Layer 3  ·  Signal  ·  ConnecTwin™

Every machine talks.
Most were never asked.

ConnecTwin™ connects what your plant already owns · a 1978 mechanical press, a 2005 Fanuc lathe, a 2024 robot cell · into one live data stream. Nothing replaced. No PLC program touched. Read-only by construction, because that is the only version your OT team will sign.

Five connectivity tiers
40+ protocol drivers
Zero PLC changes
// MACHINE 01 · 1978 PRESS
T4 · CT CLAMP
// NAMESPACE
4 / 4 resolved
// DIRECTION
READ ONLY  ·  OUTBOUND
T1–T5
Every machine classified. None excluded.
40+
Protocol drivers in the shipped library
~1 hr
Per machine for a no-port retrofit
0
PLC programs modified. Ever.
In 30 seconds

What ConnecTwin™ does.

Machine connectivity is where almost every Indian digital manufacturing programme stalls. Not at the dashboard, not at the analytics · at the moment someone walks the shop floor and finds that eleven of the forty machines have no port, no documentation and no one left in the plant who remembers the controller. ConnecTwin is the layer built for that walk.

// 01

Classifies every machine

We survey the plant and place each machine in one of five connectivity tiers · from native OPC UA down to no digital output at all. You get a firm method and a firm price per machine before anything is bought.

// 02

Reads it without touching it

Drivers for the controller families actually found in Indian plants. Where no protocol exists, a clamp-on current sensor or a discrete I/O tap. The machine never receives a command from us · the read-only path is compiled in.

// 03

Puts it in one namespace

Every reading is resolved to a named asset before it is stored, in an ISA-95 structure. Above that line, a current clamp on a 1978 press and a native OPC UA robot produce the same record shape.

The question everyone asks first

“Can my machines be connected?”

Yes · all of them. A machine with no ethernet port, no controller and no documentation still gives you run, stop, cycle count and energy per unit through a clamp-on current sensor fitted in about an hour, with no panel entry.

What changes from machine to machine is not whether it can be connected. It is the method, the cost and the resolution of what comes back. That is what the five tiers make explicit · before you commit a rupee, you know which tier every machine sits in and what each one will actually tell you.

TierWhat the machine isHow we read itTypical effortWhat you get
T1
Connect Direct2015–2026 · native
Speaks a modern open standard already. Newer Siemens, Fanuc, Rockwell, Mitsubishi, most CNCs bought in the last decade, and almost everything specified since 2018. Direct read over OPC UA, MQTT Sparkplug B or MTConnect. Configuration only · no hardware, no gateway change. Hours
Live the same day
Full structured tags · cycle, state, part count, alarms, spindle load, programme number, tool life.
T2
Connect Bridge1995–2018 · documented
Has a controller and a documented protocol, but no open standard interface. The largest single population in Indian plants and the one every integrator underprices. Read through our driver library · Modbus TCP/RTU, Siemens S7 (S7-300 to S7-1500), Fanuc FOCAS, Mitsubishi MELSEC, EtherNet/IP, Profinet, serial. 1–2 days
Per machine family
Nearly everything T1 gives, depending on what the controller exposes. Usually indistinguishable on the dashboard.
T3
Connect AdaptUndocumented · proprietary
A controller exists and a protocol exists, but the documentation is gone, the OEM has exited India, or the interface is proprietary. Often has a display and nothing else. We build a bridge for that machine family · protocol capture, mapping, validation against the machine's own display. 2–3 weeks
Once, per family
Same as T2 once built. The bridge then joins the shipped library at our cost · never billed to you as custom development.
T4
Connect Sensepre-1995 · no output
No digital output of any kind. Mechanical presses, old lathes, rebuilt imports, machines whose controller died and was replaced with a relay panel. Often no drawings left either. Clamp-on CT sensors on the supply, discrete I/O taps on existing contactors, vibration or proximity where the cycle needs it, operator entry for reason codes. ~1 hour
No panel entry, no downtime
Run, stop, cycle count, load profile, energy per unit, downtime with operator reason codes. Not part-level quality data · we say so.
T5
Declared BlockedPolicy, not physics
Technically connectable, but blocked by an OEM warranty clause, a validated-system lock in a regulated line, or a customer contract that forbids third-party access. Nothing is installed. The machine is recorded in the register with the specific clause that blocks it and the named owner of that decision.
Reviewed each renewal
A declared gap on every report, not a silent hole. An auditor sees why the row is empty instead of assuming the data was lost.

