Training · live
A 1.19-billion-parameter foundation model, trained from scratch on Canadian iron at AXE Studio Toronto. Numbers below refresh from JL1 every minute. No mocks, no marketing demos — actual loss values from the live run.
/ 01 — What it is
CASTLE binds inference, memory, data, identity, policy, and the controlled web into a single operating fabric. The model talks to the ontology. The ontology talks to the data. Every request crosses the same audit boundary in both directions.
Customers stop integrating six vendors. They start operating one system.
/ 02 — The six
/ 03 — Provenance
Every hop, every header, every hash is signed and stored append-only. The audit chain reconstructs any decision, in any order, by any reviewer with the right key.
Frontier labs don't track inference-level provenance — consumer users never demanded it. Every regulator AXE will meet does.
/ 04 — Architecture
Five layers, six systems, one audit chain. Every request enters through the gateway, traverses cognition and knowledge, and leaves a receipt with Knox on the way out.
This is the actual deployment shape — not a marketing abstraction.
/ 05 — Edge
Every weight learned from scratch on hardware in Canada. One model family, four sizes, one training pipeline. No upstream dependency.
Frontier labs sell tokens. We sell the room the model lives in.
/ 06 — Deploy
/ 07 — The live surface
MODULE 01
Real-time visualization of a sovereign AI platform operating within Canadian jurisdiction. Every request. Every model. Every decision. Under your control.
MODULE 02
Every action leaves a verifiable, cryptographically signed trail. Complete transparency for regulators, partners, and your future self.
MODULE 03
Built for operators, SREs, and security teams. Observe, decide, and act with speed and clarity.
/ Contact · we read every inquiry
.Demos, partnerships, government RFPs, technical questions. A person reads every form. You hear from someone — not a queue.
AXE Concierge · live
Answered by a local model · Canadian iron · zero external API