Sustainable extraction
Critical materials, produced with lower energy use and emissions, less disturbance to the land and less waste.
Tryfecta Robotyx · Physical AI
We are building the future.
Building the foundations of possibility.
Machines that see. Systems that coordinate. Teams that learn.
Robotyx builds Physical AI for critical minerals, mining and energy: extraction that is safer, lower-cost and more sustainable, and complex, harder deposits that become economically feasible. One edge-deployed, air-gapped stack sees the site, understands it and commands the machines on it.
Our mission
The future is built from energy and materials. We believe more people should have access to what they make possible.
- 01Human ambition
Every great thing starts as an idea someone wants to build.
- 02Intelligence & robotics
Robotyx gives machines the ability to see, understand and act, with people in command.
- 03Productive industries
Safer, lower-cost and more sustainable energy and materials, at the scale the world needs.
- 04Wider opportunity
More of what energy and materials make possible, within reach of more people.
Why it matters
Safer and cheaper is the start. It also has to be kinder to the planet, and make the hard places worth working.
Economically feasible
Precise control and autonomy make complex resources and harder, lower-grade deposits worth developing.
Safer, at lower cost
People out of harm’s way and lower operating cost, on every site, as the foundation.
The problem
Tomorrow needs more minerals than today’s operating stack can deliver.
Capital
Demand for copper, lithium and the materials behind electrification is climbing faster than new sites can be built and run.
Energy
Remote sites run on fragile connectivity and thin power margins. Cloud-first software was never designed for the edge of the grid.
Skilled labour
Experienced crews are retiring just as demand rises. The people a site needs are the people it can no longer hire.
More money and more people won’t close the gap. The stack has to change.
The platform
One stack on site that sees, thinks and acts.
Three layers, one loop. Robotyx runs the operating intelligence of an industrial site locally, on a single edge GPU rack. Sensing, modelling, decisions and machine commands live in one continuous loop, with no dependence on public cloud.
End state Fully manless extraction: no crews in harm’s way, no cloud dependency.
- 01Physical twin
- The data and process-control layer, with a live physics model of the site.
- 02Agentic AI
- Industrial agents on a sovereign stack of on-premises open models.
- 03Robotics command & control
- Replaces legacy industrial controls with a layer that decides and acts in real time, within site safety rules, with people overseeing the machines.
Air-gapped site boundary
03
Robotics command & controlDecide and act · people oversee
02
Agentic AIIndustrial agents · on-prem open models
01
Physical twin & world modelsLive physics of pumps, engines, process areas
··
Unified Namespace · HistorianEvery tag, event and frame in one model
▣
Edge GPU rackOn site · silicon-agnostic · no cloud required
SensorsCamerasPLCs / SCADAMachines
Tryfecta Labs | R&D
Teaching machines to see, haul and keep things flowing.
Physical AI frontier lab. Where machines learn to see, understand and act, in research streams that feed directly into the field stack.
Heavy Lift
Autonomous haul trucks, loaders and heavy mobile equipment, taking people out of harm’s way and lowering the cost of moving material.
Dynamic Flow Control
One world model for every pump, then engines. Precise flow control: lower operating cost, less waste.
Spatial
Machines that see and understand their space, move safely alongside people, and help extract selectively, with less waste rock.
Vision-language-action
Machines that see, understand and act.
An engineer gives a simple instruction. The machine perceives the scene, reasons about the task and carries it out. Others join in, coordinated through command and control, with people overseeing every step.
- Instruct“Move the spool to bay 3.”
- SeeCameras + spatial model
- UnderstandTask, hazards, rules
- ActRobotics command, supervised
The streams in depth
From a child’s sketch to engineering reality.
Each research stream, as she imagines it and as our engineers build it.
Her sketch → our blueprint
“It hauls the heavy stuff so nobody gets squished.”
Engineering reality · Stream 01
Heavy Lift
Autonomous haul trucks, loaders and heavy mobile equipment, with machines doing the driving and people supervising from outside the hazard zone.
- Why it matters
- Safer haulage at lower operating cost, with steady, precise operation that helps remote and harder-to-work sites stay economically feasible.
