What we were handed.
“We use AI” is easy to say and hard to show. Screenshots of chatbots all look the same, and a demo that resets every session proves nothing about running agents for real.
We wanted a piece of proof you can walk around in — AI characters with memories, relationships, and plans, running continuously on our own hardware rather than rented API calls.
The system.
Voxelburg: a persistent virtual town where AI residents live out their days — chatting with each other, forming opinions, and remembering visitors between sessions.
The agent simulation, the language models, and the world state all run on VoxelBox's self-hosted AI stack. It isn't a scripted demo; it's the same agentic machinery we use for business automation, wearing a friendlier face.
Agents with memories, relationships, and daily plans of their own.
A persistent simulation that keeps running whether or not anyone's watching.
Self-hosted LLMs doing the thinking — no per-conversation API bill.
Walkable in the browser, by anyone, any time.
What changed.
A public, always-on demonstration of agentic AI infrastructure — the same building blocks we use for client work, running where anyone can poke at them.
The next study could be yours.
Bring us the problem — the messy, manual, “there has to be a better way” kind. We'll design the system, build it, and run it.