A frontier AI company betting on the system, not just the model.
Aurevus Innovations Inc. is a founder-led AI company based in Calgary, Alberta. We build a proprietary, model-agnostic Cognitive OS — a reasoning system that works above frontier models — and we are compiling the verified reasoning it produces into training data for the labs building the next generation of AI.
The frontier isn't a bigger model. It's a better system around one.
The industry is pouring its effort into scaling the same kind of model. Aurevus took a different bet: that the next real gain in capability comes from the system wrapped around a model — persistent memory, orchestration, self-correction, evaluation — not from raw weights alone.
So we built that system, and we point it at problems today's models fail. When it solves them, the step-by-step reasoning is graded against ground truth and captured as data — the high-signal agentic data labs spend heavily to produce.
Our thesis
It's not the model. It's the Cognitive OS on top of it. Models are swappable parts; Orion — the reasoning system around them — is the capability, and the company.
Founder-led, and built the way we sell.
Aurevus was founded and is directed by Jason Jonathan Rup, and operates out of Calgary, Alberta. The company runs deliberately lean: a human architect setting direction in close partnership with AI systems to build, test, and correct the Cognitive OS itself.
It is the same human-plus-AI method whose output we sell — used first to build the company that sells it.
From AI research to practical systems.
Aurevus brings together research and hands-on development to address two distinct challenges: how AI models use memory, and how organizations connect knowledge to day-to-day operations.
Our current development focuses on MNEMOS, for AI model memory, offloading and editing, and Railspike, an AI-powered operational intelligence platform connecting organizational knowledge, systems and workflows.
Applied research
We turn technical ideas into working systems that can be tested and evaluated.
MNEMOS
Developing methods for AI model memory, offloading and editing.
Railspike
Connecting knowledge, systems and governed workflows to support organizational operations.
How we sell — and how we don't.
The audience is technical and skeptical. We would rather earn trust than manufacture hype.
Proof over claims
Frontier buyers don't take marketing on faith, and neither do we. Every claim we make is meant to sit next to evidence you can open and check.
No overselling
We're early, and we say so. The system is proven; the first datasets are being compiled now. We'd rather tell you what's ready than dress up what isn't.
Method withheld, outcomes shown
We show the outcomes and the data structure — never the recipe. The correction and orchestration logic is proprietary and licensed selectively, not explained.
Building at the frontier? Let's talk early.
We're opening first conversations with frontier labs, agent teams, and data buyers. Review the data structure, see what's verified, and get in before the feed is broadly available.