How to use ChatGPT with ECI E2 Shop System
Short answer
You can use ChatGPT with E2 Shop System in three ways: paste or upload exported data, build a custom GPT whose actions call a small API you host in front of E2, or skip it for regulated data and use a private model. The right choice depends on how sensitive the jobs and pricing are.
Many E2 users already paste a quote, a vendor email or an Excel export into ChatGPT to get a summary or a draft. That works, and for a shop with non-sensitive work it may be all you need. The limits show up when you want current answers about live jobs, or when customers' drawings and pricing are covered by contractual restrictions.
ChatGPT does not connect to E2 by itself. E2 is typically a Windows application on a local SQL Server, with no public cloud endpoint. Anything that gives ChatGPT live access therefore needs a bridge that you build and secure.
This page walks through the realistic options, from the lowest effort to the most controlled, and is honest about where each stops being appropriate.
What teams use it for
Summarise an exported open-jobs report
Export open jobs to Excel, upload it, and ask which are at risk and why. Quick, but the answer is only as fresh as the export.
Draft customer and vendor emails
Provide the job or PO details you choose to share and have ChatGPT draft delay notices, quote follow-ups or supplier chases in your tone.
Quote narrative and terms
Turn estimator notes into a clean quote scope, assumptions and exclusions, then paste it into E2 or your quote document.
Custom GPT for job status
A custom GPT with an action calling your own read-only endpoint can answer where job 12345 stands, if your security policy allows it.
Explain unfamiliar customer specs
Paste non-controlled specification text and ask for a plain-language list of process, inspection and certification requirements.
ECI E2 Shop System integration surface
- SQL Server: the usual source for any live feed. A read-only login and curated views are the safest way to build the endpoint a GPT action would call.
- Excel and report exports: manual but zero-integration, and appropriate for non-sensitive analysis.
- Your own HTTPS endpoint: a custom GPT action needs a publicly reachable HTTPS API with OpenAPI description and authentication, so E2 itself is never exposed; only a thin service you control is.
- Document stores: drawings, certs and quote files on file shares are not in the database and must be handled separately if you want them in scope.
- Version-specific web services: if your E2 version and modules offer web services, check their terms before using them as the data source.
How it works
- 1
Classify the data first
List what is in scope: customer names, pricing, drawings, export-controlled or CUI content. Decide what may never go to a third-party service.
- 2
Choose the plan and controls
Business or enterprise ChatGPT plans offer data-handling terms and admin controls that consumer accounts lack. Read the current terms rather than assuming.
- 3
Start with exports
Use filtered, de-identified exports for analysis and drafting. Strip customer identifiers where you can.
- 4
If live access is needed, build a thin API
Create a small read-only service over curated SQL views, with API key or OAuth, rate limits and an OpenAPI spec.
- 5
Create the custom GPT action
Point the action at that endpoint and write instructions so the GPT answers only from tool results and says when it does not know.
- 6
Test with real questions
Compare answers to E2 screens for a few weeks before anyone relies on them.
Example flows
Weekly at-risk summary from an export
Trigger: Scheduler exports open jobs every Monday.
- Remove customer names and pricing columns from the export.
- Upload the file to ChatGPT with a standing prompt.
- Ask for jobs with due dates inside five days and no completed operations beyond the first.
- Copy the resulting list into the weekly production meeting notes.
Result: A fast second opinion on risk, with no live connection and no sensitive columns shared.
Custom GPT job lookup
Trigger: A salesperson asks about a customer's order status.
- GPT action calls your endpoint with the job or PO number.
- Endpoint queries a read-only view and returns status, next operation and due date.
- GPT replies in plain language and cites the fields returned.
Result: Fewer interruptions to the office staff for routine status questions.
Keeping data on-prem
ChatGPT itself is a hosted service, so it cannot run inside an air-gapped E2 network. If your work carries ITAR, CUI or customer NDAs that forbid third-party processing, the compliant pattern is a private open-weight model served locally with vLLM or Ollama, reading the same curated SQL views. You keep the familiar chat experience but the data, the model and the logs stay on your premises. Many shops use a hybrid: ChatGPT for non-sensitive drafting and a private model for anything touching jobs, pricing or drawings. Your compliance owner, not the AI vendor, decides where that line sits.
Security checklist
- Never paste controlled technical data or CUI into a consumer account.
- Use business or enterprise plans with documented data-use terms if you adopt ChatGPT at all.
- Expose only a read-only, authenticated, rate-limited endpoint for any custom action, never E2 or SQL Server directly.
- Limit columns: leave out cost, margin and customer pricing unless the asker's role needs them.
- Keep a record of who uses it and what categories of data are permitted.
Ways to get it done
Manual exports and paste-in prompts
Effort: Minutes to start; labor repeats every time.
Good for non-sensitive drafting and one-off analysis in a small office.
Custom GPT with an action you host
Effort: Days to a couple of weeks, plus hosting and security review.
Suits teams comfortable running a small secured web service and whose policy allows hosted AI.
Private model with Netray or in-house build
Effort: A scoped pilot, then expand.
For aerospace, defense and NDA-bound work where data cannot go to a hosted model.
Pitfalls
- !Uploading exports with customer drawings references or pricing without checking contractual limits.
- !Trusting a GPT's confident answer when it has no tool result behind it; require citations to fields.
- !Exposing SQL Server to the internet to make an action work; always put a service in front.
- !Using a personal account for company data because it is convenient.
- !Forgetting that exported data goes stale the moment E2 changes.
Frequently asked questions
Can ChatGPT connect directly to E2 Shop System?
Not out of the box. E2 usually runs on a local SQL Server, so you need a bridge such as a small authenticated API in front of read-only views. Direct exposure of the database to the internet is not recommended.
Is it safe to upload E2 data to ChatGPT?
It depends on the data and your plan. Non-sensitive operational data on a business plan may be acceptable. Customer-controlled or export-controlled data generally should not leave your environment. Check contracts and plan terms.
What can I do today without any integration?
Export a report to Excel, remove sensitive columns, and use ChatGPT to summarise, find outliers, or draft emails and quote text. It is crude but useful and costs almost nothing to try.
Does a custom GPT keep answers current?
Only if its action calls a live endpoint. A GPT built on uploaded files is a snapshot. Live lookups require the bridge service described above.
When is a private model better than ChatGPT?
When data must remain on your network, when contracts forbid third-party processing, or when you need full audit control. The trade-off is that you operate the hardware and model yourself or with a partner.
Want ChatGPT working on your ECI E2 Shop System?
A free 30 minute assessment: we map the integration path for your version and setup, and where it should run.
Get a free assessment