# How to use ChatGPT with Microsoft Dynamics GP

> Source: https://netray.co/integrations/dynamics-gp-chatgpt
> Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems.
> Section: Integrations

## Short answer

There are three ways to use ChatGPT with GP: paste or upload exports, a custom GPT with actions that call a small API you host in front of GP, or an enterprise connector. GP has no public cloud API, so you must expose something yourself. For regulated or customer-confidential data, a private model is often the better answer.

## System

Microsoft Dynamics GP

## Capability

chatgpt

## Intro

Finance and operations staff on GP are already using ChatGPT informally. They copy a SmartList result into a chat to get a variance explanation, or ask it to write a Report Writer formula. That works for one-off help but it leaks data outside any policy and gives the model stale, partial information.

Doing it properly means deciding what ChatGPT is allowed to see and how it gets it. Unlike a cloud ERP, Dynamics GP does not ship an internet-facing REST API. Your data sits in SQL Server behind a firewall, so any live access means you build and secure a small service that ChatGPT can call.

This page covers the realistic options, what the OpenAI side can and cannot do, and the cases where you should keep the model off the public cloud altogether.

## Use Cases

### Explaining a GL variance

**Description:** Export a trial balance comparison and ask for the drivers of a period-over-period change in a named account, with the model citing which journal entries moved.

### Writing SQL against GP tables

**Description:** Developers ask ChatGPT to draft a SELECT joining the PO header and line tables, then test it on a non-production copy. Never paste connection strings.

### Drafting vendor and customer letters

**Description:** Given a short summary of an overdue account or a price change, the model drafts a polite letter that staff review before sending.

### Decoding a custom GP report

**Description:** Paste the field layout or calculated fields from a Report Writer or SmartList Builder report and ask what it actually computes, helpful when the author left the company.

### Migration planning to Business Central

**Description:** Upload a chart of accounts, customer class and item class lists and ask for a mapping draft and a list of GP customisations to review.

## System Surface

- No built-in public API: GP data is reached through SQL Server, eConnect, or a web service layer that depends on your version. Anything ChatGPT calls must be a service you host.
- SmartList and SmartList Builder exports to Excel or CSV, the simplest way to hand data to ChatGPT for analysis without any integration.
- Reporting views in SQL Server and Excel reports published from GP, useful as a stable, curated source for exports or a small API.
- Report Writer and Modifier customisations, which are worth documenting with a model but contain business logic you may not want to share externally.
- Optional: a thin REST wrapper you build over read-only SQL views, secured by an API gateway, which becomes the action endpoint for a custom GPT.

## How It Works

### Classify the data

**Detail:** Decide which GP data is public-safe, internal or restricted. Customer pricing, bank details, payroll and export-controlled part data usually belong in the restricted group.

### Start with governed exports

**Detail:** Create SmartList or SQL view exports with only the needed columns and no personal identifiers. Use an enterprise ChatGPT plan so data is not used for training by default, and check your contract.

### Build a small read-only API if you need live data

**Detail:** Host a service in your DMZ with a handful of endpoints such as open invoices by vendor. It queries SQL views, returns JSON and enforces row limits.

### Describe the API to a custom GPT

**Detail:** Write an OpenAPI schema for those endpoints and register them as actions. Use OAuth or an API key tied to a service identity and keep scopes narrow.

### Test on a copy of the company database

**Detail:** Restore a backup to a test SQL Server and point the API at it. Verify answers against GP windows before any production use.

### Decide on private versus public

**Detail:** If any restricted data is involved, replace the hosted model with an open-weight model on your own hardware using the same API.

## Example Flows

### Controller asks about aged receivables

**Trigger:** Controller asks the custom GPT for customers over 60 days with rising balances.

**Steps:**

- GPT calls the aging endpoint with a company ID and bucket.
- API reads a view over the receivables tables and returns a limited JSON list.
- GPT ranks and summarises, citing customer IDs.

**Result:** A ready collection call list, no export file passed around by email.

### Spreadsheet analysis of inventory

**Trigger:** Planner uploads an inventory by site CSV.

**Steps:**

- File is produced from a reporting view with item numbers and quantities only.
- ChatGPT identifies slow movers and items below reorder points.
- Planner reviews and records decisions back in GP manually.

**Result:** Faster planning review with no live connection or write access.

## On Prem

If the data must not leave your network, ChatGPT as a hosted product is not the right tool. The same pattern works with an open-weight model served locally: keep the API wrapper, swap the hosted model for one running under vLLM or Ollama on a GPU server in your data centre, and give users a chat front end on the intranet. You lose the polished ChatGPT interface and the strongest frontier models, but you keep control of every prompt. Many GP customers run both: ChatGPT for generic writing and SQL help with no company data, and a private model for anything touching live ledgers.

## Security

- Never expose SQL Server directly to the internet or to a hosted model. Use an authenticated gateway in front of a purpose-built service.
- Strip personal and banking fields from views used for AI access.
- Use an enterprise or business plan with training disabled and confirm retention terms in writing.
- Add rate limits and result caps to stop bulk extraction through repeated prompts.
- Publish an acceptable use policy so staff know what may be pasted into ChatGPT.

## Options

### Governed exports into ChatGPT Enterprise

**Fit:** Fine for analysis of non-sensitive data and writing help.

**Effort:** Days, mostly policy and export design.

### Custom GPT with actions over your own API

**Fit:** When users want live answers and you can host a secure endpoint.

**Effort:** One to three weeks for a small read-only API.

### Private model with Netray or in-house

**Fit:** Restricted, export-controlled or customer-confidential data.

**Effort:** Pilot in a few weeks, hardware dependent.

## Pitfalls

- Assuming ChatGPT can connect to GP out of the box. There is no ready connector for on-premises GP.
- Giving the API write endpoints early. Keep it read-only until the answers are trusted.
- Trusting model arithmetic on large ledgers. Have the API return totals computed in SQL.
- Forgetting multi-company: the GPT must pass the company database explicitly or it will mix entities.
- Pasting custom Dexterity or VBA code containing credentials into a chat.

## Faq

### Is there an official ChatGPT connector for Dynamics GP?

**Answer:** Not that we can point to as a standard product. GP is mostly on-premises with no public API, so you typically build a small service that exposes curated read-only data, then connect it to a custom GPT or another front end.

### Is it safe to paste GP data into ChatGPT?

**Answer:** It depends on your plan and policy. Enterprise and business tiers generally exclude data from model training by default, but you should verify the terms and keep restricted data such as bank details, payroll or controlled technical data out of it.

### Can ChatGPT post entries into GP?

**Answer:** It can propose them, but posting should go through eConnect or Integration Manager with a human approval step. Letting a hosted model write directly to your ledgers is a control risk most auditors would question.

### What if our auditors forbid cloud AI?

**Answer:** Use an open-weight model on your own servers behind the same API. It is less capable than the newest hosted models for some tasks, but it keeps prompts and ledger data entirely inside your environment.

### Does Microsoft Copilot replace this?

**Answer:** Copilot features are focused on Microsoft 365 and Dynamics 365 products. For on-premises GP you would still need your own data access layer, so Copilot is not a drop-in replacement for a GP integration.

## Search terms

dynamics gp chatgpt, chatgpt for dynamics gp, connect chatgpt to dynamics gp, dynamics gp custom gpt, chatgpt great plains, dynamics gp openai integration, use chatgpt with microsoft dynamics gp data, dynamics gp sql chatgpt, private chatgpt alternative dynamics gp, dynamics gp ai assistant

---
Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json