Engine · edge · app

The iGromi engine: the industrial AI platform that understands your plant

An engine that understands the plant, an edge that lives in it even when the internet drops, and an app for whoever isn’t at a computer. Each layer solves a different problem; together is where it makes sense.
Request early accessTalk to an engineer
The edge · is there

Reads the plant and keeps going offline

  • Sensors, Modbus, PLCs, MQTT and cameras
  • Offline it stores; back online, it sends what it kept
  • Web OS with the plant app suite
  • Node-RED for local automations

Industrial computer in the plant

The engine · understands

Ontology, rules, memory and agents

  • Ontology: what each asset, zone and process is
  • Rules and indicators traced back to the data
  • Knowledge with authority, validity and access level
  • Agents that propose; a person approves

Server or VM: shared, your own or wherever you say

The app · goes with you

For whoever is out on the floor

  • Alerts in sight when you open it
  • Chat: the team and the Orchestrator in one thread
  • Your plant’s apps for your role
  • Notifications without third-party services

iPhone and Android · early access

What the engine knows · ontology

An industrial ontology: what each thing is and how it chains together.

The engine doesn’t keep loose numbers. It knows what each asset, zone and process is, and how they relate.

  • Each concept with its relations and the documents it came from.
  • Built from approved memory, not from whatever the model comes up with.
  • Tap a concept and ask about it.
Concept map in the iGromi engine: the industrial ontology of a construction site, with payment status, F30-1 certificate and subcontractor linked to their source documents
Inside the engine · Memory › Map. Example data.

What the engine knows · knowledge

Knowledge with authority, validity and access level.

An authorised protocol outweighs an observation. The engine answers only with what is approved and current, and cites where it came from.

  • Each document with version, status and validity.
  • Proposals wait for review before they count.
  • Access level per document: everyone sees what’s theirs.
iGromi engine memory: company folders with approved documents, their version and one proposal awaiting approval
Inside the engine · Memory › Documentation. Example data.

How a data point travels

From the sensor to a decision, in four steps.

  1. 01

    It’s read

    The sensor delivers a number. The edge stores it and tags its quality: if it’s outside the physical range of that variable, it’s flagged and doesn’t feed indicators.

    Hands connecting a sensor to a line motor; the edge reads the value and stores it with its qualityM204_I_A · 7,8 AQuality: within physical rangeStored on the edge · 10:42
    Step 1 · the edge in the plant
  2. 02

    It’s understood

    The engine resolves which asset that tag belongs to and what it means in the process chain. It stops being a loose number.

    is the current ofdrivesfeedsM204_I_A · 7,8 APLC tag · read by the edgeMotor M-204asset · rated 7.5 ALine 2 · packagingprocessShift dispatchwhat falls behind if it stopsrule: overload
  3. 03

    It’s evaluated

    Rules open an alert when they should. It’s born deterministic: you can rebuild why it fired without asking a model.

    Alerts channel in the app: critical alert “Line 2 stopped” from an M-204 motor overload read from the PLC
    Example data
  4. 04

    It’s decided

    Someone acts, and the decision is kept as a fact with its reason. The engine checks it before proposing the same thing again.

    A person takes the shed 5 temperature alert, resolves it and it is saved as a criterion
    Example data

The rule that runs through everything

The engine proposes and measures. A person approves.

It records who approved, when and on what evidence. The edge reads the PLC read-only: no machine moves on its own.

A system that decides on its own is a system nobody is accountable for.
PLANT · LINE 2
To approve
Carla · management
For me · 1History
SENSOR ALERT
Vibration sensor · motor 4
4.1 mm/s · above normal since 10:02
iGromi proposes
Stop line 2 and replace the bearing, like on Jul 14.
Source: motor 4 sensor · WO Jul 14
Proposed by @igromi · awaiting your decision
ApproveReject
Approved by Carla · management
10:15 · work order assigned to Juan
Saved to motor 4's history
HomeTasksChatMore

AI harness

The industrial AI harness.

A harness is the layer that wraps an AI model so it can work as an agent. iGromi’s is the engine: it gives any model your plant’s ontology, your rules, your approved memory, permissioned tools and a person who approves.

  • The model is swapped; the harness keeps what your plant knows.
  • It is what makes AI useful and safe in a plant.
  • We mean the AI software harness, not a safety harness or a wiring harness.
MODELlocalswappable
  • Ontology
  • Rules
  • Authoritative memory
  • Tools
  • Permissions
  • Human approval

The edge suite

Sixteen apps on the edge Web OS. Install the ones you use.

They install from the Store as each plant needs. All talk to the engine the same way: they produce facts and ask for approval when it’s time to act. A new app doesn’t require touching the engine.

