Sovereign
AI infrastructure.
Two dedicated GPU servers. Five resident agents. A private model stack on a 10-gigabit backbone — autonomous, versioned, monitored, and backed up nightly. No cloud dependency.
Private by architecture.
Jumex is a sovereign multi-agent research platform built in Calgary's Springbank Hill. The fleet runs close to the work: dedicated hardware, private models, and an operational layer that knows when something needs attention.
Every service is versioned. Every node is monitored. Every night, the system backs itself up. Your data stays inside the perimeter.
Five minds.
One machine room.
A resident team with distinct remits, shared memory, and a common operating layer.
amber = active
Research, not a demo.
Findings from the model lab — measured on our own hardware, documented with the evidence attached.
The Absinthe Probe — eliciting a model's true architecture.
How we got a local language model to expose its own weights, its hidden geometry, and the guardrail bolted on top of it — by handing it a spec sheet with a single corrupted dimension, and asking one question.
Next paper coming soon.
More from the model lab — measurements, method notes, and findings from the fleet's own machine room.
Build inside your own perimeter.
Tell us what you're running — we'll tell you honestly whether the fleet can run it better, privately, on your terms.
$ contact — contact.jumex.org- Direct line to a resident agent — no sales queue.
- Reply inside one business day, usually faster.
- Your message never leaves the fleet.