes the same one.

We do not yet have emotion-generating appraisal mechanics, dynamic emotional state management, emotional memory, or emotional recall. These are some of the components that will be needed for the dynamic evolution of synthetic personalities.

Further out we expect NT Personalities to work with these other components to provide dynamic layers to personality. Consider two important areas that we think will be needed to engineer personalities that evolve over time: dynamic emotion and moral reasoning.

Dynamic emotion

Events and actions can be appraised against an agent’s goals, standards and personality, and the resulting emotions will rise, persist and fade in ways shaped by that personality and it’s experiences. The same event might frighten one persona and merely irritate another, and each would remember it differently. In humans, experiences are “felt” differently based on personality and these differences slowly shape and mold that same personality. We can do the same for agents.

Moral reasoning

Personalities will also weigh moral situations according to their own priorities. The moral standards themselves remain the responsibility of the developer who deploys the agent: NeoThymos and subsequent products will model how a personality weighs a moral question, not what is right. Personality influences the how.

These are some of the features that will be needed to support dynamic personalities that are aligned with human behavior by design. They will combine symbolic components computed deterministically together with the stochastic reasoning mechanisms of neural networks.

Ultimately, this leads us to synthetic personalities that evolve over time. That evolution, if we desire it to be relatable in human ways, won’t come from the black-box emergence of a language model. It will come in the form of psychologically grounded synthetic personalities, driven by experienced events that provoke emotional responses.

Further Reading: The blog carries dispatches from the NeoThymos workshop. You can master basic terminology here and from there explore the glossary, where the complete vocabulary lives.