# How to use ChatGPT with Made2Manage > Source: https://netray.co/integrations/made2manage-chatgpt > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Integrations ## Short answer ChatGPT cannot reach an on-prem Made2Manage database by itself. You either paste or upload exports for one-off analysis, or put a small read-only API in front of the data and connect it as a custom GPT action or app. For customer drawings and cost data, many shops prefer a private model instead. ## System Made2Manage ## Capability chatgpt ## Intro Plenty of Made2Manage users already paste a shortage list or a job cost report into ChatGPT and ask for a summary. It works, and it is the fastest start. The limits show quickly: the data is a stale snapshot, someone has to export it, and it is easy to upload more than policy allows. The next step is live access: a question like 'which open jobs for customer X are waiting on material' answered from current data. That needs a thin service between ChatGPT and your SQL Server, and a decision about what you are willing to send to a third party. This page lays out the three levels: manual exports, a custom GPT with actions, and a private alternative, so you can pick the lowest-risk one that solves your problem. ## Use Cases ### Explain a cost variance **Description:** Upload a job's estimated versus actual report and ask where hours and material drifted, then get a plain-language explanation for a review meeting. ### Clean up an item master export **Description:** Paste duplicate-looking part descriptions and ask for normalised names and likely duplicates, which a person confirms before any edit. ### Draft supplier and customer emails **Description:** Give ChatGPT a late PO line or a delayed order and have it draft a polite, specific email, using only the fields you provide. ### Write SQL or Crystal Reports ideas **Description:** Describe a report you want and get a draft query to review against your schema. Never run it unreviewed on production. ### Summarise a quality history **Description:** Feed a de-identified set of rework notes and ask for recurring root causes by operation. ## System Surface - No direct connection exists between ChatGPT and an on-prem Made2Manage. You must supply data by export or through an API you host. - Reports and screens can usually export to CSV or Excel, which is the lowest-friction manual path. - A read-only SQL Server login behind your own small web API (for example a REST service exposing a few endpoints) is the usual live path. Names and schema vary by version. - Custom GPT actions require an HTTPS endpoint with an OpenAPI description, so the service must be reachable, which is a real network decision for on-prem shops. - Check what your Aptean agreement and version offer for official APIs before building your own. ## How It Works ### Decide the data classification **Detail:** List what is allowed to leave: part numbers, prices, customer names, drawings. Write it down before anyone uploads. ### Start with manual exports **Detail:** Use reports exported to CSV and a ChatGPT plan with the data controls your policy requires. Treat output as a draft. ### Build a narrow read-only API **Detail:** Expose two or three endpoints such as order status by sales order and open lines by part, with row limits and a service account. ### Describe it as an action **Detail:** Write an OpenAPI spec and attach it to a custom GPT with clear instructions about when to call each endpoint. ### Secure the endpoint **Detail:** Use an API key or OAuth, restrict by IP where possible, and put the service in a DMZ rather than on the ERP server. ### Review usage and decide on private hosting **Detail:** If the useful questions all involve sensitive data, move the same API to a private model instead. ## Example Flows ### Monthly variance review prep **Trigger:** Controller exports the job cost report to CSV. **Steps:** - Remove customer names if policy requires. - Upload to ChatGPT and ask for the ten largest overruns with likely causes. - Cross-check each claim against the source rows. **Result:** A first-draft talking sheet in minutes, verified by the controller. ### Order status custom GPT **Trigger:** A salesperson asks the GPT about a sales order. **Steps:** - The GPT calls your status endpoint with the order number. - The service queries SQL Server read-only and returns a compact JSON. - The GPT answers in plain text and cites the fields it used. **Result:** Salespeople self-serve status without calling the office. ## On Prem If data cannot go to a third-party cloud, ChatGPT is the wrong endpoint, not the wrong idea. Keep the same pattern, a read-only API plus a chat interface, but serve an open-weight model locally with vLLM or Ollama and a small internal chat page. Your SQL Server stays on the LAN, the model runs on a GPU box beside it, and no part number or price crosses the firewall. You lose some raw model strength compared with the largest hosted models, which is an honest trade-off, but for lookups, summaries and drafting a mid-size open model is usually sufficient. ## Security - Use enterprise or business plans with training opt-out and retention terms reviewed by your security lead. - Never upload export-controlled drawings or technical data to a consumer account. - Give any API a dedicated read-only account and expose only the fields needed. - Authenticate every call, rate-limit, and log requests. - Remember that anything the GPT can read, any user of that GPT can ask for; scope by audience. ## Options ### Manual exports into ChatGPT **Fit:** One-off analysis with non-sensitive data and no integration budget. **Effort:** Hours; no engineering. ### Custom GPT with a small API **Fit:** Repeatable lookups for staff who already use ChatGPT, and data you are cleared to share. **Effort:** A few weeks including security review. ### Private model via a partner such as Netray **Fit:** Cost, customer or controlled data must stay in your network. **Effort:** A scoped pilot over several weeks. ## Pitfalls - Uploading a full cost file to a personal account because it was convenient. - Exposing the API to the internet without authentication to make the GPT work. - Assuming the GPT's answer is right when the export was filtered or out of date. - Letting the GPT infer business rules (like costing method) that it cannot see. ## Faq ### Can ChatGPT connect directly to Made2Manage? **Answer:** No. There is no built-in connector to an on-prem Made2Manage. You either export data manually or host a small API that ChatGPT can call. Which is right depends on how sensitive the data is and whether you can safely expose an endpoint. ### Is it safe to upload Made2Manage reports? **Answer:** It depends on the plan and the data. Business and enterprise plans offer data controls, but you must still follow your own policy and any customer contracts. Do not upload export-controlled or customer-restricted material to a public service. ### What is a custom GPT action in this context? **Answer:** It is a described API call the GPT can make during a chat. For Made2Manage that means your own HTTPS service returning order, job or inventory data. The GPT decides when to call it based on your instructions and the OpenAPI description. ### Why would we choose a private model instead? **Answer:** Mainly data control. Cost rates, customer pricing and drawings are often sensitive. A locally hosted open-weight model gives the same chat-over-data experience with nothing leaving your network, in exchange for hardware and some capability trade-offs. ### Can ChatGPT update jobs or POs? **Answer:** It should not, at least at first. Writes can bypass ERP business rules. Keep ChatGPT read-only and let it draft messages or change requests that a person enters or approves. ## Search terms made2manage chatgpt, chatgpt for made2manage, made2manage chatgpt integration, made2manage custom gpt, use chatgpt with made2manage data, made2manage openai, chatgpt manufacturing erp on-prem, made2manage private ai --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json