Tools

The engines behind our work

We build our own tools before we build content: procedural asset pipelines, headless test harnesses and numerical code. We come from math, physics and scientific computing, and that shows in everything we make.

A Kiln recipe: a few lines of JSON on the left, the chair Kiln builds from it on the right

3D asset engine Kiln A procedural 3D asset pipeline. Declarative JSON recipes go in; deterministic, game-ready glTF 2.0 comes out: props, furnished buildings, whole towns, rigged and animated characters, and music. 49 generators and counting. Explore Kiln →

The TiGen application: terrain generator with live preview, tile output and sprite viewer

2D asset engine TiGen A procedural 2D tile and sprite generator. Simplex, fBm and Worley noise become autotile sheets, environment sprites and full level layouts: 42 generators, 776 assets, the art behind Defense of Kyrath’s biomes. Explore TiGen →

Terminal output of the headless test harness: an auto-placer choosing kill zones and the result of each simulated battle

Automated testing Headless balance testing A headless Godot harness that plays Defense of Kyrath at 200× speed. An auto-placer scores path geometry to find kill zones, and batch runs across 50+ levels and a 15-tier loadout matrix drive the balance tuning. Read how it works →

Two surface plots: the true function cos(r) and the neural network's reconstruction of it from data

Machine learning Neural-net function solver A TensorFlow network that solves an inverse problem: given samples of z = f(x, y), recover the unknown function. It trains on experimental data, or on a known function for validation. Shown: the true surface and the network’s reconstruction. View on GitHub →