# Global Shop Solutions Private AI Layer Blueprint Generator

> Source: https://netray.co/tools/global-shop-solutions-private-ai-layer-blueprint-generator
> Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems.
> Section: Free Tools

## Short answer

Fill in your version, database, hosting and use cases to get an architecture plan for a private, on-prem AI assistant on Global Shop Solutions, ready to share with IT and security.

## Category

on-prem-ai

## Erp

Global Shop Solutions

## Audience

- CIO
- IT manager
- ERP administrator
- Plant manager

## Tool

**Kind:** template

**Output:** document

**Fields:**

#### Company name

**Key:** company

**Type:** text

**Default:** Ridgeline Machining Inc

#### Global Shop version

**Key:** version

**Type:** select

**Options:**

- Older release (pre-2018)
- Mid-generation release
- Current release
- Not sure

**Default:** Not sure

#### Database engine

**Key:** db

**Type:** select

**Options:**

- Actian Zen / Pervasive PSQL
- Microsoft SQL Server
- Not sure, to be confirmed

**Default:** Actian Zen / Pervasive PSQL

#### Hosting model

**Key:** hosting

**Type:** select

**Options:**

- On-prem server
- Private cloud
- Air-gapped

**Default:** On-prem server

#### Named users

**Key:** users

**Type:** number

**Default:** 90

#### Sites or plants

**Key:** sites

**Type:** number

**Default:** 1

#### Main use cases

**Key:** usecases

**Type:** textarea

**Default:** Open job status, work center load, material shortages, supplier delivery performance, traveler and certificate lookup.

#### Data that must never leave the site

**Key:** never

**Type:** textarea

**Default:** Customer pricing, job cost and margin, customer drawings under export control.

#### GPU available

**Key:** gpu

**Type:** select

**Options:**

- None yet
- One 24 GB GPU
- One 48 GB GPU
- Two or more 48 GB GPUs
- 80 GB class GPUs

**Default:** None yet

#### Target go-live

**Key:** golive

**Type:** text

**Default:** Q1 2027 for read-only phase

**Blocks:**

#### Private AI Layer Blueprint: Global Shop Solutions

**Body:** Prepared for: {{company}}

Scope: a private AI assistant for {{version}} running on {{db}}, hosted as: {{hosting}}. Named users: {{users}}. Sites or plants: {{sites}}. Target go-live: {{golive}}.

Principle: the assistant reads from a copy of the system, runs open-weight models on infrastructure the company controls, and changes data only through supported interfaces after a named person approves.

#### 1. Context and use cases

**Body:** Use cases in scope:
{{usecases}}

Data that must never leave the site:
{{never}}

Available GPU: {{gpu}}.

#### 2. Components

**Table:**

**Columns:**

- Component
- Design for Global Shop Solutions
- Notes

**Rows:**

**0:** Data access path

**1:** ODBC to a read-only reporting copy; vendor-supported import or API for approved writes

**2:** Read-only account; no direct writes to application tables

**0:** Retrieval index

**1:** Local vector store over documents and long-text fields, tagged with customer, job and document class

**2:** Embedding model runs locally

**0:** Model serving

**1:** Open-weight model on vLLM or Ollama, GPU: {{gpu}}

**2:** Sized for {{users}} named users; see sizing calculator

**0:** Tool server

**1:** Approved parameterised read actions exposed through MCP or similar

**2:** Every call logged; limits on rows returned

**0:** Identity and permissions

**1:** Global Shop user mapped to role-based views through your identity provider

**2:** Scope applied to queries and retrieval filters

**0:** Audit

**1:** Append-only log of user, prompt, tool calls, rows returned and answer

**2:** Retained on site per company policy

#### 3. Security controls for {{company}}

**Body:** Hosting: {{hosting}}. All inference, embeddings and indexes stay inside this boundary.

Data that must not leave: {{never}} This data is excluded from any external service, and outbound network access from the AI components is blocked or limited to an allow list.

Access: users sign in through the company identity provider; the assistant enforces the same data scope as the system for each user. Database credentials are held in a vault and the account is read-only.

Write-back: disabled in phases 1 and 2; in phase 3 only for named actions, through supported interfaces, with human approval showing the exact change.