  • Energyfrom the Store
  • Waterfrom the Store
  • Climatefrom the Store
  • Maintenancefrom the Store
  • Productionfrom the Store
  • MESfrom the Store
  • Formscoming soon
  • Documentsfrom the Store
  • Datafrom the Store
  • Visionfrom the Store
  • GPSfrom the Store
  • Platesfrom the Store
  • Silosfrom the Store
  • Simulationfrom the Store
  • Multimediafrom the Store
  • Messagingfrom the Store
App home at Grupo Proa’s egg plant: alerts on top and the edge apps Analytics, Pullets, Silos and Eggs
App screen with data from a Grupo Proa farm

Helpers per role

AI agents that propose, each with its own board.

Each helper works alongside a role: it watches its indicators, flags what’s due and proposes what’s next. It never releases or executes on its own.

  • The role’s indicators, current and with their source.
  • Proposed tasks a person accepts or dismisses.
  • Adjusts to your company’s rules.
AI agent boards in the iGromi engine: payment status, site progress, quality and subcontracts
Inside the engine · Boards. Example data.

Learns · under construction

It learns how your plant behaves, hour by hour.

The engine builds a baseline per measuring point and per hour, and proposes thresholds backed by history before applying them. A person approves them.

The mechanism is built; it’s gathering the history it needs.

currentproposedProposed threshold · shed 5backed by history · awaiting approval00:0006:0012:0018:0024:00hourly baseline

Where the engine runs

From free to your own, without starting over.

You start on the shared engine. When you want your own, everything moves in full and is verified before the move is signed off.

Security and where your data lives at data sovereignty.

  1. 01 · Free

    iGromi’s shared engine, with an isolated space per company. Nothing to install.

  2. 02 · Premium

    Your engine on a machine of your own.

  3. 03 · Enterprise

    Your engine wherever you say: your cloud or your plant.

Moves in fullDataKnowledgeRulesHistoryWaiting for the move

FAQ

What people ask about the engine

What is the iGromi engine?
It is iGromi’s industrial AI platform: the server where your plant’s ontology, rules and indicators, authoritative knowledge and AI agents live. The edge reads the plant and the app goes with whoever is on the floor; the engine is what understands and proposes. A person approves.
What is an industrial AI harness?
An AI harness is the software layer that wraps a language model so it can work as an agent: it gives it context, tools, memory and permissions, and controls each step. An industrial AI harness adds what a plant demands: the operation’s ontology, deterministic rules, memory with authority and validity, permissions by access level and recorded human approval. At iGromi, the harness is the engine; the model can be swapped.
What do AI agents do in the engine?
iGromi’s AI agents work alongside a role (maintenance, quality, site administration): they query the Second Brain, explain a deviation with its source and propose a task or an action. They don’t execute it on their own: the Orchestrator holds it until a person approves, recording who, when and on what evidence.
What is an industrial ontology?
It is the map of meaning of your operation: what each asset, zone and process is, and how they chain together. Thanks to it the engine knows a PLC tag is motor M-204’s current, that the motor drives line 2 and what falls behind if it stops.
Can the engine change a PLC parameter?
No. The edge reads the PLC read-only. The engine queries, evaluates and proposes; a person reviews and decides.
What happens if the plant loses internet?
The edge keeps reading and storing in the plant. When the link returns, it sends the engine everything it stored.
Which AI model does the engine use?
The Swappable Cognitive Core: a local model or an external provider via API. Ontology, memory, rules and permissions belong to the engine, not the model, so changing models doesn’t mean starting over.
Does my data train other companies’ models?
No. One company’s data is never used to train another’s models. On the free plan each company has its own isolated space on the shared engine.

Keep reading

Related guides

Industrial artificial intelligenceWhat it is and where it helps in a plant.Secure industrial AICriteria to evaluate any vendor.AI agents in the factoryWhat they do and where a person still decides.Industrial edge computingThe edge that reads the plant and keeps going offline.Industrial IoT architectureHow the plant’s Sources connect.DevelopersAPIs, MCP and the Lobe SDK.

Every shift without a record is data you lose. Start free.

  1. 1Request early access, free
  2. 2Pick your trade: maintenance, production or safety
  3. 3Your team starts with the chat and its trade helper
Get early accessBook an assessment

No credit card. Your data stays yours.

PLANT · SHIFT B
Tasks
Juan · maintenance
OVERDUE
Silo 1 level
No signal since 05:40
Sensor · supply
TODAY
Motor 4 vibration
4.1 mm/s, above normal
Detected by the sensor
PROPOSED
Check the fans
Shed at 29 °C at 11:20
AcceptDismiss
HomeTasksChatMore