Moxilang Logo
The Moxi Ecosystem

Language-Based 3D
for AI & Humans

A semantic layer where LLMs and humans describe spatial intent in the same language.

Transform natural language into deterministic, AI-friendly 3D worlds. Where intent compiles into geometry.

Moxi lets you describe worlds in language. Voxelmap is the lower-level voxel schema where that idea began.

Moxi in Action

Semantic markdown that compiles into complete 3D worlds. Read the code, understand the intent, see the result.

Tropical Island World

A complete island ecosystem with materials, atoms, and entities.

# Tropical Island World - Moxi Code

# Atoms
> Atoms are the atomic unit. Every material maps to exactly one atom.
> Two atoms can share a color (TRUNK and SOIL both brown) but remain
> semantically distinct — useful for future material logic and gameplay.

atom SAND   { color = yellow }
atom SOIL   { color = brown }
atom ROCK   { color = gray }
...

# Materials

material Sand   { color = yellow, voxel_atom = SAND }
material Soil   { color = brown,  voxel_atom = SOIL }
...


# Palm Tree
> Crown above Trunk places the blob canopy on top of the cylinder trunk.
> The relation resolver computes the exact y offset from the trunk height.

entity PalmTree {
    part Trunk { shape = cylinder(height=6, radius=0.6), material = Bark }
    part Crown { shape = blob(radius=3, roughness=0.35),  material = Leaves }
    relation {
        Crown above Trunk
    }
    resolve voxel_size = 1.0
}

...

# Output
> Bottom to top render order — ocean first, rocks last, trees over everything.

print Ocean       detail=low
print SandBase    detail=low
print SoilTerrain detail=low
print RockyPeaks  detail=low
print PalmTree    detail=low

Moxi code is plain markdown. Semantic, readable, and fully deterministic.

Compiled Result

3D world rendered from code

Tropical Island rendered

Deterministic compilation: same code = same world

View on GitHub

Semantic & Clear

Read the code like documentation. Every definition is intention made explicit.

AI-Generatable

LLMs can reason about and generate Moxi code with semantic precision.

Deterministic

Same code always produces the same 3D world. No black-box magic.

The Moxi Ecosystem

A layered architecture where language, compilation, and AI align to create 3D worlds with semantic precision.

Moxi mascot

Moxi

Semantic 3D Language

The high-level declarative language for describing 3D spatial intent. Designed to be parseable and generatable by both humans and AI systems.

RustMarkdown
GitHub
Voxelmap cubes

Voxelmap

Lower-Level Voxel Schema

Voxelmap is not the compiler for Moxi. It is an older, orthogonal Python system for working directly with voxel/tensor-based 3D structures, where elements can carry embeddings and semantic meaning. It is closer to a low-level instruction layer, and includes GPT-assisted workflows for generating Voxelmap code.

Python
DeckGitHubVoxelmap GPT

Built for AI & LLM Generation

Semantic Alignment

Moxi uses a language designed for LLMs to reason about and generate 3D structures. No more black-box outputs—every dimension, color, and relationship is explicit.

Deterministic Output

AI-generated worlds are reproducible and inspectable. Debug and refine spatial reasoning at the semantic level, not the geometry level.

Scale Generation

LLMs can generate Moxi descriptions in bulk to create entire procedural worlds. Language-based generation scales where manual modeling doesn't.

Built for AI Generation

This is a paradigm shift. Language models can now reason about and generate 3D worlds with semantic precision—not through opaque neural mechanisms, but through explicit, inspectable intent.

Semantic Generation

LLMs understand spatial intent semantically. No more guessing what hidden latents mean—Moxi makes it explicit.

Structured Output

Language models output deterministic, structured descriptions that compile directly into 3D worlds.

Reproducible Worlds

Same description generates the same world every time. Debug and iterate on semantic meaning, not geometry.

Procedural Scale

Generate entire worlds procedurally. LLMs can create hundreds of unique spatial descriptions in seconds.

Real-time Feedback

Inspect wireframe representations instantly. Refine AI outputs at the semantic level before resolving to 3D.

Fine-grained Control

Specify exact dimensions, relationships, and constraints. AI respects spatial semantics, not approximate aesthetics.

For Game & VR Developers

  • →Generate procedural worlds with LLMs while maintaining spatial constraints
  • →Create dynamic level generation systems that understand geometry semantically
  • →Inspect and debug AI-generated structures before rendering

For AI Research & ML

  • →Study how language models reason about 3D space
  • →Train models on semantic 3D reasoning with interpretable outputs
  • →Build agents that can understand and manipulate spatial worlds

The Moxilang Philosophy

3D creation moves closer to language by putting meaning first and geometry second.

1

Language Over Geometry

3D creation should start with what you want to express, not with primitive shapes. Natural language becomes the interface between intent and form.

2

Deterministic Systems

Move beyond opaque generative outputs. Moxilang compiles semantic descriptions into predictable, controllable spatial systems that you can inspect and modify.

3

Intent → Code → Preview → Form

A transparent pipeline where meaning is extracted, structured, previewed as wireframe, and finally resolved into solid geometry. Every step is visible.

4

Human & AI Aligned

Whether you write descriptions or AI generates them, Moxilang creates a common semantic layer. Both humans and intelligent systems reason in the language of intent.

The Transformation Pipeline

01Natural Language: Describe what you want to create

02Semantic Parsing: Extract meaning and structure

03Wireframe Preview: Visualize the computed geometry as green edges

04Solid Resolution: Render final 3D form with materials and lighting