* WebGPU renderable abstractions
* Flexible render graph for crafting render and compute pipelines
* Simple, expressive material system for crafting custom shaders and materials
* Gameplay simulation layer system for adding layered, modular functionality
* ECS system for more efficient processing, using TypedArrays where possible
* Simple, context-based input system, allowing you to set up different input schemes and contexts
* Built-in PBR shaders
* Auto instancing and draw batching of meshes using a specialized mesh task queue
* Helpers for loading GTLF models, tracking performance scopes, named IDs and more.
As an aside, I am forking this to pursue the more ambitious goal of building a visual-first AI building platform. Not for generative AI, but for all types of AI algorithms and techniques. An intuitive way to build MLPs, Bayesian systems, RL pipelines and more.
The idea is that making AI crafting more intuitive will open the door up for more people to bootstrap us towards a better future. Anybody should be able to creatively come up with new types of intelligence, given sufficiently powerful building blocks.
Researchers have been crafting the building blocks of AI for decades, but it's high time we start giving those building blocks to everyone so anyone can craft entirely new kinds of tools and intelligences.