Plant OS for food manufacturers

The brain of your food plant. From PLC to audit, one AI.

A Plant OS is the operating system for an industrial plant. The same way a computer OS coordinates hardware, files, network and applications without making you think about each piece separately, a Plant OS coordinates sensors, PLCs, lines, equipment, people, quality records, maintenance and compliance under a single layer.

For an SME food manufacturer — an industrial bakery, a small dairy, a mid-sized meat plant — a Plant OS is not an Industry 4.0 luxury: it is the only realistic way to run HACCP, sustain the cold chain and keep maintenance traceability without hiring three consultants and buying five licences.

Why a Plant OS and not more separate tools?

The typical SME case — a 30-person bakery or a cheese plant with 8 vats — looks like this:

The aggregate cost for this SME runs between €15,000 and €25,000 per year in licences, paper trails and back-office hours, and there is still no real-time visibility nor a single signed record for audit. Every time a demanding customer or a BRCGS auditor shows up, there is a week of scramble.

A Plant OS solves this by changing the question. Instead of "which tool do I use for HACCP, which other one for maintenance, which other one for OEE?", the question becomes:

"How is my plant right now, and what do I need to do in the next 60 minutes?"

Only a system with a single view of assets, processes, people and records can answer that.

The 7 layers and why they share an asset graph

Rela AI has seven layers that share a single asset and process graph. A bakery oven is not a maintenance asset "and also" a HACCP CCP "and also" an OEE node: it is the same node in the graph, seen from different dimensions.

flowchart TD
  Oven["Rotary oven HR-3"]
  Oven --> HACCP["HACCP CCP-COC1<br/>baking 220 C, 14 min"]
  Oven --> Cold["Cold chain<br/>exit toward tunnel"]
  Oven --> OEE["OEE line 2<br/>availability and performance"]
  Oven --> Maint["Maintenance<br/>burner, fan, motor"]
  Oven --> LOTO["LOTO<br/>electrical and gas isolation"]
  Oven --> Agent["WhatsApp agent<br/>night shift"]

  classDef food fill:#fef3c7,stroke:#92400e
  classDef ops fill:#dbeafe,stroke:#1e40af
  class HACCP,Cold food
  class OEE,Maint,LOTO,Agent ops

Sharing the graph means that when the WhatsApp agent receives the message "oven HR-3 stopped", the system knows simultaneously:

• Which HACCP CCP is affected (baking of the in-progress batch).
• The cold-chain impact on products that already exited.
• The OEE loss accumulating on line 2.
• The preventive work order scheduled for that oven.
• The LOTO that must be executed before the technician opens the combustion chamber.

Without the shared asset graph each of those threads lives in a different tool and they get reconciled by hand (or not at all).

Plant OS vs MES vs CMMS vs ERP

This is the most frequent confusion when someone hears "Plant OS" for the first time. Short answer: a Plant OS does not replace your ERP and does not overlap with a large-factory MES, but it does eat most of what an SME food maker would try to do with a CMMS, a HACCP sheet and a SCADA dashboard.

Rela's Plant OS connects to your ERP (purchasing, lots, sales), to your CMMS if you already have one and to your SCADA if it exists — but it does not require any of them. For an SME food maker without MES and without dedicated SCADA, Rela acts as the glue between the PLC, the operator, the plant manager and the audit.

Where a Plant OS starts — the HACCP + cold chain wedge

There is a deliberate decision in the order in which a Plant OS is switched on: we do not start with predictive maintenance, we start with HACCP and cold chain. The reasons are four:

1. Willingness to pay (WTP). A recall driven by a HACCP failure costs between €100,000 and several million euros, depending on product and country. A maintenance failure costs between €5,000 and €50,000. The same plant pays much faster to mitigate HACCP than to mitigate downtime.
2. Regulatory fines. FDA, EFSA and national health authorities can shut a plant down for HACCP non-compliance. There is no regulatory equivalent for missed maintenance.
3. Software-only is not defensible. SafetyChain, Tulip Food and other competitors do "digitised paper" HACCP: the operator still writes temperatures by hand. Rela reads the temperature from the PLC in real time, which is defensible in front of an auditor in a way the digital sheet is not.
4. Time to value. HACCP and cold chain can be switched on in a week in many plants by reading existing PLCs. Predictive maintenance normally requires three to five months of history — and Rela's zero-history bootstrap helps with that, but conceptually it is still slower.

Once the plant lives with stable automated HACCP and cold chain, the natural upsell is real-time OEE (same PLC, different metrics) and from there maintenance. The four layers reinforce one another.

Compound vertical SaaS

Plant OS fits a pattern identified by Parker Conrad (Rippling) and documented by SaaStr as compound startups: companies that, instead of selling a SaaS module to the department that needs it, sell the full platform to a vertical and eat eight tools at once.

The strategic argument: Tulip is horizontal (any frontline), so it cannot do deep HACCP. Augury and Tractian are horizontal-maintenance, so they do not do HACCP at all. SafetyChain is vertical food but software-only. Rela is vertical food, connects to the PLC, and combines all the layers. That combination is what makes the category defensible.

Documented use cases

How to deploy Rela as your Plant OS

The deployment has a recommended order. Every step delivers value on its own, so if you stop after any of them you are already operating better than before; but the combination is what multiplies the return.

1. Switch on the WhatsApp agent. Configure the conversational agent, its tools and approval flows. The operator already starts asking "status of chamber 4" or "assign this task to Marco" without opening any dashboard. Doc: /docs/whatsapp-agents/create.
2. Connect the PLCs over VPN. Read existing PLCs over Modbus TCP, OPC UA or MQTT, with no need to install our own hardware. A single industrial VPN serves the whole plant. Doc: /docs/vpn and /docs/use-cases/multi-machine-single-tunnel.
3. Configure HACCP CCPs with AI. The agent reads your HACCP plan from a PDF or builds it in conversation; it maps each CCP to a PLC tag and signs every sample automatically. Doc: /docs/food-safety/haccp.
4. Activate cold chain. Cold rooms, blast tunnels, retail cabinets and transport. Alarms on WhatsApp with context and a suggested corrective action. Doc: /docs/food-safety/cold-chain.
5. Switch on real-time OEE. Availability, performance and quality computed on the same PLC tags. Doc: /docs/oee.
6. Upsell: predictive maintenance. Baseline learned from the PLC, AHI per asset, RUL, alerts with canonical severity info < warning < high < critical. Doc: /docs/condition-monitoring/overview and /docs/maintenance/predictive-config.
7. Digital LOTO and permits to work. Closing the loop: the technician who gets the work order via WhatsApp executes LOTO from the same thread and leaves signed evidence for audit. Doc: /docs/safety.

A typical plant goes through steps 1 to 4 in its first month, steps 5 to 7 in the next three months, and from there enters steady-state operation. The details of each step, with figures from the reference plant, live in the use cases.