AEON ● wip

AEON is a little world that runs itself. A deterministic Python simulation ticks forward, fully reproducible — same seed, same universe. On top of it sits a local LLM I call the world-spirit, nudging events, naming things, and giving the world a sense of intention.

The trick is the balance: the model gets to be creative, but the simulation stays deterministic underneath. You can replay any moment exactly. The visual below is a live taste of that world — rendered right in your browser.

The citizens are becoming embodied: they have positions, pathfinding, and movement intents, they perceive danger and opportunity around them, and their behavior is driven by a continuous-time liquid neural network that a larger teacher model trains through a phased curriculum. It is engineering, not a claim — the emergent behavior is fascinating to watch but not yet formally measured, and I try to say so plainly.

AEON is built in the open. The architecture is written up in a technical whitepaper, and the development log tracks how it actually evolves — including the recent work that traced a trainability bug in the liquid brain to a missing normalization and fixed it. It is the world-model corner of a wider research program; the rigorously benchmarked side lives in UFM and FMM, and the whole map is on the research page.