Home / Platform
How an answer is made, and why you can check it.
Plant Cortex is a retrieval system, not a chatbot with opinions. Every grounded answer is built from passages of your own documents and carries their page numbers.
Ask, then check.
This is a real question on a real manual. The page is blurred here because it belongs to a real plant. Tap a source to see where the answer came from.
Before starting up the kiln plant, check and activate all necessary interlocks required for safety reasons. Confirm the area is clear of personnel, and that protective guards are in place.
Tap a source. It opens the actual page, with the passage that was used.
This is the screen your shift lead opens.
A real screenshot of the app, taken on a test install with made-up sample records. Nothing here is from a customer plant. What is waiting for you comes first, plant status sits under it, and the question box is always one tap away.
Four steps, none of them trusting the model blindly.
Understand
The question is cleaned up, follow-ups are resolved against the chat, and the reply language is detected. Plain words are fine.
Retrieve
Meaning-based search over your manuals, scoped to the machine you meant. When look-alike manuals exist, it asks which one with tappable choices instead of guessing.
Answer from those passages only
The model is handed only the retrieved text and told to cite it. Every citation is resolved back to a real page, never trusted as written.
Check
Numbers and part codes in the answer are compared with the source text. Anything not in the source is flagged amber. If nothing supports an answer, it says "not in the manuals".
Documents are only the start.
Each machine becomes a page that gathers everything known about it.
An equipment tree, not a folder of PDFs.
Manuals and drawings are tagged to a department, machine and component. Ask about the machine and the answer prefers its own documents. Superseded and duplicate files are flagged and left out of answers, never deleted.
- Manuals
- Drawings
- Open permits
- Stoppages and KPIs
- Near-misses
- Spare parts
- Work orders
- PM plans
Drawings found by what they show
Ask "show me the drawing of the gate" and the real page opens, including general arrangement sheets that sit inside manuals. Matches are labelled to confirm. Nothing is redrawn or invented.
Field notes beside the manual
When a stoppage or permit is closed with a real resolution, it can surface next to manual citations, clearly labelled as a field note and not OEM text.
Built for the floor, not the desk.
Voice
Speak the question or the near-miss description, in the language the reply is set to.
QR at the machine
Stick a code on the equipment. Scanning opens that machine's questions and records.
Android app, works offline
Manuals saved on the phone for search. Near-misses, stoppages and checklist results are written with no signal and sync once, with the time they really happened. A biometric lock protects the device.
It sits beside your ERP and your sign-in.
No rip and replace. Plant Cortex is the analysis and records layer next to systems you already run.
ERP data exchange
Import equipment, parts, failure codes and planned orders from CSV with a dry run that shows every change first. Export notifications, confirmations, goods issues and approved requisitions. Direct connectors to a specific ERP are planned, not built.
Single sign-on
OpenID Connect sign-in with domain limits and an optional required mode. Available, off by default, and not yet proven against a customer's identity provider.
Audit and signatures
Field-level history on permits, orders, plans and parts, sealed in batches so edits are detectable. Optional e-signature with password re-entry. Not certified to any external scheme.
Live data, proven on a test server
An OPC-UA client has been proven against a public demo server. Connecting it to real plant equipment needs an on-site agent and is not done.
See it on your own manuals.
Send a few of your real documents for one unit. We index them and you ask the questions your crew actually asks.