How to use ChatGPT with Priority ERP
Short answer
You can use ChatGPT with Priority in two ways: paste or upload exported data, or give a custom GPT or connector controlled access to the Priority OData API. Live access needs a reachable, authenticated endpoint, which is a real design decision for on-prem servers.
Many Priority users already paste a report or a spreadsheet into ChatGPT to summarize it or draft an email. That works for one-off questions, but it has limits: the data goes stale the moment you export it, someone has to remember to do it, and it is easy to upload more than you meant to.
The next step is a connection where ChatGPT can ask Priority for specific records. In practice that means defining a small set of actions that call an API you control, rather than handing ChatGPT your Priority credentials. The API you put in front of Priority is the part that decides what the model can see.
Before choosing, decide whether the data is allowed to leave your network at all. For many Priority customers in defense, aerospace and electronics supply chains the answer is no for part of the data, and the design changes accordingly.
What teams use it for
Summarize a Priority report
Export an open orders or aged receivables report and ask ChatGPT to highlight the top risks and draft follow-up messages. Good for non-sensitive, point-in-time analysis.
Custom GPT for order status
A custom GPT with one action that calls a thin API returning order header, lines and shipment status for a given order number. Sales staff ask in plain words and get a sourced answer.
Drafting supplier correspondence
Provide the purchase order lines and delivery history, and have ChatGPT draft a firm but polite chase email in the supplier's language.
Explaining a variance
Paste the cost roll-up for a part before and after a price change and ask for a plain explanation for the finance lead.
Writing SQL+ or procedure logic
Developers use ChatGPT to draft report or procedure code, then test it in a non-production environment. Output must be reviewed because syntax and field names vary by version.
Priority ERP integration surface
- The Priority OData service is the natural target for a custom GPT action: you describe a small set of endpoints in an OpenAPI schema and ChatGPT calls them on the user's behalf.
- Priority forms can usually be exported to Excel or CSV, and reports can be run and saved, which supports the manual upload route.
- Cloud-hosted Priority is reachable from the internet subject to your allow-listing. Self-hosted servers are often behind a firewall, so you would normally publish a narrow gateway rather than the Priority server itself.
- Authentication to Priority is typically a service user or application credentials. ChatGPT should never hold these. Your gateway holds them and issues limited access to the GPT.
- Exact OData paths, entity names and header conventions vary by version and tabula.ini setup, so test against your own environment.
How it works
- 1
Classify the data
List the fields the use case needs and mark any that are restricted, such as controlled technical data, customer pricing or personal data. Exclude them from the design up front.
- 2
Build a thin gateway
Create a small service that exposes two to five read-only endpoints, each mapped to a Priority OData query with fixed filters and row limits. It holds the Priority credentials.
- 3
Publish an OpenAPI description
Describe the gateway endpoints precisely, with parameter names and short descriptions, so the model calls them correctly.
- 4
Create the custom GPT or connector
Add the action, configure authentication (OAuth is preferable to a shared key), and write instructions that tell the GPT to cite the order or part number from every answer.
- 5
Pilot with a few users
Review the actual questions asked and the data returned. Tighten the endpoints where users ask for more than they should see.
Example flows
Order status through a custom GPT
Trigger: Salesperson types an order number
- GPT calls the gateway order-status action.
- Gateway queries the Priority sales order and linked delivery documents via OData.
- Gateway returns a trimmed JSON object.
- GPT replies with status, promised date and any open lines.
Result: A sourced status answer without opening Priority or asking a colleague.
Weekly receivables summary by upload
Trigger: Finance lead exports the aged debt report
- Remove customer contact details from the file.
- Upload it to a ChatGPT workspace configured for business data.
- Ask for the top ten overdue accounts and suggested wording.
Result: A first-draft collections plan, reviewed and edited by a person.
Keeping data on-prem
If part of your Priority data cannot go to a hosted model, you have two honest options. One is to restrict ChatGPT to the non-sensitive slice through the gateway. The other is to use a model running inside your network and keep ChatGPT out of the loop for the regulated data. Self-hosted Priority customers often run both: a hosted assistant for general work, and a local open-weight model behind the same gateway for anything touching controlled drawings, customer pricing or defense program data.
Security checklist
- Review your ChatGPT plan's data-use and retention terms and confirm that business data is excluded from model training before sending anything.
- Keep Priority credentials in the gateway only, and give the GPT a scoped token with a short lifetime.
- Return only the fields needed for the answer, and enforce row limits server side.
- Log every action call with the user identity so you can answer who asked what.
- Do not upload exports containing personal or controlled data to a general consumer account.
Ways to get it done
Manual export and upload
Effort: None
Fine for occasional analysis of non-sensitive reports. No integration work, but no live data and weak control.
Custom GPT with gateway actions
Effort: One to three weeks
Best for read-only lookups by a defined user group when hosted processing is acceptable.
Private model behind the same gateway, via Netray or in-house
Effort: Scoped project
Right when data may not leave your environment, or when you want one assistant across ERP, documents and files.
Pitfalls
- !Exposing the whole OData service to ChatGPT instead of a narrow gateway.
- !Relying on the GPT's memory for figures rather than forcing it to call the action each time.
- !Forgetting that exports from Priority can include remarks and attachment names with sensitive text.
- !Using one shared service account so nobody can tell who asked for what.
Frequently asked questions
Is there an official ChatGPT connector for Priority?
I am not aware of a standard one that fits every setup, and availability changes. Check the Priority partner marketplace and the ChatGPT app directory. If none fits, a gateway plus a custom GPT action is the usual route.
Can ChatGPT write back to Priority?
It can if you build a write action, but that is rarely a good first step. Start with read-only access. If you add writes, have the action create a draft that a person approves, and keep the Priority user tightly scoped.
Will my Priority data be used to train models?
That depends on your plan and settings, not on Priority. Read the current terms for your ChatGPT tier, choose a business plan with training excluded, and confirm with your legal or security team.
What if our Priority server is not on the internet?
Then a hosted ChatGPT cannot reach it directly. Either publish a narrow gateway through your DMZ, or run a local model inside the network that calls Priority over the LAN.
Is Priority's data good enough for a model to use?
Mostly yes if the questions are about orders, stock and purchasing. Quality depends on your master data. Test with real questions and fix the gaps the tests expose.
Want ChatGPT working on your Priority ERP?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment