AI for Engineering Knowledge Management
What an AI layer can actually read inside SOLIDWORKS Manage: items, BOMs, change processes and project records, plus the criteria worth applying before you buy one.
·
⏱
8 min read
Michelle Ben-David
Michelle Ben-David is a mechanical engineer and Technion graduate. She served in an IDF elite technology and intelligence unit, where she developed multidisciplinary systems integrating mechanics, electronics, and advanced algorithms. Her engineering background spans robotics, medical devices, and automotive systems.
BOTTOM LINE
SOLIDWORKS Manage holds two kinds of engineering truth that are rarely read together: the files and versions it inherits from SOLIDWORKS PDM Professional, and the items, bills of materials, change processes and project records it adds above them. The questions that cost engineering teams real time are the ones that cross that boundary. Which released items consume this component, what did we decide last time, where has the item structure drifted from the model. An AI layer worth buying reads the item layer rather than only the vault, respects the permission model already configured, cites what it returns, and stays a layer rather than becoming another system of record. Leo is built to that shape: retrieval across PDM and PLM data, replacing neither.
SOLIDWORKS Manage sits in an awkward spot. It is not the vault, and it is not a full enterprise PLM either. Dassault Systemes positions it as a set of advanced data management tools built on the file management capabilities of SOLIDWORKS PDM Professional, adding project, process and item management on top of that foundation.
The practical consequence is that a team running Manage carries two layers of engineering truth at once. Files, versions and references live in the PDM vault below. Items, bills of materials, change processes, project stages and task history live in Manage above. Both layers are structured, both are queryable in principle, and almost nobody queries them together.
Most engineers can find a file. Far fewer can answer a question that crosses both layers in under an hour. Which released items still carry the obsolete connector, which change process retired the previous revision, and what did the reviewer actually write in the comment field before approving it. The answer exists somewhere. It is spread across a vault, a set of item records, a process history and somebody's memory of a decision made eighteen months ago.
This post covers what an AI layer can realistically reach inside SOLIDWORKS Manage, what stays behind in the vault, and the criteria worth applying before a team commits to one.
What SOLIDWORKS Manage Actually Holds
Before judging any tool against Manage, it helps to be precise about what Manage stores that the vault does not. Dassault Systemes groups the product into four areas:
Project management, covering stages, milestones, timelines, resource utilisation, user tasks and timesheets.
Process management, covering workflow-driven business processes such as engineering change requests and approvals, with configurable decision points.
Item management, covering item records, bills of materials, product variations and the link between item data and SOLIDWORKS drawings.
Dashboards and reports, covering the graphical rollups configured to a company's own reporting standards.
The important structural fact is the item. A PDM vault manages files: a part file, its versions, its references and its workflow state. Manage introduces a record that is not a file. An item can exist before any geometry does, can point at several documents, can carry fields the CAD file has no place for, and can appear on a bill of materials that includes purchased parts, firmware and packaging that were never modelled at all.
That distinction is the reason cross-layer questions are hard. Ask the vault which assemblies reference a part and you get a reference graph. Ask Manage which released items consume that part and you get a different graph, drawn over different records, with different revisions and different approval history. Neither answer is wrong. They simply describe different things, and reconciling them by hand is where afternoons go. If your team is still deciding which layer it needs, the distinction is covered at length in our guide to SOLIDWORKS PLM versus PDM.
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
Where the Questions Actually Break
Teams rarely struggle with retrieval of a single known object. They struggle with questions that have to traverse records before they can be answered at all. Four shapes come up repeatedly.
The first is impact. A supplier discontinues a component, and somebody has to establish every released item that consumes it, at every level of every bill of materials, including the assemblies that reach it through three intermediate subassemblies. Manage holds the structure. Walking it by hand across a few hundred items is a day of work, and the result is only as good as the patience of the person doing it.
The second is precedent. An engineer is about to raise a change on a bracket and wants to know whether this has been attempted before. The relevant history is a closed change process from two years ago, with the reasoning sitting in a comment on a rejected stage. It is in the system. It is not findable by anyone who does not already know it exists.
The third is reconciliation. The item bill of materials and the CAD structure have drifted, usually because a configuration was added, a part was replaced in the model, or a purchased component was added at the item level only. Someone has to find the drift before manufacturing does. The mechanics of this are worth reading on their own, which we covered in our piece on multi-level BOM management across nested assemblies.
