# How to use ChatGPT with JD Edwards World > Source: https://netray.co/integrations/jde-world-chatgpt > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Integrations ## Short answer ChatGPT cannot reach a World system on its own. You either export data into it, or publish a narrow, authenticated API that a custom GPT calls as an action. For regulated or defense data, a private model on your own hardware is usually the safer second step. ## System JD Edwards World (IBM i) ## Capability chatgpt ## Intro People on JD Edwards World often start with the simple thing: run a query, paste a result into ChatGPT and ask for a summary. That works for a one-off analysis, and for a lot of finance and supply chain staff it is the first AI win. The trouble shows up when the pasted data includes customer names, cost, drawings numbers or anything under contract. The next step is live access. ChatGPT supports custom GPTs with actions, which are calls to an HTTPS API you describe with an OpenAPI schema. World has no such API by default, so the question becomes who builds and secures that endpoint, and whether the data should leave your network at all. This page covers both paths and the point where a private model makes more sense. ## Use Cases ### Ad hoc analysis of exported data **Description:** Export an aged payables or slow-moving stock extract to a spreadsheet, remove sensitive columns and ask ChatGPT to find patterns or draft a summary for management. ### Explaining legacy RPG or CL code **Description:** Paste a non-sensitive RPG fragment or a CL program and ask for a plain-language explanation to help new developers learn how a customization works. ### Drafting supplier and customer correspondence **Description:** Use order and PO facts you pass in as context to draft polite chase-ups or delay notices, then send them from your own mail. ### Custom GPT for order lookup **Description:** A GPT with an action that calls a read-only order status endpoint so a sales rep can ask about an order in natural language. ### SQL and report assistance **Description:** Ask for help writing DB2 for i SQL against F-file column names you describe, then review and run the query yourself. ## System Surface - Query output from World screens, queries and reports saved as CSV or spreadsheet via IBM i Access Client Solutions, a file transfer or the IFS. This is the manual 'export then paste' path. - DB2 for i accessed by ODBC or JDBC from a small service that you host; that service becomes the API a custom GPT action can call. - IBM i Integrated Web Services or a REST layer written in Node, Python or Java on IBM i or a separate server. Availability depends on your OS release and policy. - Existing batch jobs that already drop files into the IFS, which you can pick up for a nightly refresh of a vetted dataset. ## How It Works ### Classify the data **Detail:** Decide which files and fields are public-safe, internal, or restricted. Cost, customer pricing and export-controlled technical data usually end up restricted. ### Start with sanitized extracts **Detail:** Create a SQL view or query that drops sensitive columns and masks names. Save as CSV and load into ChatGPT for the first pilot. ### Pick the right ChatGPT plan **Detail:** Use a business or enterprise plan with the data controls your counsel accepts, such as no training on your data and admin audit. Confirm retention terms in your contract. ### Build a narrow API for live access **Detail:** Host a small service that exposes two or three read-only endpoints over a views library. Describe them in an OpenAPI schema for the custom GPT action. ### Authenticate and limit **Detail:** Use OAuth or API keys tied to a service identity, rate limits and a row cap. Never put an IBM i password in the GPT configuration. ### Decide if the data should leave at all **Detail:** If a custom GPT call would carry restricted fields, stop and move that use case to a private model. ## Example Flows ### Monthly slow-mover review **Trigger:** Inventory controller wants a talking-points summary **Steps:** - A saved query lists items with no movement in 12 months by branch, with cost removed. - The result is exported to CSV via Access Client Solutions. - The controller uploads it to ChatGPT with a prompt for grouping and a draft memo. **Result:** A first-draft memo in minutes; the controller verifies values against World before sending. ### Custom GPT order status **Trigger:** Sales rep asks 'what's the status of order 51877' **Steps:** - The GPT calls the order_status action with the order number. - Your API checks the token, runs a view over the sales order tables and returns status, ship date and open lines. - The GPT answers in plain language and cites the fields it used. **Result:** Rep gets an answer without a call to customer service, and the API logs the request. ## On Prem If data cannot go to a public cloud, ChatGPT is the wrong place for it. Keep the same pattern, a read-only tool service over DB2 for i views, but point it at an open-weight model served on your own GPU hardware with vLLM or Ollama. The chat interface can look familiar to users, and the model never sees the public internet. Some teams keep ChatGPT for non-sensitive tasks such as code explanation and use the private model for anything with part numbers, pricing or customer data. That split is easy to explain to auditors. ## Security - Check your plan's data terms: training use, retention and region. Do not rely on the free tier for company data. - Mask or drop restricted columns before export; keep a record of what was shared. - Custom GPT action endpoints must authenticate every call and return only the minimum fields. - Export-controlled technical data (for example under ITAR or EAR) should not go to a shared cloud service unless your compliance team has cleared that route. - Review who in the company can create or share custom GPTs that call your API. ## Options ### Manual exports into ChatGPT **Fit:** Occasional analysis on non-sensitive data. Good for proving value fast. **Effort:** Hours, but it does not scale and depends on discipline. ### Custom GPT with an action over your own API **Fit:** Live lookups for a small group, with data that is acceptable to send to the cloud. **Effort:** A few weeks, mostly the API and security review. ### Private model with the same tool layer, via a partner like Netray **Fit:** Regulated, contract-bound or competitive data that must stay on your network. **Effort:** Scoped pilot; reuse the API design and swap the model. ## Pitfalls - Pasting full extracts with costs and customer names 'just this once'. It becomes habit. - Giving the GPT a database login. Always use a purpose-built API with limited views. - Dates in Julian format and packed numbers confuse the model; convert in the view before export. - Assuming ChatGPT knows your customizations. World shops carry decades of local logic that the model has never seen. - No logging of what was asked and returned, which makes later review impossible. ## Faq ### Can ChatGPT connect directly to JD Edwards World? **Answer:** Not directly. ChatGPT calls HTTPS APIs you describe, and World does not ship a general REST layer. You export data manually or build a small service over DB2 for i that a custom GPT calls. Either way you are responsible for what leaves your network. ### Is it safe to put World data into ChatGPT? **Answer:** It depends on the plan, the data and your obligations. Business and enterprise plans offer stronger data controls than free ones, but contract, export-control and customer confidentiality rules can still forbid it. Classify the data first, and keep restricted fields out. ### Can ChatGPT help with our RPG and CL code? **Answer:** Yes, for explaining and drafting, as long as the code is not confidential. Treat the output as a starting point and test it on a non-production partition. It can misread fixed-format RPG or invent commands, so review matters. ### Why would we use a private model instead? **Answer:** Control. A private model on your own server keeps prompts, rows and logs inside your network, which simplifies customer and regulator conversations. You trade some raw model capability and take on operations, so match the model to the task. ### Does a custom GPT work with World's security? **Answer:** Only if you make it. World's menu and file security do not apply to a separate API, so you must replicate the access rules in your service and map each GPT user to a role. Keep the first release read-only and small. ## Search terms jd edwards world chatgpt, chatgpt for jd edwards world, jd edwards world chatgpt integration, use chatgpt with ibm i erp, custom gpt jd edwards, jd edwards world openai api, chatgpt as400 erp data, jd edwards world generative ai, private alternative to chatgpt for jd edwards --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json