#### 4. Rollout phases

**Table:**

**Columns:**

- Phase
- Scope
- Exit criteria

**Rows:**

**0:** 1: Read-only questions

**1:** Status and lookup questions over the use cases above, with the query shown to the user

**2:** Accuracy measured on a test set agreed with power users; scope tests pass

**0:** 2: Documents and drafts

**1:** Document Q and A and drafted messages or transactions that a user enters manually

**2:** Citations verified; no cross-scope leakage in tests

**0:** 3: Approved actions

**1:** A short list of write actions through supported interfaces with approval

**2:** Each action signed off by the business owner and IT

#### 5. Open questions for IT

**Body:** 1. Which database edition, version and replication options are licensed for {{db}}?
2. Who owns the read-only copy and its refresh schedule?
3. Which identity provider and groups map to roles for {{users}} users across {{sites}} site(s)?
4. Where will the GPU host sit, and how are model files approved and updated?
5. What is the retention period for audit logs?
6. Which customer or regulatory commitments restrict where data may be processed?

**Signatures:**

- Prepared by
- IT / Security review
- Business owner

**Notes:** This is a planning document, not a security certification or legal advice. Table and API details differ by release, so verify each data access detail against your own system before building.

## Content

**Intro:** This generator produces a structured architecture plan for adding a private AI assistant to Global Shop Solutions without moving data to a public cloud. It captures your version, database, hosting model, users and sensitive data, and returns a document you can send to IT and security. Export it as PDF or Word.

**Sections:**

#### How to use the generator

**Body:** Fill in the company, version, database, hosting model, user and site counts, and the use cases you want first. List the data that must never leave the site as specifically as you can, because the security section is built from it. Then export the document and use it as the agenda for a review with IT, security and the business owner.

**Bullets:**

- Start with two or three use cases, not ten
- Be specific about restricted data classes
- Pick the hosting model your policy allows
- Share the open questions with IT before the meeting

#### What the blueprint contains

**Body:** The plan sets out the context, a component table (data access path, retrieval index, model serving, tool server, identity and audit), the security controls mapped to your inputs, a three-phase rollout and a list of open questions. The component choices follow a common pattern: read from a copy, retrieve documents locally, serve an open-weight model on your GPUs, and log everything.

**Bullets:**

- Data access path: ODBC to a read-only reporting copy; vendor-supported import or API for approved writes
- Identity: Global Shop user mapped to role-based views through your identity provider
- Phases: read-only, drafts, approved actions

#### Reading the output critically

**Body:** Treat the blueprint as a starting hypothesis. Table names, interfaces and replication options vary by release, so have your administrator confirm each data access item. Check licensing for any database replication feature you plan to rely on, and decide who will own the model host, the index and the audit log after go-live.

#### Common mistakes

**Body:** Teams often start by pointing the model at the production database, letting write-back arrive in the first release, or skipping role tests. Another frequent error is ignoring documents: much of the value is in drawings, certificates and procedures that never sat in the database.

**Bullets:**

- Production database used for AI queries
- No role or scope tests before go-live
- Write-back enabled too early
- No owner for the audit log

**Cta:** Netray builds private AI layers for job shops running Global Shop Solutions on-prem. If you want this blueprint checked against your release, database and modules, ask for an expert review.

**Faq:**

#### Is this blueprint specific to Global Shop Solutions?

**Answer:** The data access path, identity mapping and component notes are tailored to Global Shop Solutions, but you should still verify table names, interfaces and licensing against your own release before building.

#### Does the plan require a public cloud?

**Answer:** No. The design keeps inference, embeddings, indexes and logs inside your boundary. You can choose on-prem server, private cloud or air-gapped hosting in the form.

#### Can I give the output to my security team?

**Answer:** Yes. It is formatted for sharing and lists controls and open questions. It is a planning document, not a certification, so your security team will still apply its own review.

#### Do I need a GPU before I start?

**Answer:** No. The form accepts none yet. The sizing calculator estimates GPU count from your usage, and a pilot can start on a single 24 GB or 48 GB GPU.

## Search terms

Global Shop Solutions AI architecture, Global Shop Solutions on-prem AI plan, private AI architecture template, on-prem LLM architecture document, AI integration plan Global Shop, air-gapped AI blueprint, ERP AI security plan, AI rollout plan template

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