There is no sixth tier called “not possible”. In four years of plant surveys we have not found a machine that gives nothing · we have found machines where the honest answer was T4 and the buyer had been quoted T1 by someone who had not yet walked the floor.

Want your own plant tiered?

Send us your machine list · make, model, year, controller if you know it. We return a tier against every line item and a firm price. Most plants get it back inside a week.

Protocol coverage

What does your machine
actually speak?

Most plant engineers know the make and the year. Very few know the protocol, because nobody had a reason to ask until now. This is the shortest route from what is written on the nameplate to what we can read off it.

Controller familyTypical vintageHow we read itTier
Fanuc 0i, 16i/18i/21i, 30i/31i series1995 → todayFOCAS over ethernet, or FOCAS via the HSSB card on older controls. MTConnect adapter where fitted.T2
Siemens SIMATIC S7-300, S7-400, S7-1200, S7-15001995 → todayS7 protocol direct on older ranges; native OPC UA server on S7-1500 and newer S7-1200 firmware.T2 T1
Siemens SINUMERIK 810/840D, 828D, ONE1990s → todayDrive/NC data via S7 or the machine's own OPC UA server on newer ranges.T2
Rockwell / Allen-Bradley SLC, MicroLogix, CompactLogix, ControlLogix1995 → todayEtherNet/IP CIP read. OPC UA on current ControlLogix firmware.T2 T1
Mitsubishi MELSEC FX, Q, iQ-R · M700/M800 CNC1995 → todayMELSEC / MC protocol over ethernet or serial. SLMP on newer iQ-R.T2
Omron CJ, CS, NJ, NX2000 → todayFINS protocol, or EtherNet/IP and OPC UA on the NJ/NX range.T2 T1
Delta, Schneider, LS, Wecon DVP, AS, M340, M580, XGB2005 → todayModbus TCP or Modbus RTU in almost every case · the workhorse of Indian panel builds.T2
Haas, Okuma, Mazak, DMG Mori imported machining centres2005 → todayMTConnect where fitted, otherwise the OEM's own ethernet interface or MDC option.T1 T2
Injection moulding Engel, Arburg, Milacron, Windsor, Ferromatik1998 → todayEUROMAP 63 file interface on older machines, EUROMAP 77 over OPC UA on current ones.T2 T1
Weighing, testing and lab instrumentsanyRS-232 or RS-485 serial capture, or the instrument's own export. Common in pharma and food QC.T2
Existing SCADA / historian WinCC, FactoryTalk, AVEVA PI, iFIXanyWe read from the system you already own rather than re-instrumenting the machine. Years of history come with it.T1
Relay panel, mechanical press, rebuilt import controller dead or never existedpre-1995Clamp-on CT on the supply, discrete I/O tap on a contactor, proximity or vibration for cycle detection.T4
OPC UAMQTT Sparkplug BMTConnectModbus TCPModbus RTU Siemens S7Fanuc FOCASMELSEC / MCSLMPEtherNet/IP CIP ProfinetProfibusFINSEUROMAP 63 / 77BACnet IEC 61850Modbus over TCP gatewayRS-232 / RS-485Dry contact I/O4–20 mA

If your controller is not on this list, it is very likely a T3 · we capture the protocol, build the bridge, and it ships in the library for the next plant. You are not billed for that work and you do not wait for a roadmap.

The machine nobody could connect

A 1978 press has no port.
It still has a current draw.

This is the machine that kills Indian digitalisation projects. It is usually load-bearing for the line, it is usually the oldest thing on the floor, and every proposal quietly leaves it out. We do not leave it out, because the plant report is worthless with a hole in the middle of it.

// Method

A clamp, not a rewire

A split-core current transformer clips around the existing supply cable. No panel entry, no isolation permit, no machine downtime, no warranty exposure. Typically fitted in under an hour during a running shift.

  • Nothing electrically connected to the machine
  • No change to any existing circuit
  • Removable without trace
// What you get

More than you expect

Current signature alone separates idle, loaded and stopped with high reliability, and the cycle shows up as a repeating pattern you can count.

  • Run / stop / idle state, continuously
  • Cycle count and cycle time variance
  • Load profile per cycle
  • Energy per unit in kWh/piece
  • Downtime with operator reason codes
// What you do not get

And we say so on the report

A current clamp cannot read part numbers, programme names, tool life or in-process measurements, because that information never leaves the machine in electrical form.

  • No part-level genealogy from the machine itself
  • No SPC values without a separate gauge feed
  • Resolution is declared on every report that uses it

The age of the machine changes what it costs to connect. It does not change what you get afterwards.

Above the namespace layer, a clamp on a 1978 press and a native OPC UA robot produce the same asset record, the same tag structure and the same report shape. A reader cannot tell from the dashboard which was which · only the provenance stamp on the data point says so, and that stamp is there deliberately, for the auditor.

Fidelity

Coverage is complete.
Resolution is honest.

Every connected machine is labelled with the resolution it actually delivers. This is the difference between a plant report you can put in front of a customer auditor and one you cannot.

F1

Continuous

Sub-second to few-second sampling, straight off the controller or the sensor. State changes and micro-stops are visible as they happen.

Supports micro-stop analysis, cycle-time variance and live OEE. Typically T1 and T2 machines.
F2

Event

Captured on change · a start, a stop, a completed cycle, an alarm. Nothing between events, but every event is timestamped at source.

Supports availability, downtime Pareto and cycle count. Common on T3 and well-instrumented T4.
F3

Periodic

Shift-level or batch-level figures, often part machine-derived and part operator-entered. Honest about the human share.

Supports shift OEE and energy per unit. Cannot produce micro-stops · we never imply otherwise.

Why this matters commercially: if your business case rests on recovering short stops, the machines carrying that case need F1. We will tell you that during the assessment rather than after the invoice, even when it means a larger scope on fewer machines.

Security posture

Read-only is a property
of the code, not a promise.

The first meeting that decides whether this project happens is usually with IT and OT security, not with the plant head. So the architecture is built to survive that meeting rather than to argue its way through it.

// 01

No write path exists

The read-only behaviour is compiled into the driver, not set by a configuration flag a support engineer could flip. There is no command function in the shipped build to enable.

// 02

Outbound only

The edge gateway initiates every connection. No inbound port, no VPN into the plant network, no firewall exception to justify to your group security team.

// 03

Air-gap supported

Where a customer contract, a PSU policy or a defence requirement demands it, the whole stack runs inside your boundary with no outbound connection at all. Same containers, different orchestration.

// 04

Nothing crosses the control loop

We sit beside the control layer, never inside it. No machine safety function, interlock or validated recipe is in the data path, which is what keeps a regulated line's validation intact.

// 05

Buffered for Indian reality

Seven days minimum local buffer on the gateway, replayed in order on reconnect. Power cuts and network drops do not produce a quietly wrong OEE figure · unrecoverable gaps are declared.

// 06

India-hosted, DPDP-aligned

Cloud deployments run on Indian infrastructure. Bring your own tenancy where data residency is contractual, or go fully on-premise where it is absolute.

Send it to your security team first.

We will give you the architecture note, the port and protocol list and the read-only statement before any commercial conversation. Most OT reviews close in one call because there is nothing to negotiate.

Deployment

Three ways to run it.
One codebase behind all three.

Nobody chooses a deployment model on technical grounds. It is decided by a clause · in a customer's supplier agreement, in a PSU tender, in a defence requirement · and it is usually decided by someone who was not in the room when the business case was made. So all three are built and supported today, rather than one being real and two being roadmap.

