# How to use ChatGPT with SAP Business One

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

## Short answer

There are three ways: paste or upload exported data, give a custom GPT or ChatGPT app limited access to a small API in front of the Service Layer, or skip the public service and run a private model. The first is quick but manual, the second needs a hardened public endpoint, and the third suits regulated or price-sensitive data.

## System

SAP Business One

## Capability

chatgpt

## Intro

Many Business One users already use ChatGPT on the side. Someone exports an item list from a query, pastes it into a chat, and asks for a summary or a cleaned-up description. It works, and for non-sensitive data it is perfectly reasonable.

The next step is wanting ChatGPT to answer questions about live data: what is the status of order 4521, which customers are over their credit limit. That is where the real design decisions begin, because the model is hosted outside your network while your Business One server probably is not exposed to the internet.

This page lays out the realistic options, what a custom GPT action needs from your side, what data controls to check, and when it is cleaner to keep the model in-house.

## Use Cases

### Clean up item descriptions in bulk

**Description:** Export the item master (OITM) to a spreadsheet, have ChatGPT normalise names and draft purchasing descriptions, then review and import the changes through the Data Import/Export tool or Service Layer.

### Explain a query result

**Description:** Paste the output of a Query Manager report and ask for the three things that changed against last month, with no live connection at all.

### Order status for sales staff

**Description:** A custom GPT with one action that calls your own small API, which in turn reads order and delivery status from the Service Layer.

### Draft customer emails

**Description:** Give ChatGPT a delivery note and open order lines, and have it draft a clear delay notice that the salesperson edits and sends.

### Write SQL and formatted search help

**Description:** Ask ChatGPT to draft a Query Manager query against the B1 schema, then test it on a copy company before saving it.

### Summarise supplier documents

**Description:** Upload a vendor contract or certificate, ask for key dates and obligations, then record the result in a user-defined field after review.

## System Surface

Exports: Query Manager, Crystal Reports and the Data Import/Export utilities produce Excel or CSV that can be uploaded to ChatGPT with no integration.

Service Layer: the REST/OData interface that a small wrapper API can call. Custom GPT actions need an HTTPS endpoint with an OpenAPI description, so the wrapper, not the Service Layer itself, should face the internet.

SQLQueries and views via the Service Layer: let the wrapper expose only a few fixed, parameterised read shapes. Availability varies with version.

Database read replica or reporting copy: a safer source than the production database when the wrapper needs heavier queries.

Authentication: the Service Layer uses a login with company database, user and password and a session cookie, so the wrapper has to manage sessions and map GPT callers to a restricted B1 user.

## How It Works

### Decide what level of access you need

**Detail:** Manual exports solve many cases. Only build a live connection when the same question is asked daily and the data is not highly sensitive.

### Check your ChatGPT plan's data controls

**Detail:** Review whether the plan you use trains on inputs, what retention applies and who administers the workspace. Business and enterprise offerings generally differ from personal accounts, so read the current terms.

### Build a thin wrapper API

**Detail:** Create a small service with a few endpoints such as order status or item availability. It logs in to the Service Layer with a restricted user and returns only a minimal field set.

### Describe it with OpenAPI

**Detail:** Write an OpenAPI spec for those endpoints and register it as a custom GPT action. Keep descriptions precise so the model chooses the right call.

### Secure the endpoint

**Detail:** Use an API key or OAuth, rate limits, IP restrictions where possible, and a gateway in a DMZ. Never open the Service Layer port directly.

### Pilot with a small group

**Detail:** Give five users access for two weeks, review the logs for odd requests, and refine tool descriptions.

## Example Flows

### Order status from a custom GPT

**Trigger:** A salesperson asks, "Where is order 4521?"

**Steps:**

- The GPT calls the wrapper's order-status action with the document number.
- The wrapper queries the sales order and linked delivery and invoice through the Service Layer.
- It returns status, promised date and carrier reference only.
- The GPT replies in plain language.

**Result:** The salesperson gets an answer without opening B1 or calling the warehouse.

### Weekly export review

**Trigger:** A finance user exports aged receivables to Excel.

**Steps:**

- The file is uploaded to a ChatGPT workspace with data controls approved by IT.
- The user asks for accounts that moved into a worse ageing bucket.
- ChatGPT produces a summary and draft reminder wording.
- The user checks the figures against B1 before sending anything.

**Result:** A manual review that took an hour is done in minutes, with no system integration.

## On Prem

If your policy forbids sending B1 data to a third-party model, keep the same wrapper-and-tools design but point it at an open-weight model hosted on your own server. The wrapper and Service Layer stay inside your network, and users reach the assistant through an internal web page or chat client. You lose some model capability compared with the newest hosted models, but for status questions, summarisation and drafting a mid-sized open model is usually adequate. This is also the cleaner answer for customers with contractual limits on where pricing, drawings or part numbers may be processed.

## Security

- Do not expose the Service Layer to the internet; put a hardened wrapper in a DMZ.
- Return the minimum fields needed and strip cost, margin and personal data unless required.
- Use a restricted B1 user for the wrapper with read-only authorisations.
- Check your plan's training and retention settings before uploading any exports.
- Log actions and review them for unexpected queries.

## Options

### Manual exports into ChatGPT

**Fit:** Best for ad hoc summaries and drafting on non-sensitive data.

**Effort:** None, but depends on discipline.

### Custom GPT with a wrapper API

**Fit:** Good for a handful of repeated lookups and a small user group.

**Effort:** One to three weeks including security review.

### Private model with a partner such as Netray

**Fit:** Appropriate when data may not leave your network or you need many tools and write-back with approval.

**Effort:** Scoped pilot, then expansion.

## Pitfalls

- Uploading customer price lists or drawings to a personal account with default data settings.
- Exposing the Service Layer directly to make a GPT action work.
- Trusting ChatGPT's SQL against the B1 schema without testing on a copy company.
- Skipping user-defined fields so answers miss customer-specific data.
- Letting a GPT action return full documents when a few fields would do.

## Faq

### Can ChatGPT connect directly to SAP Business One?

**Answer:** Not out of the box. A custom GPT can call an HTTPS API you provide, so you would build a small wrapper in front of the Service Layer. Pointing ChatGPT straight at the Service Layer is not advisable because of authentication, session handling and exposure concerns.

### Is it safe to paste B1 exports into ChatGPT?

**Answer:** It depends on the data and your plan. Review the provider's current terms on training and retention for your account type, and keep sensitive items such as cost, margin and personal data out unless your policy allows it. Many firms set rules for what may be uploaded.

### Can ChatGPT create sales orders in B1?

**Answer:** It could if your wrapper exposes a write endpoint, but that is risky. A safer design creates a draft document that a user reviews and approves in B1, so approval procedures and stock effects remain under human control.

### Why would I choose a private model instead?

**Answer:** Typical reasons are contractual restrictions, defense or aerospace data, pricing sensitivity, or simply wanting predictable costs. A private open-weight model keeps all queries inside your network, at the price of managing hardware and accepting some capability gap.

### Do I need to be on HANA for this?

**Answer:** No. The wrapper talks to the Service Layer, which works across supported database editions. HANA mainly matters if you want heavy analytic queries directly against the database. Confirm Service Layer availability for your version and database.

## Search terms

sap business one chatgpt, chatgpt for sap business one, sap b1 chatgpt integration, connect chatgpt to sap business one, sap business one custom gpt, sap business one openai, chatgpt sap b1 service layer, use chatgpt with sap business one data, sap business one gpt actions, private chatgpt alternative sap business one

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