Part 5 of 5 in our Edge to Enterprise series, and the last. We have argued for a common vocabulary, an edge that does real work, a common API and a clear line between moving bytes and moving meaning. Here is what we built on those ideas, what works on our test bench today, and what comes next.
Where the series has taken us
Over four posts we have made one argument, one layer at a time.
- A shared vocabulary. In 4 Surprising Truths About ISA-95 we showed that the plant's data needs an address and a meaning before any system can use it: enterprise, site, area, line, machine.
- An edge that does the work. In CATIM Edge IPC: the ultimate OT intelligence we argued that intelligence belongs at the data point, not three networks away, and that it must keep working when the network does not.
- One API for every application. In The End of API Chaos we explained why i3X lets an application be written once and used with any compliant platform.
- Bytes versus meaning. In i3X vs MQTT: One Moves Bytes, the Other Moves Meaning we showed that a message broker and an information interface are not rivals: each has its own altitude.
All four describe an architecture. This post describes the product we are building to deliver it.
The problem every plant still has
Ask a plant manager what their data could tell them and you get a long list. Ask when they will have it and the answer is usually "after the next integration project". The data already exists inside the controllers that run the machines. Getting it out, keeping its meaning, and putting it to work is where projects stall:
- Data is trapped. Each new consumer - a dashboard, an MES, a cloud tool - means another point-to-point connection, and renaming one tag quietly breaks something downstream.
- AI pilots never reach the line. A model that shines in a proof of concept lives far from the process, depends on the cloud, and nobody will let it near the control system.
- IT and OT pull in opposite directions. IT wants open interfaces and governance. OT wants determinism and no surprises. Both are right, and security too often comes last.
CATIM Nimbus: sense, decide, act
Since 2019 CATIM has built the Nimbus as the "brain" of an automation solution: a controller for any task that can be described as sense -> decide -> act. CATIM Nimbus is its next generation: the same brain, now software-defined. Control, data and AI run as software on one industrial unit, built on CODESYS, the widely used IEC 61131-3 programming environment.
Sense: configure, don't code
Tributary, the data-flow engine of CATIM Nimbus, reads your plant's devices over Modbus, the most common industrial protocol. In Tributary Studio, a browser editor, a technician says what to read, how to scale, filter, compare or debounce each value, and where the result goes. The change is applied from the browser; nobody rebuilds the PLC program. More on Tributary below.
Every value carries its health. In i3X vs MQTT we described the moment our bench PLC went down mid-demo and every client saw its values turn Bad instead of a frozen dashboard quietly lying. That is the same principle, applied at the source: a device that has stopped answering never looks like a process that is simply steady.
Decide: advise, don't override
The CATIMi Advisor runs on the controller itself, on its built-in AI accelerator: no cloud round trip and no extra box on the line. Running it there keeps the control cores free; it is not about raw speed.
On our test bench, a small model trained on what the controller reads from a simulated machine watched those values and the health of the link to the machine. When we made the machine misbehave - refusing replies, or jumping to an unusual value - the anomaly score rose within a second or two and raised an alarm on a live dashboard. The alarm cleared within about 40 seconds of the fault ending. Not every fault was caught: a value that froze inside its normal range looked normal to the model, and closing that gap is our next step. This was a simulated machine on a bench, not a customer plant.
One rule never changes: the model advises; the controller decides. Today the advice goes to people, on a dashboard. Next, it will reach your control logic as a checked, qualified input. It will never drive a plant output on its own.
Your data, your model
The Advisor is not a black box we ship and lock. It is a loop built around your own machine: record its normal running, train a model on a PC, and load it onto the controller, where it scores live on the AI accelerator. The result lands on the Unified Namespace, where any dashboard, historian or MES can subscribe, and is served through the i3X API, typed and carrying its quality. Our bench model was built exactly this way, on recorded data from the machine it watches. Today CATIM runs that loop with you. Next, the tools come to your engineers: recording set up from the console, a PC application that trains and signs the model, and a controller that loads only models signed by a key it trusts. Your CODESYS logic will read the score as a checked input, marked valid or not, so your logic decides what to do with it, never the model.
Act: start where your engineers already are
If your team programs PLCs today, CATIM Nimbus speaks their language: the same IEC 61131-3 languages, the same skills, the same libraries. Control logic runs on the controller and is designed never to wait on the network, for data or for advice. Your code stays yours and stays portable: CATIM extends CODESYS, it does not replace it.
Tributary: every data stream, shaped live
Most edge platforms move data. Few let you decide, while the plant runs, what that data becomes before anything else sees it. That is the job of Tributary, the data-flow engine of CATIM Nimbus.
The name is the idea. Small streams join a river upstream, and each one shapes the water before it reaches the main channel. Tributary works the same way. Each flow taps a protocol stream where it enters the controller and shapes it on the way in. It scales, filters, debounces and compares the values, combines them, logs them, and raises alarms on them. Only qualified data reaches the main channel: the platform's shared data hub, the Unified Namespace and the i3X API.