Option 01

Edge · on-premise

Air-gapped plants · defence · PSU · restricted pharma

Everything that touches a machine and everything that stores what it said runs inside your own boundary · gateway, historian, broker, dashboards. No outbound connection is required at any point. For some Indian suppliers this is not a preference at all: it is the only configuration their largest customer's contract permits.

  • Fully air-gapped operation supported
  • Data never leaves the plant network
  • Runs on modest on-site hardware
  • Same containers as cloud, different orchestration
Most chosen
Option 02

Hybrid

Most multi-plant Indian groups

The gateway keeps doing its job on site · reading machines, resolving tags, buffering · while anything that benefits from scale sits upstream: group roll-up, plant-versus-plant comparison, long retention. The link between the two is allowed to fail, because nothing on the shop floor waits for it.

  • Seven days minimum local buffer, replayed in order
  • Shop floor unaffected by connectivity loss
  • Bring your own cloud tenancy where residency is contractual
  • Group view without centralising everything
Option 03

Cloud

Single plants · fastest start · lowest entry cost

We run it on Indian infrastructure and you procure nothing. For a single plant this is the shortest distance between a signed assessment and a screen on the shop floor. The gateway still dials out rather than being dialled into, so even the cloud option does not widen the attack surface your security team is protecting.

  • India-hosted, DPDP-aligned
  • No server procurement, no capex line
  • Outbound-only · no inbound ports or VPN
  • Upgrade path to hybrid without re-implementation
// Gateway hardware
Industrial PC or DIN-rail unit · no rack, no server room
// Footprint per line
One gateway typically covers a full line, not one machine
// Local buffer
7 days minimum, replayed in order on reconnect
// Switching later
Cloud to hybrid to on-prem without re-implementation

Same containers. Different orchestration.

The air-gapped build is the same product, not a stripped cousin of it. The moment a vendor maintains a second codebase for on-premise customers, that codebase starts falling behind · one roadmap gets the attention and the other gets patches. A plant running fully air-gapped in a defence cluster takes the same release, on the same day, as a plant running cloud. It costs us engineering discipline and it is the single thing most worth checking about any vendor selling you an on-premise industrial product in India.

Not sure which one your contracts allow?

Send us the data clause from your largest customer's supplier agreement. We will tell you which deployment model satisfies it, in writing, before anything is quoted.

How it works

Four steps. Nameplate to namespace.

A single line is usually live in six weeks. Nothing in this sequence requires a machine to be replaced, a PLC to be reprogrammed or a shift to be stopped.

// 01 · SURVEY

Walk the floor, tier every machine.

Two to three weeks. Every machine is photographed, tagged, identified and placed in T1 to T5, with the method and the price attached. You own the register whether or not you proceed.

// 02 · CONNECT

Drivers, bridges and clamps.

T1 and T2 machines come up in hours or days on the shipped library. T4 machines get sensors fitted during a running shift. T3 bridges are built in parallel and do not hold up the line.

// 03 · RESOLVE

Every reading gets a name.

Raw tags are mapped to named assets in an ISA-95 structure before anything is stored. A reading that cannot be resolved to an asset is not silently kept · it is flagged.

// 04 · PUBLISH

One stream, many consumers.

The resolved stream feeds SmarTwin™ dashboards, PredicTwin™ models, VolTwin™ energy, TrusTwin™ evidence packs, your existing MES, your ERP, or your own team's code over the API.

Unified namespace

The part that makes it
worth doing once.

Connecting machines is the visible half. The half that decides whether you ever have to do this again is what happens to the data in the two seconds after it is read.

A tag called DB12.DBW4 is not data. It is a location.

ConnecTwin resolves every raw tag to a named asset in an ISA-95 hierarchy · enterprise, site, area, line, cell, machine · and publishes it on a single unified namespace over MQTT Sparkplug B. Which means the next system you add reads a structure, not forty different controllers. The second project costs a fraction of the first, and the third costs almost nothing. That is the whole economic argument for doing the signal layer properly instead of point-to-point.

