The Platform

MUSE is the first platform for applied emergent simulation at massive scale. Entities with real internal systems interact in persistent environments and produce outcomes no one scripted.

What is MUSE?

Most simulations are scripted. Set up a scenario, define the rules, run the clock. MUSE is different. Every entity in a MUSE simulation has its own internal model — biological systems, psychological states, social relationships — that drive its behavior from the inside out. Outcomes emerge from those interactions, not from a script.

This distinction matters more than it might seem. Every other simulation tool is a question-answering model — you define the question first, then build the model to answer it. MUSE is a world model — you build the world, run it, and watch what happens. The answers come out the other side, including answers to questions you never thought to ask.

A city that commissions a MUSE simulation to study a new transit line may discover that the same run reveals unexpected shifts in business district activity, public safety patterns, and neighborhood density — none of which were in the original scope. MUSE runs the world. You observe.

The MUSE Ecosystem

Gothic Grandma builds and maintains every layer of the stack.

FONT

The emergent simulation engine and kernel orchestrator. Manages entity systems, heuristics, and state at runtime.

TESSERA

Visual node graph Constructor for defining Universal Entity schemas — the blueprint for every entity in a simulation.

GESTALT

Code generator that turns TESSERA schemas into FONT kernel C++ source. No hand-writing engine code.

GLYPH

The unified internal workbench for kernel engineering, R&D, content authoring, and research.

BABEL

The LLM and natural language interface used across MUSE tools — bridging simulation state and human language.

GRIM

The consumer-facing e-reader and marketplace. How the public experiences MUSE Living Worlds.

GG.Flow

Open-source scientific pipeline framework for reproducible data workflows. Built for science, not just neuroscience.

MOIRA

Financial modeling and scenario analysis tooling built on MUSE entity simulation.

How We Build It

MUSE is GPU-native from the ground up. Entities are stored in compact Structure-of-Arrays memory layouts designed for throughput, not convenience. Behavioral heuristics use bitmask decomposition and are dispatched via FontDispatch — a custom kernel scheduling system that routes work across CPU and GPU based on biological processing constraints.

The team behind MUSE comes from neuroscience, data science, and cognitive science. The systems aren't metaphors — they're grounded in how brains, bodies, and societies actually work, then optimized for real-time performance.

Hover to preview GLYPH
Hover to preview CYPHER

Why Not LLMs?

Large language models are good at generating plausible text. They are not good at modeling the causal dynamics of complex systems. An LLM cannot tell you where the outbreak goes because it has no internal model of a pathogen, a social network, or a healthcare system. It can describe what those things sound like — it cannot simulate what they do.

LLMs are good at

  • Generating plausible text
  • Summarizing and translating
  • Answering questions from training data
  • One-shot generation tasks

MUSE is good at

  • Causal, mechanistic simulation
  • Emergent outcomes from interacting agents
  • Reproducible, auditable runs
  • Answering "what happens if" questions at scale

Why Not Existing ABM Tools?

Agent-based modeling (ABM) tools have existed for decades. MUSE is not one of them — or rather, it is something they were trying to become. The differences are architectural, not cosmetic.

Epistemix / FRED

  • CPU-bound; one-day minimum timestep
  • Agents are census demographic records — no internal physiology
  • Behavior is state machine transitions on co-location schedules
  • Built for epidemiology; limited outside that domain

AnyLogic

  • Java-based, CPU-only
  • General-purpose but DIY — you build the model from scratch
  • License model: buy the software, then build everything yourself
  • No shared scientific model library

FLAME GPU

  • GPU-scale ABM — but an academic research library
  • Requires CUDA C++ expertise to use
  • Not a commercial product; no toolchain, no support
  • Proof of concept, not a platform

MUSE

  • GPU-native (CUDA + Metal), biologicaltiming hierarchy
  • Agents have real internal drives — physiology, psychology, social state
  • Structure-of-Arrays entity storage; FontDispatch kernel scheduling
  • 50× faster than real time with 100K+ entities

GG.Flow

GG.Flow is Gothic Grandma's open-source node-based scientific pipeline framework — and the connective tissue of the entire company. It provides reproducible, composable data workflows for scientific computing, built initially for neuroscience but designed for any domain where you need to move data through processing steps reliably and repeatably.

Inside Gothic Grandma, GG.Flow does the work that holds everything else together:

The open-source core is available for any research lab to use independently. The MUSE-specific node library — the GG.Flow nodes that interface directly with FONT shared memory, run simulation ensembles, and produce MUSE diagnostics — is proprietary.

View GG.Flow on GitHub →

Talk to us about your simulation needs →