In Tributary Studio, a browser editor, an engineer draws those flows. A flow is saved from the browser, and the controller loads it within a second. Nobody rebuilds or restarts the PLC program, and the controller's published address space never changes, so every client connected to it keeps working.
Downstream, every output is a named, typed value the whole platform shares:
- Your CODESYS control reads it, so your logic can use a value the moment a flow publishes it.
- The Unified Namespace and the i3X API carry it, with its quality and timestamp, to every dashboard, historian, MES and AI agent.
- The CATIMi Advisor can learn from it and score it, and its advice can flow back in. On our bench, a flow already reads the AI's score, and a setpoint written through the i3X API, as inputs. Neither ever reaches a plant output until the flow arms it and the controller's own enable allows it.
So Tributary is not one more box in the data path. It augments the services already on the controller (control, the namespace, the AI) with any protocol stream you choose to tap, configured rather than coded. Today it speaks Modbus; more industrial protocols are planned.
Publish once, use everywhere
This is where the series comes together. CATIM Nimbus publishes its data once and lets every system above take what it needs, in the language that system already speaks:
- The i3X API, the open information interface from CESMII. A new application, or an AI agent, discovers what the data means without a briefing: the stranger test, passed. Our server reports 1.0 Compatible on the official i3X conformance suite against live controller data, and on our bench an AI agent answered plain-English questions about live PLC values through it, with no integration code written for the plant.
- A Unified Namespace over MQTT: one live, shared view of the plant's state, organised by the ISA-95 hierarchy you set (enterprise, site, area, line). Dashboards, historians and clouds subscribe without new integration work. It is the platform's data hub, and i3X is the typed window onto it.
- OPC UA, the established industrial standard, as the second door for SCADA and MES systems that do not speak i3X yet.
- An outbound connection to an industrial cloud platform. Nothing on the internet calls in.
ERP is reached through MES, as ISA-95 intends, never by wiring a machine straight into it. This is the altitude model made real: MQTT moves the bytes, the platform makes the meaning, and i3X serves it to every consumer, written once.
Secure by design
Every person, service and AI agent has its own identity and role. Every way in is a named, deliberate door. The control runtime is never exposed directly to the network, and cloud connections go outbound only. The EU Cyber Resilience Act's full requirements - secure defaults, security updates, and a software bill of materials - apply from December 2027, which is why they are in the design from the start.
What it means for you
If you are...
What CATIM Nimbus changes
An executive
Data and AI stop being an integration project. Open standards and CODESYS mean no proprietary lock-in, and outbound-only cloud links give you a security story you can explain to your board
A project manager
Time to first useful data is spent configuring flows in a browser, not commissioning custom code. A line or a site grows as a set of identical units publishing the same names and meanings, not a new integration each time
An engineer
You keep IEC 61131-3 and the tools you know. Every value arrives with its quality and timestamp, every application reads one typed API, and the AI advises your logic without ever taking control from it
Available today, and what comes next
In limited trials today:
- CODESYS control with Tributary flows, configured in a browser with no PLC rebuild
- The controller's data and the Unified Namespace through one i3X API, with a token and a role for each client
- A model trained on recorded machine data, scoring live on the controller, with its score on the Unified Namespace
- The CATIMi Advisor's anomaly score, computed on the controller and shown live on a dashboard
- An outbound connection to an industrial cloud platform
What we are building next:
- Your engineers train their own models from their own machine data, signed and loaded from the console
- The AI's score in your CODESYS logic as a checked input
- The controller's own data on the Unified Namespace
- The same names and meanings on every unit
- OPC UA as the second door for SCADA and MES
- The Nimbus Integrator Kit, so your own CODESYS applications can share and receive data on the platform by name
Products, or a partner to build with
Some teams want the platform. Others want the engineering behind it. CATIM offers both: CATIM Nimbus as a product, and CATIM's services to integrate it, or the same open architecture, into your plant - ISA-95 modelling, Unified Namespace design, i3X integration and CODESYS control.
If a data or AI project at your plant has been "nearly done" for too long, we would like to hear about it. We are looking for pilot partners who want to start on one machine, prove a measurable result, and grow from there. Get in touch.
The Edge to Enterprise series
- 4 Surprising Truths About ISA-95 - the vocabulary every system shares
- CATIM Edge IPC: the ultimate OT intelligence - why intelligence belongs at the edge
- The End of API Chaos - one API for every application
- i3X vs MQTT: One Moves Bytes, the Other Moves Meaning - moving bytes versus moving meaning
- Sense, Decide, Act: The Controller Is the Platform - the platform we built on these ideas (you are here)
Previous: i3X vs MQTT: One Moves Bytes, the Other Moves Meaning
About CATIM. CATIM - Creating Automation Technology In Motion - empowers people to engineer automation solutions that solve hard technical problems. CATIM Nimbus, the CODESYS edge controller with advisory AI, is the next generation of the Nimbus platform it has built since 2019.