// Consequence 01

Your data is not locked to us

The namespace is an open MQTT Sparkplug B broker and a REST API. Your team can build on it, your existing MES can subscribe to it, and if you replace anything above the signal layer later, the signal layer stays.

// Consequence 02

The second plant is faster

Machine families repeat across an Indian group. A driver built at plant one is configuration at plant two. Multi-plant groups usually find plant three going live in under half the time of plant one.

// Consequence 03

Provenance travels with the value

Every data point carries where it came from · machine-captured, sensor-derived or human-entered · from capture through to the printed report. That stamp is what answers an auditor in ninety seconds.

What you get

Deliverables, not a statement of intent.

01

Connectivity register

Every machine, its tier, its method, its fidelity level, and the price to connect it. Yours to keep, and useful even if you buy nothing else.

02

Edge gateway, installed

Sized to the line, running the driver library, buffering locally for seven days minimum, outbound-only.

03

Live asset-resolved stream

Every machine publishing to one ISA-95 namespace, with provenance stamped at capture.

04

Retrofit sensor kits

CT clamps, discrete I/O taps, proximity and vibration where the tier calls for them. Fitted without panel entry.

05

Open API and broker access

REST, MQTT Sparkplug B, OPC UA out and scheduled exports. Your team builds whatever view it wants without waiting for us.

06

Declared gap statement

Every T5 machine and every unrecoverable outage listed with its reason, so the report is defensible rather than merely tidy.

Where you start

Three ways in. None of them
need a board note.

Every one of these ends with something in your hands · a register, a dashboard, a price. None of them requires you to commit to a plant-wide programme to find out whether this works on your floor.

Option 01 · fastest proof

Energy baseline

Clamp sensors on the ten or fifteen highest-consuming machines. No controller access, no IT involvement, no procurement of anything permanent.

A working dashboard in about a week, with kWh per unit by machine and shift. Usually pays for the next step on its own.
Most chosen
Option 02 · the usual entry

Connectivity assessment

We walk the plant, tier every machine, and hand you the register with a method and a price against each line. This is the document that ends the guessing.

Two to three weeks, ending with a firm price for the full connection. The register is yours whether or not you proceed with us.
Option 03 · full pilot

One line, live

Every machine on a single line connected at its tier, resolved to the namespace, feeding a live dashboard on the shop floor.

First dashboard on the floor by week six, sign-off at week eight to twelve. The reference the rest of the plant is scoped against.
Questions we get asked

Straight answers,
including the unflattering ones.

These are the questions Indian plant, maintenance and IT heads actually put to us in the first meeting. Where the honest answer is “no” or “not from that machine”, it says so.

Through the machine's electrical behaviour rather than its controller. A split-core current transformer clips around the existing supply cable and gives run, stop, cycle count, load profile and energy per unit · fitted in about an hour, with no panel entry and no downtime. If the machine has a serial port, an RS-232 or RS-485 capture usually gives more. If it has a Fanuc, Siemens or Mitsubishi control with an ethernet card fitted later, FOCAS, S7 or MELSEC will read it directly. The absence of an ethernet port rules out one method, not the machine. Send us the make, model and year of the machine you think is impossible · put it on the list and we will tell you which method applies to it before anyone visits your plant.
Yes · that is the only way it works. The read-only path is compiled into the driver, not set by a configuration flag. We read existing registers and tags; we do not add code blocks, change scan cycles, or write values. Nothing about the machine's control logic, safety function or validated recipe changes. This matters for two reasons beyond security: an OEM warranty is not voided by reading, and a validated line in a regulated plant does not need requalification for a read-only observer sitting beside the control layer. Ask for it and we will send you the read-only architecture note · it is the document your maintenance head and your machine OEM will both ask to see.
It depends on the controller, not the machine. Fanuc controls use FOCAS. Siemens S7-300 and S7-400 use the S7 protocol; S7-1500 and newer S7-1200 firmware expose a native OPC UA server. Rockwell uses EtherNet/IP CIP. Mitsubishi uses MELSEC/MC or SLMP. Omron uses FINS. Delta, Schneider, LS and most Indian panel builds use Modbus TCP or RTU. Injection moulding machines use EUROMAP 63 or 77. Machines from 2018 onward increasingly ship OPC UA or MTConnect natively. If you send us the make, model and year, we will tell you which one applies before anyone visits. Send us your machine list and we will return the protocol against every line item, free, usually inside a week.
There is no honest per-machine figure, because the cost is set by the tier the machine sits in. A T1 machine is configuration only. A T2 is a day or two of driver work. A T4 needs a sensor kit and about an hour of fitting. A plant of forty machines with a good T1 and T2 mix costs a fraction of a plant of forty pre-1995 machines · and a vendor quoting you a per-machine rate before walking your floor does not know which one you are. What is fixed is the assessment: two to three weeks for a single Indian plant, priced against machine count rather than turnover, ending with a firm itemised price for the full connection rather than a range. The register it produces is yours to keep whether or not you proceed with us. Send the machine list and you get a firm figure rather than a range.
Yes, with one honest exception. Every machine sits in one of five tiers, and T1 through T4 all produce usable data · the oldest, portless machines still give run, stop, cycle count and energy per unit. The exception is T5: machines blocked by an OEM warranty clause, a validated-system lock or a customer contract that forbids third-party access. Those are a policy problem, not a physics problem, and we record them in the register with the specific clause and the named owner rather than leaving a silent hole in your report. Send us the list and we will mark the T5s honestly, including the ones we cannot connect and why.
Yes, and usually in one call · the architecture is built around that review rather than arguing its way through it. The edge gateway connects outbound only · no inbound ports, no VPN into the plant network, no firewall exception for your group security team to sign off. Read-only is enforced in the driver rather than promised in a contract. Nothing sits inside the control loop. Where a customer clause or a PSU policy requires it, the whole stack runs on-premise and fully air-gapped with no outbound connection at all. We hand over the architecture note and port list before any commercial conversation, so the review usually closes in one call. Ask for the architecture note and port list · we send it before any commercial conversation, so your security team can start reviewing while procurement is still thinking.
By never publishing a figure across a gap you cannot account for. Two things have to be true. First, the gateway has to survive the outage: ours buffers to local disk and replays in sequence on reconnect, sized so a plant losing power daily does not lose a shift of data. Second, and this is the part most systems skip, anything that still cannot be recovered is marked as a declared gap rather than averaged away. An availability figure computed across four silent hours is not slightly wrong, it is confidently wrong, and someone will hold a review meeting on it. Tell us your typical outage pattern when you enquire · it decides the buffer sizing and the gateway spec, not the price. Get your plant sized.
Both, on different machines in the same plant. Anyone who tells you a clamp replaces controller data is selling. A clamp gives state, cycle count, load profile and energy reliably. It cannot give part numbers, programme names, tool life, in-process measurements or SPC values, because that information never leaves the machine in electrical form. What it does give is complete coverage · your plant report has no blank rows on day one. Every data point is labelled with how it was captured, so a figure that is 70% human-entered says so on its face rather than hiding inside an average. Which of your machines need which is exactly what the tiering answers, before you spend anything. Send the machine list.
A unified namespace is a single structured place where every machine's data is published under a name that means something · site, area, line, cell, machine · instead of a raw controller address like DB12.DBW4. It follows the ISA-95 model and, in our implementation, publishes over MQTT Sparkplug B. The commercial reason to care: the next system you add reads one structure rather than forty controllers. Your second project costs a fraction of the first. Without it, every new application means re-integrating the whole plant, which is how a group ends up paying for connectivity three times. If you want to see what your own plant looks like as a namespace rather than as a theory, start with the assessment · that is precisely what it produces.
