Gothic Grandma. Laboratories

Research platforms, simulation engine, and scientific tools

Most AI is impressive until you ask how it works.

Black-box models. Training data no one can inspect. When you ask "why did the character do that?" the answer is "because the probability distribution said so." Which is fine for autocomplete. Less fine when you're simulating what it's like to be human.

Gothic Grandma. Laboratories builds the opposite: deterministic simulation where every computation traces back to a biological or psychological model. When something needs adjusting, we can find exactly where and fix it surgically—not retrain everything and hope.

Why Deterministic Simulation, Not Generative AI

Most AI characters use large language models—statistical pattern matching that predicts what sounds right. MUSE takes a fundamentally different approach.

Generative AI Approach

  • Pattern matching
    Predicts what should come next based on training data
  • Black box
    Cannot explain why a character acted a certain way
  • Stochastic
    Same input produces different outputs—no reproducibility
  • Training data biases
    Reproduces patterns from internet text, not authentic biology
  • Hallucinations & forgetting
    Makes up facts, loses context, contradicts itself

MUSE Simulation Approach

  • Biological systems
    Hunger, fatigue, stress—real physiological constraints
  • Transparent causality
    Every action traces back to specific biological/psychological models
  • Deterministic
    Same conditions = same results—enables true scientific reproducibility
  • Science-grounded
    Based on peer-reviewed research, not internet scraping
  • Causal memory
    Persistent, interconnected memories—no forgetting or contradiction

Where We Use AI: Interface Translation Only

Large language models handle one job: translating simulation state into natural language prose. Think of it like a thermometer—the mercury (simulation) determines the temperature, the numbers (LLM) just display it in readable form. The simulation is ground truth. The AI just reports it.

Deterministic simulation means every experiment is reproducible. Given identical initial conditions and inputs, MUSE produces identical results—something stochastic language models fundamentally cannot guarantee. This makes MUSE viable as a research instrument, not just an entertainment product.

Because every behavior traces through inspectable biological and psychological models, researchers can isolate variables, test interventions, and validate outcomes against peer-reviewed literature—the same rigor expected of any scientific tool.

The Research Platform

FONT

The simulation engine where characters have needs, memories, and constraints—not scripts. Behavior emerges from interacting biological, psychological, and environmental systems. When a character snaps at you, it's because they're exhausted and hungry and you reminded them of something painful. Not because a language model calculated that "snapping" was statistically likely.

CYPHER

The research workbench where you design, test, and perfect the models that drive FONT. Visual system designer, real-time analytics, hotfix manager—think of it as a laboratory bench for simulation science. When something isn't behaving right, CYPHER lets you find exactly where and fix it surgically.

Hover or tap to play

CLIO

Development management and knowledge tracking. Every decision, every insight, every late-night fix—preserved and searchable. CLIO is how a small team builds something this ambitious without losing its mind.

CLIO development management workbench

PYTHIA

Population-level research for scientists, policy makers, and anyone who needs to understand what happens when you change the conditions. Twin studies, epidemiology, social dynamics—controlled experiments at the scale of entire communities, impossible in the real world but rigorous in simulation.

Visual preview coming soon

BABEL

The language layer. Like subtitles for a foreign film, except the film is a living world. BABEL translates raw simulation state—hunger levels, emotional valence, memory associations—into natural language prose that reads like literature. The simulation is ground truth. BABEL just reports it.

Visual preview coming soon

Looking for GRIM, our content authoring system? That lives on the Studios side.

GG.Flow

Every lab has that folder. The one full of scripts no one documented, results no one can reproduce, and a pipeline that only works on one person's machine.

GG.Flow is an open-source scientific pipeline framework we built for our own research on emergent simulation. We opened it because the same foundation applies to nearly any batch-processed scientific pipeline—and we think it could do real good.

Learn more about GG.Flow →

Labs and Studios, Feeding Each Other

Research becomes products. Products generate data. Data improves research. Laboratories builds the science that powers Studios—and every experience Studios ships sends back real-world feedback that makes the science sharper. It's not a pipeline. It's a loop.

For Grant Reviewers & Research Partners

Understanding how MUSE differs from conventional AI systems is essential for evaluating our research potential.

Deterministic, Not Stochastic

Every behavior traces through biological models to first principles. Given identical initial conditions and inputs, MUSE produces identical results—enabling true scientific reproducibility.

Simulation, Not Generation

Characters have actual biological systems (metabolism, hormones, cognition). Behavior emerges from authentic processes, not pattern matching in training data.

Interface, Not Creator

AI translates simulation state to prose. It doesn't create content—it reports ground truth. Like a sensor translating voltage to temperature reading.

Auditable, Not Opaque

Every computation is traceable. Community concerns can be investigated at the level of individual heuristics and verified against scientific literature.

Why This Enables Better Research

  • Reproducibility: Same conditions = same results
  • Transparency: Every behavior has clear causality
  • Privacy: Simulation runs locally, no cloud data
  • Validation: Models verified against peer-reviewed research

Interested in research collaboration? We're actively seeking academic and policy research partners. All partnerships must align with our research ethics standards.

The story continues...

Contents

Build With MUSE

Curious about the technology? Interested in integration or collaboration? We'd love to hear from developers, researchers, and builders.

Get in Touch