How to use ChatGPT with SYSPRO
Short answer
You can use ChatGPT with SYSPRO in three ways: paste or upload exports, give a custom GPT actions that call a small API you host in front of SYSPRO, or move to a private model when the data is too sensitive. Live access needs a middle layer, because ChatGPT cannot reach an on-prem SYSPRO server on its own.
People on SYSPRO usually start with ChatGPT the same way: export a grid to Excel, paste it into a chat, and ask for a summary. That works for a one-off analysis, and it is a useful way to find out what questions are worth automating. It does not scale, because the data is stale the moment it is exported and someone has to repeat the steps every time.
The next step is giving ChatGPT live access. A custom GPT can call external actions described by an OpenAPI schema, which means you can publish a few well-defined endpoints, such as stock by item or order status, and let the GPT call them. SYSPRO does not ship that API for you in a form ChatGPT understands, so you build a thin service over e.NET or SQL Server.
The third question is where the data goes. For finished-goods stock levels this may not matter. For customer pricing, costing, drawings or defense work it often does, and that is where private model options enter.
What teams use it for
Explain a variance report
Upload a job costing or inventory movement export and ask ChatGPT to explain the largest variances and suggest what to check in SYSPRO next.
Draft customer replies on order status
A CSR pastes an order number's status from SYSPRO, and a custom GPT drafts a clear reply with ship dates and any caveats.
Write SQL and report logic
Planners and analysts describe a report in plain English and get draft SQL for a read-only SQL Server login or logic for SYSPRO Reporting Services, which they test before use.
Summarise supplier performance
Feed in receipt history exports and ask for on-time delivery patterns by supplier, with caveats about small samples.
Clean up item master data
Upload an item list and ask for suggested naming standards, duplicate candidates and missing fields, then load corrections through a controlled import.
Prepare internal training material
Turn procedures for goods receipt, job issue or ECN handling into short step-by-step guides for new staff.
SYSPRO integration surface
- Exports: SYSPRO grids and reports export to Excel, CSV or PDF, and SYSPRO Reporting Services can schedule exports to a folder. This is the simplest and least risky path for ChatGPT file uploads.
- e.NET business objects: query objects such as INVQRY or SORQRY return XML you can translate to JSON in a wrapper. Names and parameters vary by version, so check the e.NET documentation for the build you run.
- SQL Server: a read-only reporting login on the company database lets a wrapper run fixed, parameterised queries over tables like InvMaster, SorMaster and PorMasterDetail.
- A published wrapper API: ChatGPT custom GPT actions need an HTTPS endpoint with OpenAPI schema and OAuth or API key authentication, so you host a small service in a DMZ or cloud that calls back into SYSPRO.
- SYSPRO on-prem networking: most installs are not reachable from the internet, so any live ChatGPT connection needs a reverse proxy, gateway or tunnel decision that your IT team signs off.
How it works
- 1
Decide what ChatGPT is allowed to see
List the data classes involved: stock, prices, costs, customer names, drawings. Mark which can leave your network under your policies and contracts.
- 2
Start with controlled exports
Schedule a few SYSPRO exports to a shared folder, strip sensitive columns, and let staff upload them to an approved ChatGPT workspace.
- 3
Build a small read API
Create endpoints such as GET /items/{code}/stock and GET /orders/{number}/status that call e.NET or parameterised SQL and return trimmed JSON.
- 4
Describe the API in OpenAPI
Write an OpenAPI file with short, unambiguous descriptions for each operation. The description text is what the GPT uses to decide when to call it.
- 5
Configure the custom GPT
Add the schema as an action, set authentication, and write instructions that tell the GPT to cite which call produced each figure and to say when data is missing.
- 6
Test with real questions
Gather 30 questions from planners and CSRs, run them, and check every number against SYSPRO before rolling out.
- 7
Review the privacy position
Confirm your ChatGPT plan's data-use and retention terms, admin controls and regional options, and decide whether sensitive workflows move to a private model.
Example flows
Order status from a custom GPT
Trigger: A CSR asks, 'Where is order 004512?'
- The GPT calls the order status action with the order number.
- The wrapper queries SORQRY-style data and the dispatch note tables.
- The API returns lines, quantities shipped, and next ship date.
- The GPT drafts a reply using only the returned fields.
Result: The CSR gets a grounded draft in seconds and sends it after a quick check.
Weekly inventory review from an export
Trigger: A planner uploads the Friday slow-moving stock export.
- ChatGPT groups items by warehouse and age bucket.
- It highlights items with no movement and open purchase orders.
- It proposes questions to ask the buyer before cancelling anything.
Result: A review agenda with specific items instead of a raw spreadsheet.
Keeping data on-prem
ChatGPT is a cloud service, so an on-prem SYSPRO cannot be queried by it directly without an inbound path. Some sites accept a narrow, authenticated gateway for non-sensitive endpoints. Others decide that costing, customer pricing or export-controlled data must not leave the premises and use an open-weight model hosted next to SQL Server instead, with the same OpenAPI-style tool layer. The custom GPT can then be kept for public or low-risk content only. Whatever you choose, document it so auditors and customers can see where each data class is processed.
Security checklist
- Use a business or enterprise ChatGPT plan with admin controls, and confirm current terms on training use of your data rather than assuming them.
- Expose only the fields the question needs; leave out cost rollups, bank details and personnel data from API responses.
- Authenticate the wrapper with OAuth or signed keys, rotate credentials, and bind each GPT user to an identity you can log.
- Do not paste export-controlled technical data or customer drawings into any cloud tool unless your compliance owner has approved it.
- Log every action call with user, endpoint and row count so you can audit what the GPT retrieved.
Ways to get it done
Manual export and upload
Effort: Hours.
Fine for ad hoc analysis on non-sensitive data and for discovering which questions matter.
Custom GPT with actions over your own read API
Effort: About 2 to 6 weeks including security review.
Good for live order, stock and item lookups for a team already on ChatGPT.
Private model with a Netray-style tool layer
Effort: A pilot in weeks, plus GPU hardware or a private cloud.
Right when data cannot leave the building, or when you want consistent behaviour independent of a cloud vendor.
Pitfalls
- !Letting ChatGPT do arithmetic on exported rows leads to quiet errors, so have the API return totals computed in SQL.
- !Large exports exceed context limits and get truncated without warning, so aggregate before upload.
- !Item codes and customer IDs in SYSPRO are often padded or case-sensitive, and the GPT may normalise them wrongly, so validate inputs in the API.
- !Publishing an endpoint with weak authentication to make the demo work leaves a permanent hole in the network.
- !Skipping a data classification step means someone eventually pastes something they should not.
Frequently asked questions
Can ChatGPT connect directly to my SYSPRO database?
No, and you should not want it to. ChatGPT cannot reach a private network, and a direct connection would bypass access control. Use a small API you control that runs fixed queries and returns limited results, then let a custom GPT call that API.
Is there an official SYSPRO ChatGPT plugin?
Availability changes over time, and SYSPRO has been adding its own AI capabilities. Check the current SYSPRO release notes and partner marketplace. If nothing fits, a custom GPT over your own API gives you control over scope and security.
Is pasting SYSPRO exports into ChatGPT safe?
It depends on the data and your plan. Business and enterprise plans offer admin controls and data-use commitments, but your own contracts and export rules still apply. Strip sensitive columns and get compliance approval for anything involving customer or technical data.
Will ChatGPT give correct numbers from SYSPRO?
Only when numbers come from a live call or a clean export and are computed by the database, not the model. Ask it to cite the source call, and verify a sample against SYSPRO screens before relying on results.
When should I use a private model instead?
Choose a private model when you handle defense, aerospace or customer-confidential data, when policy forbids cloud processing, or when you need predictable cost at high volume. The same tool-layer design carries over, so early work is not wasted.
Want ChatGPT working on your SYSPRO?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment