How to use ChatGPT with Macola
Short answer
There are three ways to use ChatGPT with Macola: paste or upload exports for ad hoc analysis, connect a custom GPT to a small API gateway that fronts Macola, or skip OpenAI hosting and run a private model. The right one depends on how sensitive the data is and whether a person or a workflow is asking.
Many Macola users already use ChatGPT on the side. A planner pastes a shortage report and asks for a summary, an accountant uploads an aging export and asks for the top risks. It works, and it is also an unmanaged data flow out of the ERP. The question for IT is how to make it deliberate.
Macola does not ship a ChatGPT connector that I can point you to, so any live access is something you build or buy. Because many Macola systems are on-prem behind a firewall, a custom GPT cannot simply reach into them. You need an intermediate service that you control.
This page walks the realistic paths, from the lowest risk (exports) to live actions, and explains where a private model starts to make more sense than ChatGPT.
What teams use it for
Ad hoc analysis of a Crystal Reports export
Export open orders or aging to CSV, strip columns you do not need, and ask ChatGPT to find late customers, concentration risk or odd patterns.
Drafting customer and supplier emails
A user pastes an order's status lines and asks ChatGPT to draft a polite delay notice or an expedite request in the right tone.
Explaining a report or customization
Paste a Crystal formula or a stored procedure into ChatGPT and ask for a plain explanation before changing it.
A custom GPT for read-only lookups
A custom GPT with an action calls your gateway for item availability or order status, returning only what the user's role may see.
Spreadsheet cleanup before import
Normalize a supplier's price list or item descriptions with ChatGPT, then review and load through Macola's normal import tools.
Macola integration surface
- Exports: Macola reports, Crystal Reports and data export tools produce CSV or Excel. This is the simplest and most controllable way to put data in front of ChatGPT.
- Database or ODBC: a gateway service can read curated views from the Macola database (SQL Server on newer lines, Progress ODBC on older ones) and expose a few HTTPS endpoints.
- Web services or SDK: where your Macola release has them, they can back write endpoints. Names and availability vary by version.
- Network position: ChatGPT actions call a public HTTPS URL. For on-prem Macola this means a gateway in a DMZ or a reverse proxy, which is itself a security decision.
- Application security: Macola user roles do not follow a database login, so the gateway has to reimplement who may see costs, margins and customers.
How it works
- 1
Decide what data may leave
Classify fields: public, internal, restricted. Pricing, customer lists and drawings are usually restricted. Write the rule down before anyone builds anything.
- 2
Pick the ChatGPT plan with the right controls
Business or enterprise plans offer admin controls and options about training on your content. Verify the current terms with OpenAI rather than relying on a blog, including data retention.
- 3
Start with governed exports
Publish a few saved reports designed for ChatGPT use, with restricted columns removed, and put them in a shared location with an acceptable use note.
- 4
If you need live data, build a gateway
A small service reads curated Macola views and exposes read-only endpoints with an OpenAPI spec. It authenticates the user and applies row and column rules.
- 5
Create the custom GPT action
Import the OpenAPI spec as an action, use OAuth or an API key per your plan's options, and limit each endpoint to small, typed results.
- 6
Review usage
Log gateway calls, review them monthly, and retire anything people do not use.
Example flows
Planner uploads a shortage export
Trigger: Planner finishes the weekly shortage report.
- Planner exports the report to CSV with part numbers and quantities only.
- Uploads it to ChatGPT with a standard prompt.
- ChatGPT groups shortages by supplier and likely cause.
- Planner copies the supplier groupings into the follow-up emails.
Result: Faster weekly prep with no live connection and a clear boundary on what left the company.
Custom GPT order status
Trigger: Sales rep asks 'where is order 12345?'
- GPT calls the gateway's order status endpoint.
- Gateway checks the user, queries the Macola view and returns header, ship dates and invoice status.
- GPT summarizes in plain language.
Result: A status answer in seconds, scoped to what that rep may see.
Keeping data on-prem
If policy says ERP data stays in the building, ChatGPT is the wrong place for it, however convenient. The alternative is to keep the same interaction pattern but swap the model: an open-weight model such as Llama, Qwen or Mistral served on your own GPU, fronted by the same kind of gateway, with a chat interface on the intranet. Users keep the experience they like, and prompts, outputs and logs stay on your network. A hybrid is also reasonable: ChatGPT for general drafting and coding help with no ERP data, and a private model for anything that touches orders, costs or customers. Defense, aerospace and export-controlled work usually belongs in the private bucket.
Security checklist
- Never put customer, pricing or controlled technical data into a personal ChatGPT account. Use a managed workspace.
- Gateway endpoints must be read-only, rate limited and authenticated per user, not a shared key.
- Return the minimum fields needed for the answer, not entire tables.
- Review the vendor's current data retention and training terms and record the date you checked.
- Do not expose the Macola database directly to the internet. Put the gateway in a DMZ with strict inbound rules.
Ways to get it done
Governed exports with a usage policy
Effort: A day or two to define reports and rules.
Small teams who want value this week and have non-sensitive reports.
Custom GPT over your own gateway
Effort: About 3 to 6 weeks including security review.
Teams that want live read-only lookups and accept an internet-facing service.
Private model with a similar interface, with a partner like Netray
Effort: A scoped pilot of a few weeks plus GPU procurement time.
Regulated or on-prem-only environments.
Pitfalls
- !Pasting full customer lists or price books into a consumer account because it was quick.
- !Assuming Macola roles carry through to the gateway when they do not.
- !Large exports that exceed context limits and silently get truncated, leading to wrong totals.
- !Trusting arithmetic from the model on financial data. Have it explain, and compute totals in SQL.
Frequently asked questions
Is there an official ChatGPT connector for Macola?
I am not aware of a vendor-supported one, so treat live access as something you build through a gateway or obtain from a third party. Check with your Macola reseller for anything current, since add-ons change.
Can ChatGPT connect straight to our Macola database?
Not directly and not safely. ChatGPT actions call HTTPS endpoints, so you need a service in between that handles authentication, limits and logging. Exposing the database server itself is not advised.
Is it okay to upload Macola exports?
It can be, if the export is limited to non-restricted fields and you use a managed workspace with the right data controls. Define allowed reports in policy so users are not guessing.
Why would we choose a private model instead?
When data is restricted, such as export-controlled designs or contract pricing, or when you must keep a full audit trail on your own systems. Open-weight models are good enough for lookups, summaries and drafting.
Can ChatGPT change data in Macola?
Technically it can if you build a write endpoint, but that is rarely wise early on. Prefer proposals that a person approves, executed by a service through a supported Macola path.
Want ChatGPT working on your Macola?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment