Orebex logo

Orebex

The concentrator runs itself. You decide how far.

Seed · 2026 Mineral processing autonomy Active

The autonomous operations layer for mineral processing and ore concentration — autonomous crushing and grinding to liberation size at minimum energy, flotation reagent, air and froth control, classification and dewatering, froth-vision grade and recovery sensing, whole-plant recovery, throughput, energy and water optimisation, and an as-run concentrator digital twin that hits grade, recovery and throughput before the shift.

Company profile. Orebex details — founders, products, and open roles — are taken from the company's own materials at orebex.com. AWSM VC backed Orebex's $1M seed round.

The stack

Frothix

The perception layer for the concentrator: grade, recovery and froth seen in real time.

Grindex

The autonomous crush-and-grind agent that grinds to liberation size at minimum energy.

Floatex

The autonomous float-and-reagent agent for flotation, reagent, air and froth control.

Recovex

The plant-wide optimisation brain for recovery, throughput, energy and water together.

Oretwin

The as-run concentrator digital twin that hits grade, recovery and throughput before the shift.

The team

Robert Ellison

Founder & CEO

Owns Orebex's design-partner relationships, pilot structure and commercial model, deciding which concentrator and which circuit go first and what recovery, grade, energy or water number a pilot is judged on.

LinkedIn →

Justin Blake

CTO

Owns the plant-edge runtime, the grinding, flotation and froth-vision models, the as-run concentrator twin and the audit architecture behind every Orebex agent.

LinkedIn →

Jobs at Orebex

View all jobs →
Founding Edge Perception Engineer

London / Remote-friendly · Engineering

Apply →
Founding Controls and OT Integration Engineer

London / Remote-friendly · Engineering

Apply →
Founding Simulation Engineer

London / Remote-friendly · Engineering

Apply →
Founding Metallurgist

London / Remote-friendly · Process

Apply →

Own a constraint like this?

If you're building the layer Physical AI can't scale without, we want to hear about it.

Tell us your constraint →