# How to use ChatGPT with Epicor Kinetic > Source: https://netray.co/integrations/epicor-kinetic-chatgpt > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Integrations ## Short answer There are three workable routes: export BAQ results for ad hoc analysis, build a custom GPT whose Action calls a small gateway in front of Kinetic's REST API, or skip ChatGPT and run a private model. Always put a gateway between ChatGPT and Kinetic rather than exposing the ERP directly. ## System Epicor Kinetic ## Capability chatgpt ## Intro Someone on your team is already pasting Epicor reports into ChatGPT. It works for one-off analysis, and it also moves part numbers, customer names and pricing into a third-party service without a policy. The question is how to make that useful and controlled. ChatGPT cannot reach an on-prem Kinetic server by itself. A custom GPT can call an HTTPS API you describe with an OpenAPI schema, which means you must publish a small, authenticated endpoint that wraps a handful of BAQs. Kinetic's own REST API is too broad to hand over as is. This page walks through the options in order of risk, and says plainly when the right answer is a model you host. ## Use Cases ### Explain a variance **Description:** Export a job cost BAQ (estimated versus actual labor, material and burden) and ask ChatGPT to explain the largest variances in plain language for the plant manager. ### Order status for sales **Description:** A custom GPT with one Action returns sales order and shipment status by order number, so inside sales stops opening four Kinetic screens. ### Draft supplier communications **Description:** Give ChatGPT a late-release list from a BAQ and ask for polite, specific emails per supplier. A buyer edits and sends. ### Translate and summarise customer POs **Description:** Paste a customer's long PO terms or an engineering change notice and ask for the points that affect your routing or lead time. ### Write BAQ and report logic **Description:** Ask ChatGPT to propose the join logic for a new BAQ or an SSRS-style report from field descriptions, then validate in the Query Builder yourself. ## System Surface - BAQs executed through the BaqSvc OData endpoint, or exported from the BAQ Designer to Excel or CSV, are the cleanest data to hand to ChatGPT. - The Kinetic REST API v2 (OData) can back a gateway, but expose only specific read calls, not the whole business object catalogue. - Kinetic API keys and service accounts: a gateway should hold these server side, never the GPT configuration. - Epicor Functions let you publish a purpose-built endpoint that returns exactly the shape the GPT needs. - Epicor's own Data Discovery and dashboards are alternatives when the question is purely exploratory. ## How It Works ### Decide what data class may leave your network **Detail:** Separate harmless fields (part descriptions, dates) from sensitive ones (customer pricing, export-controlled part numbers). ### Create narrow BAQs **Detail:** One BAQ per question, with only the columns the answer needs and a row cap. ### Stand up a gateway **Detail:** A small HTTPS service that authenticates the caller, calls the BAQ with a service account, filters columns, and returns JSON. It logs every call. ### Describe the gateway with OpenAPI **Detail:** Create a custom GPT, add the schema as an Action, and use OAuth or an API key that you can revoke. ### Set instructions and test with real questions **Detail:** Tell the GPT to answer only from returned data and to say when it has none. Test with ten questions people actually ask. ### Review workspace controls **Detail:** Use a business or enterprise ChatGPT plan with the retention and training settings your policy needs. ## Example Flows ### Where is my order **Trigger:** A salesperson asks the custom GPT for the status of an order. **Steps:** - The GPT calls the gateway's order status Action with the order number. - The gateway runs the BAQ and returns lines, ship dates and carrier tracking. - The GPT summarises, flagging lines that are behind. **Result:** A plain answer in seconds, with no Kinetic login for the salesperson. ### Monthly scrap review **Trigger:** A quality manager uploads an exported scrap BAQ. **Steps:** - The file is analysed for top reason codes by work center. - ChatGPT proposes three hypotheses and what to check. - The manager copies the findings into the review deck. **Result:** Faster review prep, with no live ERP access granted. ## On Prem If your Kinetic server is on-prem and the data is controlled, ChatGPT is usually the wrong destination. A custom GPT needs a publicly reachable endpoint, which means opening a path from the internet to something that fronts your ERP. Prefer a private deployment: an open-weight model served with vLLM or Ollama inside your network, using the same BAQ-backed tools, with an internal chat UI. You keep the pattern (narrow tools, role-scoped service account, logs) and drop the third-party data flow. Netray builds this kind of private setup for ERP data. ## Security - Do not give ChatGPT direct credentials to Kinetic. Credentials live only in your gateway. - Return the minimum columns and rows. Mask customer pricing unless the asker's role allows it. - Use enterprise plan settings so that conversations are not used for training, and verify them yourself. - Rate-limit and log every Action call with the identity of the human behind it. - Export-controlled technical data should not go to ChatGPT at all. ## Options ### Manual export and paste **Fit:** Occasional analysis on non-sensitive data. **Effort:** None, but needs a clear policy. ### Custom GPT with a gateway **Fit:** Read-only lookups for a small trusted group. **Effort:** One to three weeks including security review. ### Private model with Netray or in-house **Fit:** Regulated, competitive or on-prem data. **Effort:** Weeks for a pilot on a GPU server. ## Pitfalls - Publishing the full Kinetic REST API as the GPT's Action and letting the model choose any call. - Treating screenshots and CSVs as harmless when they contain customer pricing. - Trusting the model's arithmetic on large exports. Do totals in the BAQ. - Not deciding who owns the GPT, so it breaks when its builder leaves. ## Faq ### Can ChatGPT connect directly to Epicor Kinetic? **Answer:** Not on its own. A custom GPT can call an HTTPS API you define, so you build a gateway in front of Kinetic's REST API or BAQs. Pointing it straight at the ERP is not advisable because it widens the attack surface and hands the model more than it needs. ### Is it safe to upload Epicor exports to ChatGPT? **Answer:** It depends on the data and the plan. Business plans offer stronger data controls than consumer accounts, but export-controlled, customer-confidential or pricing data may still be barred by contracts or regulation. Check your policy first. ### Do I need a custom GPT, or is the API better? **Answer:** A custom GPT is the fastest way to give a team a chat interface. If you want the answers inside your own portal or Teams, use the OpenAI API with the same gateway and tools instead. ### What about the cloud version of Kinetic? **Answer:** The pattern is the same, but your gateway reaches the Epicor cloud tenant's API with its own key. You are bounded by the API limits and terms of your subscription, so ask Epicor about volume. ### When should I choose a private model instead? **Answer:** When data cannot leave your control, when you need predictable cost at high volume, or when an auditor needs to see exactly where data flows. Open-weight models are good enough for most lookup and summarisation tasks. ## Search terms epicor kinetic chatgpt, chatgpt for epicor, epicor chatgpt integration, custom gpt epicor, chatgpt epicor baq, connect chatgpt to epicor kinetic, epicor openai integration, chatgpt erp data security epicor, epicor kinetic gpt actions --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json