Skip to content
Koidra
Book a demo

Process Optimization

Physics-aware AI for process plants

More good output, steadier quality, less energy.

What Process Optimization does

Proof

2×
Winner, Autonomous Greenhouse Challenge 2018 and 2022, a live crop each time
17%
Higher net profit than the expert growers 2018 competition, with 6% higher yield
19.5%
Yield increase 29.7 against 24.9 kg/m² in the growers' own zone
17%
Energy savings lower electricity cost per unit, same lights in both zones

The autonomous control in the KoPilot plan has been tested in international competition and in commercial greenhouses. In 2018, the technology behind it was the only AI entry to beat the expert growers on both yield and net profit. It does more than recommend a climate strategy. It makes the control decisions and applies them, day and night.

Three plans

Essential Pro KoPilot

Process Optimization · Essential

Universal data integration: all your operational data in one place, live and historical

  • One data layer for your PLC, SCADA and sensor data, with metrics you define
  • Your operational data, modeled as a knowledge graph
  • Condition-based alerts and alarms, on every phone
  • MCP, so Claude, ChatGPT and your own AI tools can work with your operational data
The Essential plan at a glancePLC, SCADA and sensor readings and the entries operators make on their phones come together in one data layer. From there the plant builds its own metrics, gets alerts and alarms on every phone, and its data in Claude, ChatGPT or its own AI tools through MCP.PLC, SCADA and sensorsOne data layerOperator entriesYour own metricsAlerts and alarmsYour AI tools, through MCP

Data integration

One data layer for everything your plant records

All the data from your PLCs, SCADA, meters and sensors, live and historical, read-only from the systems you already run. Organized around the same equipment and timeline, and turned into the metrics you run the plant by.

Process and energy data
Data from your existing PLCs, sensors, meters and control systems, collected at the frequency they already provide it.
Production batches
Runs, campaigns, crop cycles or other production periods that connect operational data to what was being produced.
Your own metrics
A formula turns raw readings into the number you manage by, such as energy per tonne. It updates live, and calculated metrics are free.
Creating a calculation metric named Energy per ton, with the formula daily(Pelletizing energy meter) divided by daily(Wet Hammermill 1 output).
Energy per ton, from the pelletizing line's energy meter and the hammermill's output. The formula is the whole setup.

Semantic operational data

Your plant, modeled as a knowledge graph

Koidra models how your plant is built, and links every number and note to the part of the plant it describes. Sensor data, form entries, photos, alarms and work orders all point to the same areas, machines and production batches, so a question about one machine finds everything that belongs to it.

Metrics, one model for every time series
A sensor tag, a setpoint, a number an operator types, a formula you write and each move KoPilot makes are all metrics. One chart, one alarm rule and one analysis work on any of them.
The plant as it is built
Sites, areas and lines, the machines in them and their parts, the products you make and each production batch. A metric belongs to an area, a batch, or that batch in that area.
What people record, in context
Form entries, photographs, alarm responses and work orders, each tied to the machine and the time they are about.
Your plant as a knowledge graphSensor tags and setpoints, the entries operators make, and the formulas a team writes all become metrics, one time-series model. Each metric belongs to an area or line of the plant and to a production batch. Areas hold the assets and their parts. Alarm rules watch metrics and open work orders on assets, and photographs are attached to the assets they show.belongs tobelongs toholdswatched byopenonofSensor tags and setpointsMetricsone time-series modelOperator entriesYour formulasAreas and linesProduction batchesAssets and partsAlarm rulesWork ordersPhotos
Every metric, record and work order points to the same areas, machines and batches, so an analysis can follow a problem from a reading to a repair.

Alarms and incidents

Alarm rules written like spreadsheet formulas

Most control systems raise an alarm when one reading crosses one limit. That is right for a fast reaction, and it misses problems that build up over hours or involve several readings at once. In Koidra, your team writes a rule as an Excel-style formula on live metrics, such as a spark detector tripping three times in ten minutes, or a burner cooling while its feed screw still runs. A precise rule fires less often by mistake.

