How to use ChatGPT with Baan IV and Baan V
Short answer
ChatGPT cannot reach Baan directly. You either paste or upload exports, or you put a small API in front of a read-only copy of the data and connect it as a custom GPT action. For sensitive costing and customer data, check your policy before using any hosted model.
Most Baan users who try ChatGPT start the same way: export a session to Excel, paste it in, and ask for analysis. That works for one-off questions, and it is a fair way to test whether the idea is useful. It does not scale, it is stale the moment you export, and it puts whole spreadsheets into a third-party service.
The next step is giving ChatGPT a controlled way to ask questions of Baan data. That requires an API, because ChatGPT calls HTTPS endpoints and Baan IV and V do not publish one. Someone has to build the bridge.
This page covers the realistic paths, the data handling questions your IT or compliance team will ask, and when to choose a private model instead.
What teams use it for
Explain a messy export
Upload a bill of material or routing export and ask for discrepancies, missing components or unusual quantities.
Write report specs from plain language
Describe a report a manager wants and have ChatGPT draft the field list and logic for your Baan report developer, who then builds it in the Baan reporting tools.
Decode legacy 3GL and report scripts
Paste a Baan 3GL script or a report script you do not understand and ask for a plain explanation, then verify the explanation against the behavior.
Supplier email drafts
Give it a list of late lines from a query and have it draft polite, specific follow-ups.
Custom GPT over an inventory API
A custom GPT with an action calls your own endpoint for stock by item and warehouse, with no free SQL.
Baan IV / Baan V integration surface
- Session exports: Baan sessions can print to file or export to ASCII, and many sites already schedule these for Excel use.
- ODBC and the underlying database: a read-only replica can serve a small HTTPS API that a custom GPT action can call.
- Baan reporting tools and print devices: reports can be directed to a file device, a handy low-effort route to feed documents to ChatGPT.
- AFS: for write-back to Baan from an approved action; do not expose it directly to ChatGPT.
- Table and field names vary by version and customization, so any API must hide them behind business names.
How it works
- 1
Decide which data may leave
Classify Baan data: item masters and public specs may be fine, while customer pricing, cost rolls and drawings may not.
- 2
Start with exports
Use scheduled exports of non-sensitive data to test usefulness before building anything.
- 3
Build a narrow API
If tests are valuable, write a small HTTPS service over a read-only copy with endpoints like stock_by_item and open_orders_by_customer.
- 4
Describe it as an OpenAPI spec
Custom GPT actions are defined by an OpenAPI schema. Keep descriptions clear so the model calls the right endpoint.
- 5
Add authentication
Use an API key or OAuth supported by the GPT builder, and map users to Baan roles so one person cannot see another's data.
- 6
Review data controls
Check your ChatGPT plan's retention and training settings with your legal team; business plans have different defaults than consumer accounts.
Example flows
Weekly open order analysis
Trigger: Planner exports open sales orders each Monday.
- Export goes to a folder, with customer names replaced by codes.
- Planner uploads it to ChatGPT.
- Asks which orders are at risk and why.
- Planner checks the top items in Baan before acting.
Result: Faster triage, no live connection needed.
Stock lookup in a custom GPT
Trigger: Sales rep asks whether an item is available.
- GPT calls the stock endpoint with the item code.
- Endpoint queries the replica and returns on hand and allocated.
- GPT answers in plain language with the as-of time.
Result: Quick answers without opening a Baan session.
Keeping data on-prem
ChatGPT is a hosted service, so by itself it is not an on-prem option. If your Baan data must stay inside your network, the usual substitute is an open-weight model on your own GPU server with the same custom API in front of Baan. You get a similar chat experience, but the model, logs and prompts never leave your environment. Some firms use both: ChatGPT for general drafting with no ERP data, and a private model for anything that touches Baan.
Security checklist
- Do not upload cost, pricing, customer or export-controlled technical data without explicit approval.
- Use a business or enterprise plan with training on your data disabled, and confirm retention terms in writing.
- Keep API endpoints read-only and rate limited.
- Authenticate users and enforce row-level filtering by role.
- Log every action call.
Ways to get it done
Manual exports into ChatGPT
Effort: Hours.
Early exploration with non-sensitive data.
Custom GPT with your own API
Effort: Weeks, mostly for the API and security review.
Repeated questions on stock, orders or items where policy allows hosted AI.
Private model with a Baan data layer, built with a partner like Netray
Effort: A scoped pilot, then expansion.
Regulated or sensitive data, or a policy against sending ERP data out.
Pitfalls
- !Pasting live customer prices into a personal account.
- !Trusting an explanation of a Baan script without testing it, since the model has seen little Baan code.
- !Stale exports presented as current stock.
- !An API that exposes table names and invites guesswork.
Frequently asked questions
Can ChatGPT connect directly to Baan?
No. Baan IV and V have no native HTTPS API for it to call. You need an intermediate service, or you work with exported files. The intermediate service is what you control, secure and audit.
Is ChatGPT good at Baan-specific code?
Only partly. Baan 3GL and report scripts are rare in training data, so explanations can be wrong. Treat them as a starting point and verify in a test environment.
Is it safe to upload Baan exports?
That depends on the data and your plan. Business plans offer stronger controls, but your policy and any customer contracts decide what may leave. Strip sensitive fields first when in doubt.
What about a custom GPT for the shop floor?
It can work if the endpoints are narrow and read-only. For shop floor users, response speed and clear answers matter more than model size, and a private model may be simpler to run.
When should we skip ChatGPT?
If your data is controlled, if contracts forbid outside processing, or if the work needs live write-back. In those cases use a private model with approvals.
Want ChatGPT working on your Baan IV / Baan V?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment