AI for Engineering Knowledge Management
Most CAD teams without a PDM still have structure, just no database under it. Here is what an AI agent can and cannot learn from a plain network folder.
·
⏱
7 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
A network drive without a PDM is not a lack of structure, it is structure without a database underneath it. An indexing agent working against that structure can read file system metadata and CAD-embedded custom properties, and it can interpret filename and folder conventions as revision signals. What it cannot do is invent check-out locking, enforced lifecycle states, or bill-of-materials traceability that a team never built into its folders in the first place. Leo's desktop app indexes exactly this kind of folder using native file watchers, so a team can get working search today and layer in a formal vault later if and when it makes sense, without redoing the indexing work already done.
Most engineering teams that say "we don't have a PDM" do not mean their files are chaos. They mean their files live on a network drive, organized by project or customer folder, with no database sitting underneath telling anyone which file is current, who owns it, or what assembly it belongs to. That distinction matters the moment you try to point an AI agent at those files and ask it to find a part. An agent cannot query a vault that does not exist. It has to work from the same thing an engineer does: a folder tree, a set of filenames, and whatever properties happen to be saved inside each file. This post walks through what that actually gives an agent to work with, where the gaps are, and how Leo indexes a plain CAD folder without asking a team to stand up a PDM system first.
Why "No PDM" Usually Means a Folder, Not Nothing
A shared network drive is still a system, just an informal one. Reseller guidance for teams running without a PDM typically starts with the same recommendation: fix a folder hierarchy before work begins, and point every workstation at the same reference-document location so external CAD links do not break when a file moves between folders. That structure usually holds up reasonably well for a small team working off one server, especially early on, when the number of active projects is small enough that everyone roughly remembers where things live.
What that structure does not have is a database layer. There is no central table mapping part numbers to files, no lifecycle state marking a design "Released" versus still under review, and no single record of who last touched a given assembly or why. Everything an agent, or a person, can learn about a file has to come from the file system and the file itself, not from a query against a managed index the way it would against a PDM vault.
This is also why the transition to a PDM often gets delayed longer than teams expect. The folder-and-convention approach is not obviously broken day to day; it degrades slowly, as the number of projects, engineers, and revisions grows past what any one person can hold in their head. An indexing agent is useful precisely at that point, because it can hold far more of that structure in working memory than a person scanning folders by hand.
IN PRACTICE
The geometry search has been invaluable, helping me find standard parts instead of designing new ones, saving a huge amount of time and effort. The search system is smart and CAD-aware.
- Eytan S., R&D Engineer
What a File System Can Actually Tell an Agent
Without opening a CAD application, an indexing agent scanning a network folder has three layers of information available. The first is the path and filename. The second is file system metadata: created date, last modified date, and file size. The third, and the most useful, is whatever custom properties are saved inside the CAD file itself, things like part number, description, revision, material, or engineer of record.
That third layer does not require a CAD license to read. SolidWorks, for example, ships a separate Document Manager API specifically so custom properties and configuration data can be pulled out of a part or assembly file without loading its full geometry kernel or opening the application at all. Other major CAD platforms expose comparable lightweight metadata access through their own SDKs. This is the raw material an indexing agent works from. It is real, structured data, but it is scattered across thousands of individual files rather than organized into the single attribute table a PDM vault would maintain.
The practical effect is that an agent can build something that looks like a database, part number, revision, material, last-modified date, per file, purely by reading what is already sitting on disk. No one has to go back and manually tag every existing file first. The properties an engineer already filled in while modeling the part are the same ones the agent reads back out.
Reading Revision Signals From Filenames and Folder Habits
Without a managed vault, revision history usually lives in the filename itself: a suffix like _v2, _RevB, or _FINAL, a date embedded in the name, or a split between a released subfolder and a work-in-progress one. None of that is a single authoritative field the way a PDM revision attribute is. An agent has to treat it as a pattern to interpret, not a value to read, and the pattern breaks down exactly when it matters most: under deadline pressure, when naming conventions are the first thing a team lets slide, engineers can end up with three files that all plausibly look current.
Renaming and copying add another failure mode. Reseller guidance for non-PDM teams specifically warns against renaming or moving CAD files directly on disk, since doing so can silently break the external references between a part and the assemblies that use it; the safer path is a built-in repackaging step that updates those references together rather than a plain file-system move. An agent reading the folder after that kind of manual rename has no way to recover a link that was already broken before it ever scanned the drive; it can only flag that a reference looks missing. Teams that have standardized their naming and revision conventions up front give an indexing agent a meaningfully cleaner signal to work from, and that discipline pays off well before anyone considers a PDM migration.
What Indexing Without a Vault Still Cannot Do
Reading a folder well does not manufacture governance a team never put in place. Three gaps do not go away just because an agent is watching the drive. There is no check-out and check-in locking, so two engineers can open and edit the same file at the same time with no warning to either of them. There is no enforced lifecycle state distinguishing a released design from one still under review, only whatever a folder name or filename suffix implies, and nothing stops a "final" file from actually being an intermediate draft someone forgot to rename. And there is no automatic where-used or bill-of-materials traceability connecting a given part file to every assembly that actually references it; that link exists only inside each assembly file's own internal references, if it exists there at all.
An agent can surface likely candidates, flag when two files look like they might be the same part at different revisions, and point out a modified-date conflict that suggests concurrent edits. It cannot promise a single source of truth the way a proper PDM search layer is built to provide. That is a real limit, worth stating plainly, and it is the honest answer any team should expect before treating folder indexing as a full substitute for a managed vault rather than a useful step before one.
How Leo Indexes CAD Folders Without Requiring a PDM
Leo's desktop app watches local and mapped network CAD folders directly, using native operating-system file watchers rather than requiring a team to first migrate into a managed vault. As files are added, edited, or renamed, Leo indexes the folder structure, the file system metadata, and the embedded custom properties each CAD file already carries, and surfaces likely-current parts and their properties in response to a plain-language search rather than a folder-by-folder manual hunt. For a team whose actual system today is a shared drive and a set of habits, this means the search layer can go in before a PDM rollout, not after one.
Leo also offers integrations with leading PDM and PLM platforms, including SolidWorks PDM, Autodesk Vault, PTC Windchill, Siemens Teamcenter, Arena PLM, and others, so a team that later does adopt a vault does not lose the indexing layer already built on top of its folders; it extends into the new system instead of starting the search problem over from scratch. The same underlying part-finding problem that drives engineering teams to redesign parts that already exist shows up just as often on a folder-based team as it does inside a poorly searched vault, and a shape-based search layer that does not depend on metadata being complete works the same way in either setup, since it is reading the geometry itself rather than the filename around it.
FAQ
See Leo Index Your CAD Folders
No PDM required. Point Leo at your network drive and start searching.
Leo watches your CAD folders directly and indexes parts and revisions as they change, so your team can search what it already has before deciding on a PDM.
Schedule a Demo →
#1 New AI Software Globally - G2 2026
Enterprise-grade security
Trusted by world-class engineering teams
