About — ORIAS

We are interested in systems where useful intelligence accumulates through interaction with complex environments — where understanding is not only generated, but retained, tested, corrected, and made to compound.

Most of today’s AI intelligence exists inside transient models. A model is trained, it answers, a session ends, and whatever it understood in that moment evaporates. The intelligence is real. It simply has no place to settle. It does not accumulate.

ORIAS is interested in persistent intelligence around worlds. Not intelligence floating inside a model, but intelligence anchored to a place — a world it can observe, act on, and learn from over time.

Chimera is the first major programme, and the first serious test of this thesis. It is persistent engineering intelligence for software.

Why software first

Software is the first world because it offers something almost no complex environment can: unusually precise observation, execution, branching, replay, evidence, and reversibility.

It can be inspected. Run. Copied, changed, branched, replayed, tested, compared, rolled back. Few environments allow an idea to be tested this cleanly, or a mistake to be undone this reliably.

We start here not because software is the most important world, but because it is the one where these ideas can be tested most honestly. A thesis this large should be proven in the smallest place where the evidence is unambiguous before anyone asks it to travel further.

How we work

We keep what exists, what is being built, and what is still research strictly separate. We say “does” only about what can be demonstrated today. We prefer a falsifiable question to a confident claim, and where the evidence is not yet in, we say so.

This is not modesty for its own sake. A system built on the difference between knowledge and belief has to practice that difference itself.

Chimera is where the thesis is being tested, one repository at a time. It is a product and an experiment at once: a working system for teams, and a controlled setting in which we can measure whether persistent intelligence actually helps. The goal is not to win an argument about AI. It is to find out, under repeatable conditions, whether the idea works.

The larger thesis

If a world can be represented,
if interventions can be made or simulated,
if consequences can be observed,
if evidence can be settled,
and if useful experience can persist,

then perhaps some of the underlying architecture can generalize beyond software.

Further out, ORIAS is interested in systems that can maintain evolving world understanding, learn from interaction, explore possible futures, compare interventions, reason about consequences, retain grounded knowledge, and become progressively better at acting in a world.

We are not building a universal simulator of reality. We are not claiming any of this has been achieved. These are directions, and most of them remain unproven.

If you are interested in the questions more than the claims, the Research page is the honest map of what remains open.

Software is the first world.

We expect some of the underlying ideas to travel further. We are not claiming that yet.