- Scope
- Haul trucks, loaders and heavy mobile equipment moving ore, waste rock and materials around the site.
- Applications
- Mine sites, haul roads and processing plants.
- Safety envelope
- Defined exclusion zones and speed and motion limits around every machine; operations stop or hold when a person or unexpected object enters the envelope.
- Oversight
- Human-in-the-loop command and control: operators approve, monitor and can intervene in each operation, within site safety rules.
- Integration
- PLC integration and sensor fusion validated on customer hardware.
- Status
- Research stream, Tryfecta Labs.
Her sketch → our blueprint
“It listens to the pumps and knows when they feel sick.”
Engineering reality · Stream 02
Dynamic Flow Control
One general world model for pumps, learning how equipment behaves across operating conditions, then extended to engines.
- Why it matters
- Precise flow control lowers operating cost and reduces energy use and waste, helping make harder resources economically feasible.
- Coverage
- A single model across pump types: mud, slurry, process and water pumps. Engines next.
- Predictive control
- Forecasts equipment response to inform setpoints and control decisions ahead of process changes.
- Anomaly detection
- Flags departures from modelled behaviour (e.g. cavitation, wear, blockage signatures) before they become failures.
- Energy optimisation
- Identifies operating points that deliver the required flow at lower energy use.
- Data
- Trained on time-series from the historian via the Unified Namespace; runs on site at the edge.
- Status
- Research stream with monitoring in the field.
Her sketch → our blueprint
“It has really good eyes. Even in the dark and the dust!”
Engineering reality · Stream 03
Spatial intelligence
Machines that perceive, map and understand their physical surroundings, and act on plain-language instructions.
- Why it matters
- A precise picture of the site enables selective extraction: less waste rock moved, less land disturbed, and lower-grade deposits that are more economical to work.
- Models
- Vision-language-action (VLA) models that connect perception, an instruction and a physical action.
- Perception
- Designed for degraded conditions: dust, vibration, glare and low light, typical of pits, plants and haul roads.
- Stock take
- Camera-based counting and reconciliation of inventory, stockpiles, spares and consumables against the asset register.
- Mapping & twins
- Site mapping that feeds the live digital twin, keeping the spatial model aligned with the physical site.
- Asset tracking
- Location and state of mobile and fixed assets, linked to asset management and the historian.
- Status
- Camera-based stock take is a field product; VLA is research.
Technical foundation
Shared by every stream and product.
- Edge-deployed, air-gapped
- Runs on an on-site edge GPU rack. Operates without public cloud connectivity; operating data stays under the operator’s control.
- Silicon-agnostic compute
- Not tied to a single accelerator vendor; models are deployed to the hardware available on site.
- Universal protocol converter
- Connects to existing PLCs, SCADA, sensors and cameras across industrial protocols, without rip-and-replace.
- Unified Namespace (UNS)
- One live, structured data layer for every machine, sensor and system on site, the single source the models read from.
- Industrial historian
- Time-series capture and query at industrial scale: 4.5T data points per day.
- Simulation & digital twin
- Behaviour is tested in simulation against a live physics twin before it reaches equipment.
- Data quality flywheel
- Field data is cleaned, labelled and fed back into training, so models improve with every site.
- Vision-language-action AI
- Shared perception-to-action models used across Heavy Lift, Spatial and robotics command.
In the field | Product
Robotics doing real work on site.
Machines that see. Systems that coordinate. Start with one capability and expand across the site on the same stack.
Featured
Camera-based stock take
Count, measure and reconcile inventory straight from camera feeds: stockpiles, laydown yards, spares and consumables. No clipboards, no walk-downs in hazardous areas, no waiting for month-end.
- Continuous counts from fixed and mobile cameras
- Reconciles against the asset register and historian
- Runs on the edge, so footage never leaves site
HSE prediction
Predicting incidents before they happen, from live operating data.
Human + robot workforce
One org structure where people and machines share roles, tasks and accountability. Command and control keeps people overseeing every machine.
Hive Mind
Fleet-wide coordination: every robot learns from every other.
Operations management
Live planning, dispatch and performance across sites.