The fourth is status. A project stage is late and the cause is somewhere in the task history, the linked change processes and the documents that never cleared approval. The data is complete. The reading of it is the work.
Items, BOMs and Variants: Where Retrieval Earns Its Keep
Item management is the part of Manage where an AI layer pays for itself fastest, because it is the part with the most structure and the least tolerance for a wrong answer.
A useful AI layer over item data does three things. It resolves a question into the right records rather than the right keyword, so that a question about a discontinued connector returns the items that consume it rather than the documents that mention it. It traverses the bill of materials to the depth the question requires, rather than to the depth a person has the patience for. And it reports where the structure is ambiguous instead of smoothing over it, because an item BOM that disagrees with the CAD structure is a finding, not a formatting problem.
Variants raise the difficulty again. Product variations in Manage mean that a single design intent can resolve into many item records that differ by a handful of fields. The question is almost never "what is this item", it is "which of these seventeen variants did we actually release to the customer in that region, and what changed between them". That is a comparison over records, and it is the kind of comparison that people quietly stop doing when it takes too long.
This is also where the reuse argument shows up. Engineers design a new part because they could not establish that a suitable one already existed, not because they wanted to. Every one of those parts then acquires a drawing, a supplier, an inspection routine and a place in the item structure, and the cost of that decision keeps arriving for years. When the item layer becomes searchable in the way engineers actually ask, that failure mode shrinks without anyone being asked to change how they work.
Change Processes and Projects as Searchable History
The process and project modules are usually treated as operational machinery. They route an engineering change request, collect approvals and close out. Once closed, the record is filed and effectively forgotten.
Treated differently, those closed records are the most valuable engineering knowledge a team owns. A completed change process contains the problem as originally stated, the options considered, the objections raised at each stage, the decision, and the identity of the person who made it. That is a reasoned engineering judgement, captured at the moment it was made, with none of the reconstruction error that creeps in when somebody is asked about it later.
An AI layer that can read process history turns that archive into something an engineer can consult before raising a new change rather than after. The questions it answers are the ones that currently get asked in a group chat: has anyone tried this, why was it rejected, who decided, and what were the conditions at the time. The project module adds the schedule dimension, which matters because the reason a decision was made is often that a stage gate was two weeks away.
There is a discipline point here that no tool solves. A process history is only as good as what people wrote in it. An AI layer makes thin records visible very quickly, because the questions it cannot answer are exactly the ones where the reasoning was never written down. Most teams find that useful. It is a short list of places where the next engineer is going to be stuck.
How to Evaluate an AI Tool for SOLIDWORKS Manage
Most tools marketed at engineering teams were built for documents and then pointed at engineering data. Five criteria separate the ones that survive contact with a Manage deployment.
Does it read the item layer, not only the file layer. A tool that indexes the vault and stops has solved the easy half. The items, bills of materials and variants are where the questions live, and they are covered in our write-up on AI for PDM and PLM integration.
Does it respect permissions as configured. Manage inherits the vault's security model, and an answer that quietly includes a record the asker is not entitled to see is a compliance incident, not a feature.
Does it cite. An uncited answer about a released revision is unusable, because verifying it costs more than finding it manually would have.
Does it read the drawings and documents as engineering content. Tolerances, notes and callouts carry meaning that generic text extraction loses.
Does it stay a layer rather than becoming another system of record. A tool that wants to own the data has recreated the migration problem the team was trying to avoid.
Leo is built to sit above this stack rather than replace any part of it. It offers integrations with leading PDM and PLM platforms, including SOLIDWORKS PDM, Autodesk Vault, PTC Windchill, Siemens Teamcenter and Arena PLM, and connects to the rest of an organisation's knowledge base: local and network directories, document stores and ERP. The vault stays the vault, Manage stays the system of record for items and processes, and the retrieval layer answers across both with citations an engineer can open and check. The same evaluation applied to the layer below is set out in our review of the best AI tool for SOLIDWORKS PDM, and the comparable analysis for a neighbouring platform is in our look at AI for Fusion Manage.
On security, the answer has to be specific rather than reassuring. Leo is SOC-2 certified and GDPR compliant, no customer data is used to train models, and customer intellectual property stays the customer's.
FAQ
See Leo read your Manage data
Items, BOMs and change history, answered with citations
Bring Leo to the stack you already run. It reads across PDM and PLM data, cites every answer, and leaves your vault and your item records exactly where they are.
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams
