
AI for Engineering Knowledge Management
What an AI assistant can actually read inside Oracle Agile PLM, which objects the API exposes, and where Agile stops and the CAD vault begins.
·
⏱
8 min read

Dr. Maor Farid
Maor Farid is the Co-Founder and CEO of Leo AI, the first AI platform purpose-built for mechanical engineers. He holds a PhD in Mechanical Engineering and completed postdoctoral research at MIT as a Fulbright fellow. A Forbes 30 Under 30 honoree and former AI researcher and Mechanical Engineer in an elite military intelligence, Maor leads Leo AI's mission to transform how engineering teams design better products faster.

BOTTOM LINE
An AI assistant against Oracle Agile PLM is worth having for the record layer and will disappoint anyone expecting it to understand geometry. Agile exposes items, changes, bills of material, approvals, the approved manufacturer list, quality records and compliance declarations through a documented SOAP surface, and an assistant that reads them with citations turns multi-tab hunts into single answers. Design objects hold CAD files but track them rather than interpret them, so questions about what actually changed in a model need something that also reads the vault. Judge tools on version-specific reads, citation quality, attachment coverage and a clear write boundary, and keep the December 2027 Premier Support date in view when weighing how deep an Agile-specific integration should go.
Oracle Agile PLM is still the system of record for product data at a large number of manufacturers, and most of those sites are running a 9.3.x release on their own hardware. Oracle's published Agile PLM roadmap puts Premier Support for 9.3.6 at December 2027, which means teams are making tooling decisions for a platform they will keep using for several more years and will eventually move off. That combination is exactly why the question comes up: if an AI assistant is going to sit beside Agile, what can it actually read?
The honest answer has two halves. An assistant does well against the record layer, because Agile stores items, changes, bills of material, approvals and compliance data as structured business objects with documented interfaces. It does poorly against the geometry layer, because the CAD models themselves sit in file containers that Agile tracks but does not interpret. Knowing where that line falls is the whole evaluation.
This is a look at the object model, the three documented ways in, the questions an agent answers well today, and the point where Agile hands off to the vault. For the wider decision across platforms, the evaluation framework for AI tools across PLM covers the criteria that are not specific to Oracle.
What Oracle Agile PLM Hands an Assistant
Agile organises product data into business object classes, and the SDK Developer Guide maps each class to a programmatic interface. That mapping is the practical inventory of what any assistant can reach. The classes that matter most for engineering questions are:
Items, covering parts and documents, with their revisions and bills of material.
Changes, which include change orders, change requests, manufacturer orders and site changes.
Manufacturers and Manufacturer Parts, which together carry the approved manufacturer list.
Suppliers, Prices and Requests for Quote on the sourcing side.
Quality Change Requests and Product Service Requests, which hold the complaint and corrective action record.
Declarations, Specifications and Substances, which carry compliance data.
Sites, modelled as manufacturing sites, so item data can differ by plant.
File Folders and Designs, the containers that hold attachments and CAD files.
Two caveats come straight from Oracle's own documentation and both shape what a tool can promise. First, not every business object is exposed through the Agile API, with some report objects given as the example. Second, every read is filtered through the privilege roles assigned to the account the tool connects as, down to individual fields. An assistant connected with a restricted service account will quietly return less than the same assistant connected with a broader one, and neither you nor the tool will see an error. Any vendor claiming complete coverage of an Agile instance is describing a permission grant, not a capability.
IN PRACTICE
It integrates directly with PLM and existing workflows, making past designs, standards, and calculations instantly available. The result is fewer errors, faster decision-making, and a more consistent process across teams.
- Sergey G., Board Member
The Three Doors In, and What Each One Costs
Agile exposes three documented integration surfaces, and an AI tool has to pick among them. The Agile API is a Java software development kit that works against interfaces such as IItem and IChange, which suits code running close to the application server. Agile PLM Core Web Services wraps much of the same ground in SOAP, built on Apache Axis with standard service description files, and is the surface most third-party tools target because it does not require Java hosting. Engineering Collaboration Web Services are a separate offering aimed at the CAD side, and they are not part of Core Web Services.
Core Web Services give a retrieval tool most of what it needs. There are operations to fetch an object by type and number, to load a table off that object, to run quick and advanced searches, to read revisions, to list approvers, to audit a routable object's history, to pull attributes and table metadata, and to check a privilege before acting. For an assistant answering questions, that is a workable surface.
The documented limits are worth knowing before a demo, because they explain behaviour that otherwise looks like a bug:
Web Services do not support random access to table rows, so a tool cannot fetch row seventeen of a bill of material and update it in place. Tables are loaded and worked through.
The lifecycle phase or workflow status of an object cannot be set while that object is being created, which is why well-behaved tools create first and route second.
Deletion through Web Services is a soft delete that marks the object rather than removing it, and items with revision history or pending changes cannot be deleted at all.
History tables and related object tables are read-only, which is good news for an assistant: the audit trail it cites cannot be edited through the same door it was read from.
None of this blocks retrieval. It does mean that a tool promising to reorganise your Agile data as easily as it reads it is promising something the interface does not offer.
The Questions an Agent Answers Well Today
The useful test is not whether a tool connects. It is whether it collapses a multi-tab hunt into one answer with a citation. Agile stores the answers to the following questions as records, which is why retrieval works against them:
Where is this part used, and which assemblies does it appear on at which revision?
What changed between revision C and revision D of this item, and which change order carried it?
Who approved that change, when, and what comments did they leave on the routing slip?
Which approved manufacturer parts are released for this item, and does the list differ by site?
Which open quality records reference this part number, and are any of them linked to a corrective action?
Does this item have a compliance declaration on file, and which specification does it claim against?
Each of those is a question an engineer currently answers by opening an item, reading a tab, following a link to a change, and opening another tab. The time saved is small per question and large per week, and the accuracy gain matters more than the time: an assistant that cites the object number and revision it read from is auditable in a way that a colleague's recollection is not.
The pattern holds across record-keeping systems rather than being specific to Oracle. The same reasoning applies to purpose-built engineering integrations against PDM and PLM, and if the vocabulary in your organisation is still unsettled, the distinction between PLM and PDM decides which system holds the answer you want.
Where Agile Stops and the CAD Vault Begins
Agile does hold CAD data, through Design objects that extend the File Folder class. A Design object carries the model files, and its Structure tab records parent and child relationships between Designs, with each row pointing at a specific version rather than simply the newest one. Oracle's user guide notes that a structure row can reference a version that is not the latest, and the interface flags it as such. The title block tracks version, label, lifecycle phase and the full checkout trail of who has the file and since when. The files tab records name, type, size, category and a thumbnail.
Read that list again and notice what is in it. All of it is bookkeeping about geometry. None of it is geometry. An assistant working from Agile alone can tell you that a model is at version three, that it was checked out on Tuesday and by whom, and that it hangs under a particular parent Design. It cannot tell you that the bore grew by two millimetres, that a boss moved off the pattern, or that the part no longer clears its neighbour. Oracle's own documentation is direct about the division of labour here: most creation and maintenance of the CAD data structures is handled by the Engineering Collaboration CAD integration rather than by hand in Agile.
This is the boundary that decides what to buy. A tool that reads only Agile is a record assistant. A tool that reads only CAD is a geometry assistant. Engineering questions tend to cross the line, because the reason a change order exists is usually visible in the model and the reason a model changed is usually written in the change order.
Leo was built for that crossing. It is an AI intelligence layer on top of the systems a team already runs rather than a replacement for them, and it offers integrations with leading PDM and PLM platforms including SolidWorks PDM, Autodesk Vault, PTC Windchill, Siemens Teamcenter, Arena PLM and others, alongside local and network directories and ERP. It is trained on more than a million pages of standards, books and articles as a Large Mechanical Model, and it returns a citation with each answer so an engineer can open the source and check it. For teams weighing what their existing platform already does before adding anything, a comparison of the built-in AI each vendor ships is the right starting point, and seat-level assistants are a different reader again, covered in the roundup of AI tools for Fusion 360.
What to Check Before You Commit to a Tool
Most of the difference between tools in this category shows up in a structured trial rather than in a feature table. Six checks separate them:
Ask which interface it uses and which account it connects as. A tool on Core Web Services with a documented service account is auditable. A tool that will not say is not.
Give it a question whose answer lives on a version that is not the latest. Tools that silently read the newest row will get it wrong, and the Structure tab is full of rows that are deliberately pinned to older versions.
Require a citation that names the object number and the revision. An answer without one cannot be checked, and an engineer who cannot check an answer will stop using the tool within a month.
Test an attachment, not just a title block. Ask something whose answer is inside a document in a File Folder rather than in a field on the item.
Confirm the write boundary. A retrieval tool should not be editing released items outside a change object, and the interface makes casual edits awkward for good reason.
Check the security posture against your own obligations. Leo is SOC-2 certified and GDPR compliant, no customer data is used to train models, and customer intellectual property stays protected, which is the baseline worth holding any vendor to.
One further point is specific to Oracle Agile. With Premier Support for 9.3.6 running to December 2027 on Oracle's published roadmap, a tool whose value depends on a deep, Agile-specific integration is an investment with a known horizon. A tool that treats Agile as one of several sources it can read is easier to carry forward when the system of record changes underneath it.
FAQ
Agile Product Lifecycle Management SDK Developer Guide, Using Agile APIs, Oracle. Agile Product Lifecycle Management Agile PLM Core Web Services User Manual, Oracle. Agile Product Lifecycle Management Product Collaboration User Guide, Working with Design Objects, Oracle. Agile PLM Roadmap, Oracle.
See Leo read your engineering data
A short working session against your own items, changes and models.
Bring a question that currently takes four tabs to answer. We will run it against your own items, changes and models, and show the citation behind every answer.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams


