-
Nice to have.
-
It is mandatory.
-
How many bids can you afford to lose to a faster shop?
So, OK… AI.
Start with one decision.
Ask someone on your shop floor or in your front office what they need to decide today. Which part would they trust their AI to help with?
Fine. Easy enough.
What comes next?
Do you grab a subscription to ChatGPT, Copilot, Claude, etc., and tell your folks to go nuts?
Hang on.
What should their AI actually be allowed to touch and how do we make 100% sure it can’t get outside its box?
Your AI can only use the tools your plant approves.
Pocket-E checks who’s asking, enforces your rules, and logs every call.
Pocket-E has spent 17 years implementing ERPs, keeping mission-critical MES running 24/7, and building smart-factory systems.
We use Pocket-E as the middle layer between the AI and your systems.
Examples of AI tools
- ChatGPT
- Copilot
- Claude
The AI can call only the workflows your plant approves.
- ERP
- MES
- CRM
Your plant systems
Pocket-E checks the user’s permissions, enforces your business rules, and logs every call.
You never connect an AI directly to your ERP, your MES, or whatever else runs your plant.* You never expose those systems in a way that lets the AI even try to connect.
You use a sane operating layer that gives the AI the tools it needs to help your people. Your plant decides what those tools can do.
Bottom line, your people need to be able to use the AI tools.
You MUST have a controlled layer between the power of LLMs and your critical systems.
So here’s a real-world example we’re writing for one of our customers right now.
(Maybe one of your competitors?)
This customer already subscribes to Grok and Claude. Put yourself in the buyer’s chair. You’ve got a purchase order ready to release. Open either one and ask…
“Before I release this PO, is this supplier approved to make the revision I’m ordering? Show me the records you checked.”
- Claude
- Grok
Read-only revision check
Pocket-E checks the buyer’s permissions, reads the approved records, and logs the call.
-
Supplier PPAP
XLSX workbook
- Approved revision
- C
-
CAD drawings
Shared directory
- Drawing revision
- C
-
QAD
Plant ERP
- Part master
- C
- Purchase order
- B
Hold up. The PO says B.
The supplier approval, the drawing, and the QAD part master all say C. The buyer gets the four records behind the mismatch. The check flags the order.
Could B still be right?
It could. Maybe Quality signed off on revision B for this lot. That belongs in an exceptions log so the next shift can see who approved it and which lot it covers.
Real plants are messy. Sometimes the approval is buried in an email. Sometimes it happened over the phone and never made it into the log. If the check can’t find it, the buyer calls Quality to confirm that B is cleared for this lot before the PO goes out.
“Cool. So you made a chatbot I can ask about my parts.”
More than a dumb chatbot.
Once Pocket-E knows how to check that revision, you can build whatever your plant needs around that logic. A PO changes? Run the check. An operator scans a part? Run it there, too. Here are three ways to put it to work, all inside Pocket-E.
Flag the order.
The dashboard flags revision mismatches so Purchasing can see which orders need a second look before release.
| Order | PO rev | Approved |
|---|---|---|
| PO 4817 | B | C |
Awaiting buyer review
Notify the right person.
Quality gets a notification when revisions don’t match and can check whether an approved exception covers the lot.
To Supplier Quality
Stop it at the scan.
The operator scans the part at the line, and if that revision isn’t approved for the job, Pocket-E won’t let them use it.
Revision not approved
- Scanned
- Rev B
- Job requires
- Rev C
Let’s get your first skill running.
-
We drop Pocket-E in.
We install Pocket-E alongside your systems and work with your security team to give the skill access to just what it needs. It runs on the same secure, stable framework Pocket-E has refined over 17 years to keep major factories running 24/7.
-
Your people set the rules.
We sit down with your business folks. “How do you make that call? Tell us about the last time the usual rule didn’t apply.” We draw on decades of ERP and process consulting to help your people spell out what the skill should check and when it should ask them.
-
Try it. Then turn it on.
We build the skill. Your team tests it against decisions they’ve already made. Does the answer check out? Does it flag the exceptions they’d want to see? Give us the nod and we turn it on in the MCP-compatible AI your people already use.
What do you want your AI to help with first?
We’re happy to trade a few emails to see if there’s a fit. If you’d rather jump on a call and talk it through, that works too. Your pick.