# How to use ChatGPT with Microsoft Dynamics NAV

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

## Short answer

There are two realistic routes: export NAV data and upload it to ChatGPT for one-off analysis, or give a custom GPT or app an action that calls a NAV OData endpoint. Both need an internet-reachable, authenticated endpoint, which is awkward for on-prem NAV. For regulated or export-controlled data, a private model is usually the safer choice.

## System

Microsoft Dynamics NAV

## Capability

chatgpt

## Intro

Teams on Dynamics NAV (often still called Navision) frequently ask whether ChatGPT can just answer questions about their items, orders and costs. Out of the box it cannot see your data. ChatGPT only knows what you paste, upload or connect through an action.

That shapes the decision. The quick route is exporting a report to Excel and asking ChatGPT to analyze it. The structured route is a custom GPT with an action that calls a small API in front of NAV. The third route is not using ChatGPT at all for the sensitive parts and running an open model on your own servers.

This page walks through the first two honestly, including where on-prem NAV makes them hard, and when you should skip them.

## Use Cases

### Explain a variance in an exported cost report

**Description:** Export the Inventory Valuation or Value Entry data for one item family to Excel, upload it, and ask ChatGPT where the cost per unit moved. Works well for a single analyst with non-sensitive data.

### Draft vendor emails from open PO lines

**Description:** Paste a filtered list of late purchase lines and have ChatGPT draft polite, consistent reminders. The buyer reviews and sends from Outlook.

### Translate and summarize customer order notes

**Description:** Sales order comments in several languages are summarized into one status line for the account manager.

### Custom GPT for item lookup

**Description:** A custom GPT with an action calls a read-only endpoint for item availability and standard cost, returning formatted answers to planners.

### Help writing C/AL or AL snippets

**Description:** Developers ask ChatGPT to explain a legacy C/AL function or draft a codeunit skeleton. No live NAV data needed, and the output is reviewed before use.

## System Surface

- Excel and CSV exports: most list pages and reports can export, and RDLC reports can render to Excel or PDF. This is the zero-integration route.
- OData web services: published pages return feeds that a small API gateway can reshape into JSON for a GPT action. Version support for the OData flavor varies.
- SOAP web services: usable but awkward for GPT actions, which expect OpenAPI-described REST endpoints. Put a thin REST wrapper in front.
- Direct SQL: possible behind your own API, but ChatGPT should never get a database login. Expose parameterized views only.
- Authentication: NAV supports Windows, user name and password, and Azure AD in later versions; the wrapper translates that to an API key or OAuth for the GPT.

## How It Works

### Decide what data may leave

**Detail:** Classify the fields involved. Pricing, customer names and drawings may be fine or restricted depending on your contracts. Settle this before building anything.

### Start with exports

**Detail:** Define two or three standard exports from NAV with pre-filtered columns. Teach staff to upload those instead of full ledger dumps.

### Wrap NAV in a read-only API

**Detail:** Build a small service that calls published OData or SOAP pages and returns clean JSON with only needed fields. Add an OpenAPI description.

### Publish through a custom GPT or app

**Detail:** Create the GPT with an action pointing at your API, with authentication and instructions that tell it to quote the record numbers it used.

### Reach the API from the internet

**Detail:** On-prem NAV means you need a reverse proxy or API gateway in a DMZ with rate limits and IP filtering. Evaluate whether this exposure is acceptable.

### Review usage and tighten

**Detail:** Log requests, check what users actually ask, and trim fields. Move sensitive questions to a private model.

## Example Flows

### Planner asks about stock for a product family

**Trigger:** A planner asks the custom GPT whether item 10450 has enough stock for next week.

**Steps:**

- The GPT calls the availability action with the item number.
- Your API queries the published Item page and open supply lines.
- JSON returns on-hand, reserved and expected receipts.
- The GPT replies with a short answer and the numbers it used.

**Result:** A fast read-only answer without opening three NAV windows.

### Analyst uploads a variance export

**Trigger:** Finance exports Value Entries for a cost center to Excel after a close.

**Steps:**

- Analyst removes customer and vendor names from the file.
- Uploads it to an enterprise ChatGPT workspace.
- Asks for the top five drivers of cost change versus last month.
- Verifies the answer against NAV before reporting.

**Result:** A first-draft variance narrative in minutes, still checked by a person.

## On Prem

ChatGPT is a hosted service, so on-prem NAV has to reach out or be reached. If policy forbids exposing an endpoint, stay with sanitized exports or use a private model served inside your network. A common compromise is a private open-weight model for anything touching production data and ChatGPT only for generic writing and developer help with no company data. If you do build an action, place the API in a DMZ, allow only the necessary egress and ingress, and keep NAV itself unreachable from the internet.

## Security

- Use ChatGPT Enterprise or an equivalent plan with data controls if any company data is involved, and confirm the training and retention terms yourself.
- Never give the GPT a database credential or a NAV super user; expose a read-only wrapper API.
- Strip or hash fields subject to ITAR, export control or customer NDAs before they leave your network.
- Rate limit and log the API so unusual volume is visible.

## Options

### Manual exports into ChatGPT Enterprise

**Fit:** Small teams with non-sensitive analysis needs.

**Effort:** Days to define standard exports and a usage policy.

### Custom GPT with an action over a wrapper API

**Fit:** NAV owners who can host a secure API and accept hosted-model data flow.

**Effort:** About 3 to 6 weeks including security review.

### Private model with Netray or in-house

**Fit:** Regulated, defense or confidential data that cannot leave your network.

**Effort:** Several weeks for a pilot; ongoing model operations.

## Pitfalls

- Uploading full ledger dumps leaks far more than the question needs.
- ChatGPT can do arithmetic wrong on big tables. Make the API do the sums.
- Exposing an on-prem NAV endpoint without a gateway is a common and serious mistake.
- Staff adopt personal ChatGPT accounts if you offer nothing sanctioned. Provide an approved path.
- Assuming answers are right because they sound confident. Require record numbers in every reply.

## Faq

### Can ChatGPT connect directly to Dynamics NAV?

**Answer:** Not natively. You can upload exports or build a custom GPT action that calls an API you host in front of NAV's web services. Direct database access is not advisable. On-prem NAV needs a secured gateway for the action to reach it.

### Does ChatGPT train on our NAV data?

**Answer:** That depends on the plan and settings. Enterprise and business plans generally state that data is not used for training by default, but you should read the current terms yourself and involve legal before sending customer or export-controlled data.

### Is there a ready-made ChatGPT connector for NAV?

**Answer:** I am not aware of an official one, and third-party offerings change often. Check Microsoft AppSource and your NAV partner. Most teams end up with a small custom wrapper API.

### When should we not use ChatGPT with NAV?

**Answer:** When data is regulated, contractually restricted or simply sensitive pricing, and when you cannot safely expose an endpoint. A model running on your own servers avoids those concerns at the cost of operating it.

### Can it write data back to NAV?

**Answer:** Technically yes through a codeunit action, but start read-only. Any write should be a proposal approved by a person and applied through NAV validation, not a direct table update.

## Search terms

dynamics nav chatgpt, chatgpt for dynamics nav, chatgpt navision, dynamics nav custom gpt, connect chatgpt to dynamics nav, dynamics nav openai integration, dynamics nav chatgpt api, navision chatgpt integration

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