How to use ChatGPT with DELMIAworks (IQMS)
Short answer
You can use ChatGPT with DELMIAworks in three ways: paste or upload exports for one-off analysis, build a custom GPT with actions that call an API you host, or put a gateway in front of the data. For regulated or customer-restricted work, a private open-weight model on your own hardware is often the better fit.
Plants running DELMIAworks, the system many still call IQMS, are asking the same question: can ChatGPT answer questions about our orders, molds and inventory? It can, but how you connect it decides whether the result is a useful tool or a data leak waiting for a security review.
The honest starting point is that ChatGPT does not talk to your Oracle database on its own. Something has to carry data to it, and that something should be designed, scoped and logged.
Below are the realistic paths, from a five-minute export to a governed integration, and where each one breaks down.
What teams use it for
Explaining a variance export
A controller exports an inventory or margin report to CSV and asks ChatGPT to group the top drivers and draft the commentary for the monthly close.
Drafting supplier and customer emails
A buyer pastes a PO line summary and asks for a polite but firm expedite request, without sending any pricing.
SQL help for report authors
An analyst asks ChatGPT to explain a view or draft a query pattern using column names only, then tests it against a non-production copy.
Work instruction and SOP rewriting
Engineers paste a process sheet for a molding job and ask for a clearer operator version, with no ERP data involved.
Custom GPT for order status
A custom GPT with one action calls a small internal endpoint that returns order status by number, so sales can ask in plain language.
Analyzing downtime reasons
Production exports downtime events by press and asks for patterns across shifts and molds.
DELMIAworks (IQMS) integration surface
- Exports: DELMIAworks reports and grids can usually be exported to Excel or CSV, which is the simplest way to hand data to a chat tool.
- Oracle views: an internal service can query curated read-only views and return JSON for a custom GPT action to call.
- Vendor web services: where your release exposes REST or SOAP style endpoints, a gateway can wrap them with your own auth and field filtering.
- Scheduled data drops: a nightly job can write a sanitized extract to a controlled location for use with ChatGPT file analysis.
- Version caution: endpoints and tables differ across releases, so confirm names with your administrator.
How it works
- 1
Classify the data
Decide which fields are public, internal, customer-confidential or export-controlled. Anything restricted by a customer contract or by regulation should not go to a public model service.
- 2
Choose the ChatGPT tier
Business and Enterprise plans offer admin controls and a commitment not to train on your data by default. Free and personal accounts do not offer the same controls, so ban them for ERP data.
- 3
Start with sanitized exports
Strip names, prices and part numbers where possible, or replace them with codes, before upload.
- 4
Build the action endpoint
For a custom GPT, host a small HTTPS service with an OpenAPI description that exposes a few read operations over curated views.
- 5
Authenticate and scope
Protect the service with OAuth or API keys tied to a role, and return only the fields the role may see.
- 6
Log and review
Record every call the GPT makes, who invoked it and what came back. Review weekly for the first month.
- 7
Decide on private
If restricted data keeps coming up, move the same tools to an on-prem model rather than loosening the controls.
Example flows
Month-end commentary from an export
Trigger: Controller finishes the margin variance report.
- Exports the report to CSV with customer names replaced by codes.
- Uploads it to a company ChatGPT workspace.
- Asks for the top five drivers and a draft paragraph.
- Checks the figures against the source report before using the text.
Result: Commentary drafted in minutes with the numbers still verified by a person.
Order status custom GPT
Trigger: A salesperson asks 'Where is order 48213?'
- The GPT calls the internal get_order_status action.
- The service queries a read-only view and returns status, ship date and last event.
- The GPT answers in plain language and cites the fields used.
Result: Sales gets an answer without opening the ERP, and the plant sees every request in the log.
Keeping data on-prem
ChatGPT is a hosted service, so by definition data you send leaves your network. That is acceptable for some plants and not for others, especially those handling ITAR, controlled unclassified information or customer NDAs that prohibit third-party processing. For those, keep DELMIAworks and its Oracle database on-prem and serve an open-weight model locally with the same tool layer, so the experience is similar but the data stays inside. The action endpoint you design for a custom GPT can be reused unchanged as the tool layer for a private model, which keeps the work from being thrown away.
Security checklist
- Use a business or enterprise workspace with admin controls, not personal accounts.
- Never upload data covered by export control rules or customer terms that bar third-party processing.
- Expose only read operations from the action endpoint and filter fields server side.
- Authenticate every call and log the user, request and response size.
- Review your vendor agreement for retention and regional processing terms.
Ways to get it done
Manual exports
Effort: Immediate, no build.
Fine for finance commentary and non-sensitive analysis by trained users.
Custom GPT with actions
Effort: One to three weeks including security review.
Good for read-only lookups on non-restricted data when a small internal API is acceptable.
Private model via partner such as Netray
Effort: A few weeks to a pilot.
Best when data is restricted or when you want ChatGPT-like chat over live ERP data on-prem.
Pitfalls
- !Staff using personal ChatGPT accounts with real ERP exports because no approved option exists.
- !Trusting generated numbers without checking them against the source report.
- !Exposing the Oracle database directly to an action instead of a narrow service.
- !Forgetting that custom fields can carry customer-confidential notes.
- !Treating a successful demo on clean data as proof it works on messy production data.
Frequently asked questions
Can ChatGPT connect directly to my DELMIAworks database?
Not by itself. You need a service that sits between them, reads from curated views or the vendor API, and returns limited JSON. Pointing a model at a production Oracle instance with broad access is a security and stability risk.
Is it safe to upload IQMS exports to ChatGPT?
It depends on the data and your plan. Enterprise or business workspaces add controls, but contracts, export rules and customer NDAs may still forbid it. Classify the fields first and sanitize anything restricted.
What is a custom GPT action?
It is a described API that a custom GPT can call during a chat. You host the API, document it with an OpenAPI file, and decide which data it returns. The model never gets your database credentials.
When should I use a private model instead?
When data cannot leave your network, when volume makes hosted costs unattractive, or when you want deep access to live ERP data without exposing it. Open-weight models served locally can handle most lookup and summarizing tasks.
Does ChatGPT replace DELMIAworks reports?
No. Reports remain the system of record for numbers. ChatGPT helps explain, summarize and draft around them, and any figure it states should be traced back to a report or query.
Want ChatGPT working on your DELMIAworks (IQMS)?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment