FAQs

Fanuc FOCAS, Siemens S7, Mazak, Haas and Mitsubishi on the CNC side; Allen-Bradley, Delta, Modbus TCP and EtherNet/IP for PLCs; and OPC-UA and MQTT as the open standards. They are drivers, not gateways you have to hand-build, and none of them locks you to a vendor.

Rugged HMI and edge hardware that taps your controllers and sensors securely. It goes in without stopping production and without replacing anything you already run.

Yes. CNC, PLC, robots, standalone sensors and PLC-less assets all connect. Where there is no controller to tap, retrofit sensors for vibration, current and temperature carry the machine's state instead, so the oldest asset on the floor appears in the same live view as the newest.

Access is role-based. Operators, supervisors and management each see the part that matters to them, and only that part.

Every stop is captured and classified the moment it happens, then ranked by what it costs. That is what turns a downtime total into a list of losses you can work through in order.

Both ways. Open APIs and standard industrial protocols push live production, downtime and quality into your ERP, MES and BI tools, and read back from them.

Both, and they are the same system. The app is live on any device and role-based, from the control room to a phone on the floor, with alerts the moment reality drifts from plan.

Every signal is cleaned, normalised and unified into one model, so there is a single source of truth rather than a plant view and an office view. Your ERP, MES and BI read the same numbers your control room does, at the same moment.

Controllers and sensors are polled continuously and the model is current to the second across the whole plant. Machine state is streamed second by second, not written up at the end of a shift.

Yes. The platform runs from a single monitored machine up to enterprise-wide across multiple lines and plant locations.

The same platform behind any camera also does assembly verification, PPE and safety compliance, zone and intrusion monitoring, process and anomaly monitoring, counting, sorting and inventory, OCR and traceability, ergonomics and activity analysis, and perimeter security.

Accuracy, false-accept and false-reject rates are measured against your acceptance criteria before go-live, not after. If it does not clear the criteria you set, it does not go on the line.

Inline, on the edge device beside the line. That is what lets it inspect every part at line speed and fire sort or alert actions immediately.

It is designed to go the other way. New edge cases from your line are fed back and the model is retrained, so accuracy compounds instead of decaying.

Yes, where you want it to. A flagged part can trigger PLC sort or reject signals inline, alert an operator, and leave a traceable record, all from the same decision.

No. It works alongside the process you already run and plugs into major MES and SCADA platforms, so it becomes another gate in your system rather than a parallel one.

Audit-ready logs for traceability and compliance, with the image kept alongside every decision, so a judgement can be reviewed later rather than taken on trust.

Yes. It reads labels, lot codes, serials and barcodes, checks print quality, and builds a traceability record from what it reads.

Models are trained and augmented specifically for it: lighting, orientation and rare-defect variation are part of the training set rather than something discovered on the line.

No, and that is the point. Every defect and image flows into the same intelligence layer as your machines, so quality sits beside OEE and downtime in one view instead of in a separate system.

SMART Patrolling does autonomous security and surveillance rounds: checkpoint navigation, perimeter watch and incident verification. SMART Assistance is a task unit for heavy, hazardous and repetitive work: material movement, work in high-risk zones and jobs that cause ergonomic injury. Both report into the same platform.

Gantries, cable trays, gratings, kerbs and wet or uneven floor stop a wheeled platform. Four legs keep a steady sensor payload over all of it, which is what makes an unmanned round possible at the hours nobody wants to walk it.

No. Routes are planned by the robot around whatever is actually in the aisle that shift. There are no floor magnets, no guide tape and no line marking to maintain when the layout changes.

LiDAR and depth sensing for navigation and obstacle avoidance, HD and thermal vision for incident verification and heat signatures, and gas and acoustic sensing for leaks, smoke and abnormal sound. Decisions are made on board.

Analogue and digital gauge readings, thermal hotspots on motors and panels, and leaks or abnormal noise. Every reading is timestamped to the route point it was taken at.

That is one of the main reasons to use one. Confined spaces, high heat and live electrical zones get inspected without putting a person in them, and the task unit handles extreme heat, chemical and confined areas.

Yes. Roofs, stacks, high racking, yard stock and the outer fence line are a scaffold job or a long walk on foot; from the air they are a scheduled flight, and the footage lands in the same place as everything else.

The unit gives first-level visual confirmation with HD or thermal imaging, classifies what it is seeing, and transmits the feed straight to the control room. On-site verification happens within minutes rather than at the next manual patrol.

At any point. Edge autonomy covers navigation and routine decisions without constant remote control, and a human can always take over or review what it did.

In four steps: we map the task, the safety zones and the integration points with your team; the robot learns the environment, checkpoints and no-go areas; a supervised pilot runs alongside your people while behaviour, logs and alerts are validated; then feeds connect to the control room and it rolls out to more routes or shifts.

No. There is no catalogue to pick from. The patterns on the AI Agents page are examples of problems real teams have described, not products on a shelf; the engineering underneath is the same and it gets pointed at your process.

Then we say so. Working out whether an agent is the right answer is part of scoping, and telling you it is not is a possible outcome of it.

A typical build runs 12 to 20 weeks, depending on how many systems it has to touch. You see working software well before the end of that.

Limits set before anything is built. Scoping records what it must never do alone, and where a human has to stay in the loop is decided at design time rather than added afterwards. Nothing acts past the approval you configured.

Because that is built as its own piece of work: test sets, scored runs, thresholds it has to clear before it goes live, and a defined way for it to stop and ask instead of guessing.

It says so. An agent that invents an answer is worse than one that declines, so not knowing is a supported outcome rather than a failure mode.

Yes. Manuals, SOPs, drawings, contracts, tickets and history become queryable, and answers come back with the source and revision cited. Permissions are respected, so people see only what they are allowed to.

The ones you already run: ERP and CRM through to MES, ticketing and email. The point of the integration work is that a process completes without a hand-off queue in the middle of it.

No. The same pattern is used in manufacturing, logistics and supply chain, finance and accounting, banking and insurance, customer service, IT and internal ops, healthcare and pharma, construction and projects, retail and e-commerce, energy and utilities, legal and compliance, and HR. If the work is repetitive, judgement-heavy and spread across systems, the shape is the same.

TARA is the operations agent, and it is the one that runs today. It reads live machine, downtime, production and quality data, explains which loss driver and which asset and which moment, and can notify maintenance or set a standing watch. The other patterns on that page are builds scoped to your systems.