Any metric, any time window
A rule can combine sensors, formulas and the numbers operators type, and look back over minutes or days.
Every incident gets an answer
Each alarm carries its severity and stays open until someone records its cause and how it was resolved. A rule that keeps firing for nothing shows up, and gets fixed instead of ignored.
From alarm to work order
When the fix takes longer than a shift, the alarm becomes a work order.
Two alarm rules written as formulas: a dryer's spark detector tripping three times in ten minutes, and a burner cooling below 700 degrees while its feed screw still runs, each with its frequency, severity, and the explanation the responder reads. Alarm Condition Frequency Severity Explanation and recommendation Grecon drum dryer detects continuous sparks CumSum(DIFF({Dryer1 Dr201 Grecon}) = 1, 10) >= 3 1x consecutively Warning Grecon sensor indicates spark 3 times in 10 minutes. Burner's temperature drops below 700 Celsius degree. AND({Dryer1 Burner T Lower} < 700, {Dryer1 Burner T Upper} < 700, moving_average({Dryer1 Sc502 Speed}, 30) > 15) 15x per hour Info Burner's temperature drops when fuel is still being fed. Check SC501/SC502/ SC505 feed screw condition and transmission for stuckage, then check material condition if it causes the stuckage, remove material stuckage point.

AI tools

MCP is available for interacting with your own AI tools

The Essential plan supports MCP, the Model Context Protocol. With it, general-purpose AI tools such as Claude and ChatGPT can work with your operational data.

Your operational data in the AI tools you already useClaude, ChatGPT and a plant's own AI tools connect through MCP, the Model Context Protocol, to the plant's data in Koidra.ClaudeMCPChatGPTYour own AI toolsYour operational data in Koidra
Claude, ChatGPT or your own AI tools ask through MCP, and Koidra answers from your operational data.

Process Optimization · Pro

The AI Consultant: an AI process engineer, available 24/7

Everything in Essential, plus:

  • Root cause on request
  • Recommendations accepted in one click
  • Talk with the agent
  • Answers grounded in physics and data
The Pro plan at a glanceThe AI Consultant works on the plant's data from the Essential plan. A question from the team, or a change in the data, starts it. It ranks the root causes with their evidence and recommends what to do, and a person accepts the recommendation in one click.Your operational dataEssentialAI ConsultantYour question,or a change in the dataRoot cause, rankedRecommendation,accepted in one click

AI Consultant

An AI process engineer, available 24/7

Root cause on request
Ask why a number moved, and get the causes ranked with their evidence.
Recommendations accepted in one click
A work order, an alarm change or a new metric.
Talk with the agent
Ask about your plant and get answers from its own data, with a Koidra expert behind it.
The weekly brief for a biomass energy plant: the week summed up with its key numbers in bold, each citing a chart, then To act on, eight findings, 3 high, 3 normal and 2 low, each folded to its priority, kind, title and impact.
A biomass energy plant's weekly brief. Every number in the summary cites the chart it comes from.

Physics-informed AI

Answers grounded in physics and data

A language model leads the conversation. Underneath it, Koidra’s physics-informed machine-learning models do the analysis, so every answer rests on the physics of your process and on your own data.

Answers grounded in physics and dataThe plant's own data and the physics of its process feed Koidra's physics-informed machine-learning models. The AI agent works on top of those models, and what reaches the team is analysis, recommendations and a conversation about the plant.Operational dataKoidra's physics-informed modelsProcess physicsThe AI agentAnalysisRecommendationsConversations
A language model alone can describe your numbers. Koidra's AI computes against the physics behind them, so a recommendation can be a better way to run, such as a setpoint to move or a changeover to reschedule, as well as a repair.

Process Optimization · KoPilot

Autonomous process control: proven recommendations, applied every minute

What your plant gets

  • Every shift runs at your best shift's settings
  • Steadier quality as raw material changes
  • Energy and quality balanced every minute

Everything in Pro, plus:

  • It applies proven recommendations through the control system you already run
  • Physics-informed decisions that stay within what your equipment can do
  • Safety guardrails: limits on settings and conditions, alarms that can switch it off, a switch for every control module
The KoPilot plan at a glanceRecommendations from the Pro plan that have proven themselves become autonomous control. Every minute, each setting passes through the safety guardrails before it reaches the plant's control system, which runs the process. Live plant data flows back, and the loop starts again.every minutelive plant dataProven recommendationsProAutonomous controlSafety guardrailsYour control systemYour process

Autonomous control

What KoPilot does

You know what the right setpoints are, and they still depend on who is on shift. KoPilot reads real-time plant data, determines the right setpoints, and sends them back to the control system you already run. It responds continuously as conditions change, so the best night your plant has ever run becomes every night.

What it replaces
Setpoints tuned by hand, differently on every shift, and comfort margins nobody has revisited.
How you measure it
Energy cost per unit of output, and the output variance between shifts.
How Koidra's AI runs a plantProcess sensors and energy meters feed the PLC and SCADA, which send live data to Koidra's AI. The process engineer sets the targets and the limits. The AI decides the setpoints round the clock and writes them back to the SCADA, which moves the dryer, the mill and the boiler.readingslive datatargets and limitssetpointsProcess sensors and metersProcess engineerPLC and SCADAKoidra's AI, 24/7Dryer, mill, boiler
Your engineers set the targets and the limits. KoPilot decides the setpoints round the clock, and your control system moves the equipment.

Safety

Safety guardrails, built in

Your team sets the limits. KoPilot works inside them.

Limits on settings and conditions
Engineers set a low and high limit for every setting KoPilot moves, and a safe range for each condition, such as a temperature. When the actual temperature leaves its range, KoPilot moves the settings to bring it back.
A switch for every control module
Switch a module off and control goes back to your existing settings at once. Your interlocks stay in charge, and an operator can take back control at any moment.
The control's general configuration at a greenhouse: temperature limits 14 to 28 °C, T24 limits 17 to 20 °C, a maximum pre-night cooling speed, and humidity deficit limits by day and night. System is ready.
The limits at a greenhouse: the range temperature and humidity must stay in, and how fast the temperature may fall before night.

Safety

An alarm can switch KoPilot off

The condition-based alarms from the Essential plan keep watching the process while KoPilot runs it. When one fires, it can switch KoPilot off on its own. Control goes back to your existing settings, and the alert reaches the person on call.

An alarm can switch KoPilot offWhen a condition-based alarm fires, it can switch KoPilot off on its own. Control goes back to the plant's existing settings, and the alert reaches the person on call.A condition alarm firesKoPilot switches offYour existing settings take overThe person on call is alerted
The alarm stops the AI by itself, and your existing settings take over the moment it fires.

Control model

Physics-informed control

The control combines process data with models of how the physical system behaves, so a control decision stays within what the equipment can actually do rather than only within patterns seen in historical data.

How quickly equipment can heat or cool
A dryer heats only as fast as its mass allows, and the model plans within that.
How humidity responds to ventilation
Opening a vent changes humidity and temperature together, and the model knows by how much.
How one control action affects conditions later
The model weighs each move over the next four hours, beyond the current minute.
Why the model is built from physicsA decision is checked against how the equipment behaves, how fast it heats or cools, how humidity answers ventilation, what one action does later, so it stays inside what the plant can do, even in conditions the history never showed.Process datawhat is happening nowThe control modelPhysics of the planthow fast it heats or cools,how humidity answers ventilation,what an action does laterA setpoint the equipment can reacheven in conditions never seen before
Process data says what is happening now. The physics says what the equipment can do next. Together they give a setpoint the plant can reach, in conditions the history has rarely shown.

Getting started

How a KoPilot project starts

Every project starts with an assessment of your process.

We look at the quality and coverage of your instrumentation, the reliability of the control system, how well the process can be modeled, and the operating constraints it must respect. A supervised trial comes before autonomous operation.

Most customers start with the Essential plan and add Pro. Those plans build the data, the operating history and the process understanding that make autonomous control reliable.

Tell us about your process, and we will show where autonomous control fits, what comes first, and how a supervised deployment would run.

Earning autonomy one decision at a timeKoidra first sees the plant's data and explains drift and its likely causes, both read-only. It then recommends changes, and a person approves each one. Only once those recommendations have proven themselves does it write setpoints, through safety guardrails, with the plant's safety systems still in charge.proven in shadow firstSeetrusted dataUnderstanddrift and causesRecommendyou approve each changeControlwith safety guardrails
Koidra first shows the plant's data and explains drift and its likely causes. It then recommends changes, and a person approves each one. Once those recommendations have proven themselves, it writes setpoints, inside your limits.

Two complementary AI-powered solutions

  1. Process Optimization All your operational data in one place, an AI process engineer available 24/7, and autonomous process control with safety guardrails built in. Essential Universal data integration: all your operational data in one place, live and historical Pro The AI Consultant: an AI process engineer, available 24/7 KoPilot Autonomous process control: proven recommendations, applied every minute
  2. Intelligent Asset & Maintenance Management Work orders, preventive maintenance, inspection routes and spare parts, on every technician's phone. Essential Modern maintenance management (a CMMS), set up by your own team Pro Predictive maintenance, from evidence to work order

See Koidra on your own plant data

Bring a week of readings from one line. We will show you what the platform sees, what it would have flagged, and what it would have scheduled.

Book a demo