Manufacturing SystemsERP Platform

Windchill PLM + private AI

AI for PTC Windchill That Respects Your Access Rules and Keeps Product Data Inside

Short answer

AI for PTC Windchill connects a privately hosted language model to Windchill's OData REST domains and document vault so engineers, change analysts and quality staff can ask plain-English questions about parts, BOMs, change requests and CAD metadata and get answers with the source objects cited. It runs on-prem or in a private cloud, evaluates every request as the asking user, and keeps drawings and technical data inside your boundary.

ERP: PTC Windchill
Industries: Aerospace, Defense, Industrial Machinery, Medical Devices, Electronics
Written for: PLM Administrator

Discuss your PTC Windchill environment See use cases

Windchill is where the product definition lives: the WTPart structure, the EPMDocument records behind Creo models and drawings, the change objects that move a design from released to revised, and the lifecycle states that decide who may see what. For most manufacturers it is the system of record for engineering, and the ERP receives a downstream copy of the released structure.

That position makes it both the most valuable and the most sensitive dataset in the company. A question like which open change notices touch this assembly and which of them affect a part we are already buying is answerable, but only by someone who knows how to navigate part usage, affected objects and resulting objects, then cross-check the ERP. Most users file a ticket or ask the one person who knows.

Public AI assistants are a poor fit for this data. Drawings, CAD attributes and BOMs often carry export-controlled technical data or customer-restricted IP, and a general chatbot has no notion of Windchill contexts, domains or access control lists. Anything that bypasses them is a compliance finding, not a convenience.

This page describes what private AI on Windchill looks like in practice: which interfaces it uses, how it keeps Windchill's own security model in force, where it genuinely saves time for engineering and change teams, and where PTC's own tooling or a plain saved search is enough.

What usually gets in the way

The problems we hear most from plm administrator teams running PTC Windchill.

Change impact is assembled by hand

Finding every assembly affected by a change means walking where-used trees, affected objects and resulting objects across several screens, then repeating the exercise in the ERP to see on-hand stock and open purchase orders.

Search finds objects, not answers

Windchill search returns parts and documents matching keywords. It does not tell you which released revision of a bracket is used on active programs, or why a past change was rejected, unless someone reads the change notice attachments.

Change paperwork is repetitive

Problem reports, change requests and change notices need consistent justification text, impact statements and affected-item lists. Authors copy from previous changes and quality varies, which slows review boards.

Reuse is invisible

Designers create new parts because finding an existing similar one by attributes, geometry notes and supplier history is slow. Duplicate parts then ripple into ERP item masters and purchasing.

PLM and ERP structures disagree quietly

Released BOMs flow to the ERP through an integration, but unit-of-measure, effectivity and substitute handling differ. Mismatches surface months later as shortages or wrong costs.

Where AI earns its place in PTC Windchill

Each use case names the ERP objects it reads or writes, so your ERP team can judge the integration effort before anyone commits budget.

Change impact summaries

Given a change request or change notice, assemble affected objects, where-used parents, linked documents and a plain-language impact statement for the review board, with each claim linked to its source object.

Touches: WTPart, WTPartUsageLink, ChangeRequest2, ChangeNotice2, ChangeActivity2, affected and resulting objects

Outcome: Impact packages that previously took a few hours of navigation are drafted in minutes and reviewed by the analyst instead of built from scratch.

Natural-language BOM questions

Ask for the indented BOM of a released revision, where a component is used, or which assemblies reference a superseded part, without building a report.

Touches: WTPart versions, usage links, view (Design vs Manufacturing), effectivity

Outcome: Routine where-used and BOM-compare requests stop queuing behind the PLM administrator.

Draft change descriptions

Produce a first draft of reason for change, impact on form, fit and function, and disposition text from the selected affected items and prior similar changes. The author edits and submits.

Touches: ChangeRequest2, ChangeNotice2, problem reports, attached documents

Outcome: More consistent change records and fewer review-board bounce-backs for missing justification.

Similar part and reuse search

Find existing parts by description, attributes and past usage before a designer creates a new one, and show approved sources from the ERP item and vendor records.

Touches: WTPart attributes, classification, EPMDocument metadata, ERP item master

Outcome: Fewer near-duplicate parts entering the vault; savings depend on how much duplication exists today.

PLM to ERP BOM reconciliation

Compare the released Windchill structure to the manufacturing BOM in the ERP and list differences in quantity, unit of measure, revision and effectivity.

Touches: Windchill released BOM, ERP BOM tables and item master, integration logs

Outcome: Mismatches are caught at release rather than at the first shortage.

Document and spec retrieval

Answer questions from specifications, test reports and supplier documents stored as WTDocuments, citing the document, version and page.

Touches: WTDocument, content holders, document lifecycle states

Outcome: Engineers find the governing requirement in seconds instead of opening many PDFs.

Release readiness checks

Before a change moves to release, check for missing approvers, unreleased children, missing drawings and attribute gaps against your own rules.

Touches: Lifecycle state, workflow tasks, EPMDocument association, required attributes

Outcome: Fewer rejected releases and less rework late in the change process.

Reference architecture

A retrieval and agent layer sits beside Windchill, never inside it. It reads through supported interfaces using the identity of the asking user, assembles evidence, and has a private model write the answer with citations. Writes to Windchill are drafts that a human submits.

1
Access layer
Windchill OData REST domains such as product management, change management and document management, called over HTTPS with delegated user credentials or a service identity that is scoped and logged. Info*Engine tasks or read-only reporting views for gaps.
2
Security enforcement
Each query runs as the requesting user so Windchill contexts, domain policies and access control rules decide what is visible. The AI layer does not hold a broader view than the user.
3
Retrieval layer
Structured lookups for parts, BOMs and changes plus a vector index over documents, with object identifiers, versions and lifecycle state kept as metadata for citation and filtering.
4
Model layer
Open-weight models served with vLLM or Ollama on the customer's GPUs or a sovereign cloud, with no calls to public APIs.
5
Governance layer
Prompt, retrieved objects and answer logged per user; drafts only for changes; evaluation sets built from past changes and known BOM questions.

Integration notes for your ERP team

Windchill exposes OData REST under the Windchill servlet path with domains for product management, change management and documents; confirm the version-specific domain list on your release.
Use read-only service roles for indexing and per-user delegation for question answering so access control lists remain authoritative.
Index released versions with their lifecycle state and view (Design or Manufacturing) so answers state which revision they describe.
For ERP comparison, read the ERP BOM through its own API or SQL views rather than through the PLM-to-ERP connector, so the comparison is independent of the integration being checked.
CAD content stays in the vault; index metadata and neutral formats only, and do not send native Creo files to the model.
Heavy bulk exports can load Windchill's method servers; schedule indexing off-peak and use incremental sync by modification timestamp.

Deployment options

On-premises next to Windchill

Defense, aerospace and export-controlled programs with self-hosted Windchill

Model, vector store and agent services run on a GPU server in the same network zone as Windchill. No external egress; suitable for air-gapped enclaves.

Private cloud or VPC

Companies running Windchill in their own cloud tenancy or on a managed host

Models run on dedicated GPU instances in the customer's tenancy with private networking to the Windchill servers and customer-held keys.

Hybrid with Windchill+

Teams on PTC's SaaS offering that want model hosting under their own control

The AI service runs in the customer's environment and calls the SaaS APIs for only the objects a user is entitled to see. Confirm tenant API access and data-transfer terms with PTC first.

Compliance and data control

How the architecture supports your obligations. Certification and accountability stay with your organisation; the design keeps the evidence straightforward.

ITAR and EAR technical data

Drawings, models and BOMs often qualify as technical data. Keeping retrieval and inference inside a controlled boundary, with user-level access enforcement and logging, lets the customer meet its own obligations. Citizenship and deemed-export checks stay with your access program.

CMMC Level 2 and NIST SP 800-171

The AI components are in scope when they process CUI. Access control, audit logging and encryption in transit are implemented in the stack and documented for the customer's system security plan.

AS9100 configuration management

Answers that reference configuration baselines cite the released version and effectivity. Draft change text is reviewed and submitted by an authorized person, preserving the formal change record.

FDA 21 CFR Part 820 and ISO 13485 design controls

Design history and change records are read, never altered. Generated text is clearly marked as a draft until a signer approves it in Windchill.

21 CFR Part 11 electronic signatures

The AI layer never signs or approves. Signatures remain Windchill workflow actions by named users.

Where Netray fits

Custom build

Windchill data models, workflows and customizations vary widely, so the retrieval layer and change-drafting agents are built to your object types and rules.

DataRay

Useful when questions must span Windchill documents, supplier files on shares and ERP tables in one private chat.

ERPray

Covers the ERP side of PLM-to-ERP comparisons with query display and read-only access.

How an engagement runs

Phase 1 . 2 weeks

Discovery

-Inventory of Windchill version, customizations and interfaces
-Top 20 real questions from engineering and change teams
-Access and export-control review
-Success measures agreed

Phase 2 . 4 to 6 weeks

Pilot build

-OData connector with user delegation
-Part, BOM and change retrieval
-Citation-first answer interface
-Evaluation set from historical changes

Phase 3 . 4 weeks

Workflow agents

-Change impact summary draft
-PLM to ERP BOM compare
-Draft-only change text with human submit
-Audit log review

Phase 4 . 2 to 3 weeks

Hardening and handover

-Security test and access regression checks
-Runbook for upgrades
-Model update procedure
-Training for administrators

Questions to ask any vendor, including us

A short list that separates real PTC Windchill AI work from a chatbot demo.

Does every query run with the asking user's Windchill permissions, or with a broad service account?
Which Windchill interfaces are used, and are they supported on our release?
Where do prompts, retrieved objects and answers go, and how long are they kept?
Can you show answers with the source part, document and revision cited?
What does the system write to Windchill, and who approves it?
How are native CAD files and export-controlled documents excluded or handled?
What is the upgrade impact when we move to a new Windchill release?
Which tasks would PTC's own capabilities or saved reports handle just as well?

Frequently asked questions

Does AI for Windchill need access to native Creo files?

No. Most valuable questions are answered from part, BOM, change and document metadata plus text documents. Native CAD stays in the vault. If geometry-aware features are wanted later, they are scoped separately, with neutral formats and clear data-handling rules.

Can the AI see parts a user is not allowed to see?

It should not. Queries are executed on behalf of the asking user so Windchill contexts and access rules apply. Any design that indexes everything under one broad account and filters afterward is a risk and should be challenged in review.

Will it replace the change review board?

No. It drafts impact summaries and justification text and flags missing items. Reviewers still decide, and approvals and signatures remain Windchill workflow actions by named people.

Is Windchill+ supported?

Possibly, through the SaaS APIs, with the AI service in your own environment. API availability and data-transfer terms depend on your PTC agreement, so confirm them before design starts.

How does this differ from PTC's own AI features?

Vendor features are fine for tasks inside the product UI. A separate layer makes sense when you need your own model hosting, cross-system answers spanning ERP, quality and documents, or customer-specific rules for change drafting.

Can it write BOMs or release changes automatically?

By default no. It produces drafts and comparison reports. Direct writes to released engineering data are a poor first step, and any later write-back should use a staged workflow with human approval and full logging.

What hardware is required?

A single server with one or two data-center GPUs serves a pilot with a mid-size open-weight model for a few dozen concurrent users. Sizing depends on document volume, concurrency and model choice and is settled in discovery.

Related guides

Digital thread + AI

AI Across PLM and ERP: Teamcenter, Windchill, and Your ERP

AI that reads Teamcenter or Windchill alongside your ERP to trace engineering change impact, BOM sync gaps, and where-used questions, on-prem.

Agile PLM + ERP AI

AI across Oracle Agile PLM and your ERP: engineering changes without the swivel chair

Ground AI on Agile PLM engineering change orders and BOMs alongside ERP, so engineers and planners get accurate ECO impact answers without swivel-chair lookups.

Know what an ECO actually touches

AI for Engineering Change Impact on BOMs, Routings, and Open Orders

AI reads a proposed ECO against live BOM, routing, and open order data to show engineering exactly what a change touches before it gets released.

RAG + SQL + permissions

A Private LLM Grounded on Your ERP Data

How a private LLM answers questions on your ERP data: RAG plus text-to-SQL, role-based permissions inherited from the ERP, and where each fits.

ITAR + on-prem AI

ITAR-Compliant AI for ERP Technical Data

How to add generative AI to your ERP without creating a deemed export under ITAR. On-prem architecture patterns an Empowered Official can sign off on.

Agents + approval gates

AI Agents for ERP, Running On-Prem

A practical guide to on-prem AI agents for ERP: what they can safely automate, where human approval belongs, and how to design the guardrails.

Plan it with numbers

Free Tool

Talk it through with an engineer who knows PTC Windchill

Bring one real question your team cannot answer from the ERP today. We will map the data path, the model, and where it runs, and tell you honestly if AI is the wrong tool for it.

Start the conversation See ERPray

All AI for ERP guides