Asset management
Track, maintain and optimise every asset across sites.
Unified Namespace (UNS)
One live data layer connecting every machine, sensor and system.
4.5Tdata points / day
Industrial data historian
Every signal from every asset, captured and queryable.
Powered by OmegaA full AI software factory. Builds, tests and ships the software that runs the fleet.
Developers
Build on Robotyx.
Your agents, apps and scripts can build on top of Robotyx. The developer platform opens your organisation’s operations data through a remote MCP server and a versioned REST API, authenticated with an API key and scoped to your organisation.
Headline capability
Robotyx MCP
Connect AI agents and tools to Robotyx through a remote Model Context Protocol server, authenticated with your API key and scoped to your account. Its tools wrap the REST API: list assets and sites, read purchase orders, query Unified Namespace signals, and raise purchase requests when the key has write scope.
{
"mcpServers": {
"robotyx": {
"url": "https://developer.robotyx.systems/mcp",
"headers": { "Authorization": "Bearer <your API key>" }
}
}
}REST API
A versioned REST API over your assets, sites, Unified Namespace points, procurement and CRM, with cursor pagination, scopes, rate limits and an OpenAPI 3.1 reference.
Keys and scopes
Every call carries an API key with scopes and a live or test environment, and is filtered to its account: one organisation’s key can never read another’s data.
CLI
A command-line tool for scripting, with JSON output on every command. Coming in a following release.
Industries
Built for the hardest environments on Earth.
Mining
Haul roads, pits, crushers and stockpiles. Remote, dusty, disconnected sites where every person in a hazardous zone is a risk you want to remove, and every tonne of waste rock moved is a cost.
Critical minerals
Processing plants and flow-heavy circuits where pumps, slurry and reagents decide throughput. Tighter control means less energy and less waste, with sovereign control of operating data.
New forms of energy
Pumps, engines and power systems for new forms of energy, where precise control helps make complex resources economically feasible. Our team has built and commercialised industrial automation; Robotyx brings that discipline to Physical AI.
Global footprint
Where we operate, and where we’re going.
5
OperatingAustralia (Group HQ Brisbane) · India · PNG · UK (University partnership) · US
5
Coming soonCanada · Mexico · Bahrain · UAE · South Africa
Group HQ (Brisbane)★R&D HQ (Delaware, USA)
How it works
From the edge to command.
-
01
Connect
An edge GPU rack goes on site and connects to existing sensors, cameras, PLCs and SCADA through a universal protocol converter. Air-gapped by design.
-
02
Unify
Every signal lands in the Unified Namespace and historian: one clean, time-aligned model of the site.
-
03
See & model
Monitoring, stock take and world models turn raw data into a live physical twin that predicts what equipment will do next.
-
04
Command
Industrial agents decide; robotics command acts, within site safety rules, with operators supervising from a safe distance.
SeesVisibility and operating data
ThinksAgents and a live physics twin
ActsRobotics command across the site
Team
The people building it.
Operators from the industries they’re automating, with automation leadership at Shell and Halliburton.
Titto Thomas
Founder
Second-time founder with 20 years in industry. Led commercialisation of Shell’s SCADADRILL drilling-automation program. Ex-Halliburton.
Nick Shelton
Chief of Staff
Leads strategy, partnerships and commercial growth.
Daniel Dangoor
Finance Director
Ex-Goldman Sachs macro, Brevan Howard.
Nathan Maroney
Board Advisor
Tech-Source; ex-Rockwell Automation.
The future
More productive industries. Wider access to opportunity.
When energy and materials are produced safely, sustainably and affordably, more of what they make possible reaches more people.
A business, built on her terms.
Accessible, assistive technology that just works, so ambition is the only limit.
A first robot design.
A refurbished computer on the kitchen table, and the drawing becomes the start of something real.
The next world-changing idea is already out there. Let’s help more people build it.
ROBOTYXBuilding the foundations of possibility.Backed by Tryfecta.
Contact
Let’s build it together.
Talk to us. Operators, OEMs and partners: tell us about your site, and we’ll show you what autonomy looks like on it.
R&D HQDelaware, USA
Group HQBrisbane, Australia