If you have SCADA covering the machines that matter, we read from it rather than re-instrumenting · and years of existing history come with it. That is usually the cheapest T1 in the plant. But SCADA typically covers the process side and leaves the discrete machines, the older presses and the utilities uncovered, and it stores tags rather than asset-resolved records. The gap we fill is the machines your SCADA never reached and the structure your historian was never asked to provide. Tell us what your SCADA already covers and we will scope only the gap rather than the whole plant, which is usually a much smaller number than people expect.
The ones at T1 and T2, and usually fewer than the quality team assumes. Clause 9.1.1.1 asks you to monitor and measure manufacturing processes and keep the evidence. A machine read at controller level produces that evidence continuously and timestamped at source. A T4 machine on a clamp produces state, cycle count and energy, which supports availability but not in-process measurement, so it cannot carry the SPC half on its own. The uncomfortable case is a plant that has already committed to a customer audit date with eleven machines it has never instrumented. The tiering tells you which machines hold your evidence before you make that commitment, not during the audit. Send the customer’s requirement and your machine list and we will map one against the other, clause by clause.
No, and we would advise against it. The usual sequence is one line, live in six to eight weeks, used as the internal reference the rest of the plant is scoped against. An even lighter start is an energy baseline on the fifteen highest-consuming machines, which needs no controller access and no IT involvement and gives a working dashboard in about a week. Both leave you with something real before anything plant-wide is committed. Partial connection is not a compromised deployment · the namespace is designed to be extended, not rebuilt. The lightest possible start is the energy baseline · about a week, no controller access, no IT involvement. Ask for that one.
It becomes a T3 · we capture the protocol, map it, validate the readings against the machine's own display, and build the bridge. That typically takes two to three weeks and runs in parallel so it does not hold up the rest of the line. The bridge then joins the shipped driver library at our cost and is never billed to you as custom development. We take that position deliberately: a driver built once at an Indian plant should make the next Indian plant cheaper, not create a dependency. Send the make and model · if it is a T3 we will tell you upfront, with the timeline, rather than discovering it on site.
You own it, and yes, you can take it out. The namespace is an open MQTT Sparkplug B broker plus a REST API, with OPC UA out and scheduled exports available. Your own team can build on it, your existing MES or historian can subscribe to it, and anything you replace above the signal layer later leaves the signal layer intact. We would rather be kept because the layer is good than because leaving is expensive · and in this category, a buyer who feels trapped tells the whole industrial belt about it. Ask for the API and export documentation before you sign anything · we send it on request, which is a reasonable test to apply to every vendor you are considering, not only us.
Yes · fully air-gapped is a supported deployment, not a special build. The gateway, the historian, the namespace broker and the dashboards all run inside your plant boundary with no outbound connection. Defence suppliers, PSUs and certain restricted pharma lines need this because a customer contract or a government policy says so rather than because of a technical preference. The on-premise build runs the same containers as the cloud build under different orchestration, so it is never a fork and never a release behind. If you are air-gapped, say so at the first enquiry · it changes the sizing conversation, not the product or the roadmap.
You choose between three models. Cloud runs on Indian infrastructure and is DPDP-aligned, with nothing for you to procure. Hybrid keeps acquisition, buffering and time-critical compute at the edge while roll-up and long-term retention sit in the cloud · ours, or your own tenancy where residency is contractual. On-premise keeps everything inside the plant with no outbound connection. Most multi-plant Indian groups choose hybrid, because it gives a group view without centralising everything. Moving between the three later does not require re-implementation. Send us the data clause from your largest customer's supplier agreement and we will tell you in writing which deployment model satisfies it, before anything is quoted.
An industrial PC or a DIN-rail unit in the existing panel room · no rack, no server room, no air conditioning project. One gateway typically covers a full line rather than one machine, so the hardware count stays small even on a plant with a hundred assets. It buffers to local disk for a minimum of seven days, and it connects outbound only. For an on-premise deployment the compute sits alongside it on modest on-site hardware rather than anything specified for a data centre. The assessment states the exact gateway count and specification for your plant, so it never becomes a surprise line in the quote. Get yours scoped.
Start with the register

Send us your machine list.
We will tell you what each one can give.

Make, model, year, controller if you know it · a photograph of the nameplate is enough where you do not. We return a tier against every line item, the method, the fidelity level and a firm price. Most plants get it back inside a week, and it is the first time anyone has put that document in their hands.

India work emails only. No sales call until you ask.