// the platform

Train it on your history. Run it beside your controls.

Ronin builds models from the tags your historian already logs. Koios runs them on your site, inside the limits your PLC enforces. Curtis answers questions about your models and your organization.

Talk to an engineerRead the docs
An open control cabinet with a PLC rack and terminal blocks, in a plant full of pipework
Ronintrains on your historyKoiosruns it beside your controlsCurtisanswers your questions
// three products

Ronin builds. Koios runs. Curtis explains.

Each has its own page with the detail. This is how they divide the work, and what you would use each one for.

A Koios trend of a measured wet bulb temperature against a model's 15 minute forecastA Koios trend of a measured wet bulb temperature against a model's 15 minute forecast
On your site

Koios

The edge platform next to your controls, in the cabinet, on a server or air-gapped.

  • Reads and writes over OPC UA, Modbus TCP and EtherNet/IP
  • Runs your models and your own Python live
  • A tamper-evident record of every write
See Koios →
Ronin replaying days a forecast never saw, its forecast an hour ahead next to what really cameRonin replaying days a forecast never saw, its forecast an hour ahead next to what really came
Ai-OPs Cloud

Ronin

Model training for engineers, on your own process history.

  • Forecasts, soft sensors, anomaly detection
  • Graded on history it never trained on
  • Exported as an ONNX file Koios runs
See Ronin →
Today

The lab value lags the line by about six hours, so the model is graded against that.

Which tags would help this soft sensor most?

3 Ronin lookups

Feed rate and inlet temperature. FI-2101 and TI-2104 both lead the lab value by about a shift. Worth trying on the next run.

Ai-OPs Cloud

Curtis

Ask Curtis. An assistant that knows your Ronin projects and your account in the portal.

  • Plant insight from your own data
  • A partner for brainstorming the next model
  • Plain answers, with sources
See Curtis →
// where it sits in your control stack

From a forecast on a screen to a hand on the valve.

Pick the level that fits the loop. The same products cover every one of them, and your existing controls stay in charge.

Advise
What it isForecasting and anomaly detection
Ronin trainsA forecaster or anomaly detector
Koios writesA trend and an alarm
You keepEvery decision
Infer
What it isSoft sensors
Ronin trainsA soft sensor
Koios writesA live estimate, written to a tag
You keepThe lab sample as the check
Supervise
What it isSupervisory control
Ronin trainsA forecaster for your own MPC or optimizer
Koios writesSetpoints, inside limits you set
You keepThe loop and its PID
Act
What it isBase-level control
Ronin trainsA deep reinforcement learning controller, built with our team
Koios writesThe output to the equipment
You keepA native controller underneath, ready to take back
Act is where our deep reinforcement learning work lives, and where experience matters most.How we do DRL →
// one canvas

PID, MPC and AI, wired side by side.

Koios puts the controller you run today, an MPC and a model on the same canvas, with a select between them. You decide which one has the loop.

  • Write the PID and MPC you trust as Python components
  • Your native controller keeps running underneath, ready to take the loop back
  • Every switch and every write lands in the audit log
componentPIDyour tuned loop
componentLevelMPCyour Python
modellevel-rl.onnxtrained on your history
core librarySelectmode: PID
output tagLIC2.MV42.6 %
// what it talks to

No rip and replace.

If your control system can expose a tag, Koios can read it, and write back to it where you allow. Nothing in between, and nothing to rewire.

Protocols
OPC UAModbus TCPEtherNet/IPSLC 500 and MicroLogixRESTSOAPMicrosoft SQL
Runs on
Edge hardware in the cabinetA server on premisesA virtual machineA cloud VMAir-gapped
Models
ONNXTensorFlow LiteYour own Python
Where Koios runs on your network →Writing your own components →
// how you buy it

A license for each plant. A seat for each person.

Koios is licensed per site and keeps running whether or not anything upstream does. Ai-OPs Cloud is a seat per person, and each seat gets Ronin and Curtis.

Per site

Koios

Install it where the loop is and activate it with a small file, carried on a USB drive if the network is air-gapped.

  • Protocols, models and canvases on one license
  • Choose how many machines and how many years in the customer portal
  • Support from the people who build it, with every license
Talk to an engineer
Per seat

Ai‑OPs Cloud

Seats are for the people who build and question the models. Your administrator assigns seats in the portal.

  • Ronin and Curtis with every seat
  • Usage, invoices and seats in one place
  • Add or move a seat whenever the team changes
Open the portal
// start small

Start with the loop that keeps drifting.

Try Koios on one loop for 30 days. Reach out with any questions, or to start your trial.

Talk to an engineerSee the industries
  1. 01
    Ask for a trial

    Tell us about the site and the loop you have in mind.

  2. 02
    Install Koios

    On a spare machine or a VM, connected to one controller.

  3. 03
    You decide

    See what it does with your own tags over 30 days.