I build a simulation engine, based on a high performance distributed actor system library I wrote for just this purpose. It is generic, but a lot of the performance trade-offs are with Citybound's needs as the main driver.

I see you didn't use Actix or even Tokio, could you please explain why?

Their actor implementation / event loop works on a higher level than mine. Mine is super simplistic intentionally to offer high performance. Being optimised for cache locality, processing similar actors in batches, using custom allocators and having encoding-free communication, it has many parallels to data-driven game engine design or even entity component systems.

cargo +nightly run of the examples in your Kay project appears to be broken, I'd love to explore using it in a future project.