To license CAD files for AI training models, first trace where each file came from and who can grant permission for the proposed use. The fact that a company has a CAD file in its project archive does not settle that question. A model might be the company's own design, a customer deliverable, a supplier part, or a contractor's work. Different agreements may apply within one folder.
This is a practical review checklist, not a legal conclusion about any particular archive. Have counsel review the relevant contracts and jurisdictions before signing a license.
Sort files by source, then by permission
Start with a sample from each project type. For each group, record:
| Question | Evidence to find | Why it matters |
|---|---|---|
| Who created the design? | Project records, author information, contractor agreements | The file creator and the company holding the archive may be different parties. |
| Was it made for a customer? | Master services agreement, statement of work, purchase order | A customer contract may limit reuse or disclosure. |
| Does it contain third-party content? | Supplier components, imported models, licensed libraries | Rights to the overall project may not cover every embedded part. |
| What use is being granted? | Draft license terms | Training, evaluation, redistribution, model sharing, and outputs can require different language. |
| Can the source be documented? | Project ID, original path, date, contract reference | A buyer needs a way to audit what entered the dataset. |
Keep the answer with the file or project ID. If a group cannot be cleared, exclude it until the missing agreement is found. Do not describe the whole archive as licensable because one project passed review.
Specify the AI use in the agreement
“AI use” is too broad to tell either side what will happen to the files. State whether the buyer may copy files for training, run evaluation, retain derivatives, share data with contractors, sublicense access, publish samples, or use the resulting model commercially. Define the licensed dataset version, permitted users, term, territory, security requirements, deletion obligations, and any excluded projects.
The same care applies to public datasets. A file being downloadable does not make every use permitted. Creative Commons' AI training primer explains why copyright rules vary by jurisdiction and use. It also explains the conditions attached to CC licenses, including NonCommercial terms. Check the actual license and source record for the files you plan to use.
Build a rights sheet a buyer can inspect
A simple sheet can have one row per project or file group: source company, project ID, creator, customer restrictions, third-party content, contract reference, allowed uses, exclusions, and reviewer. Add a status such as cleared, needs review, or excluded. Keep the underlying contracts private unless the parties agree to share them, but be ready to show how the status was reached.
This sheet is useful to both sides. An archive owner can avoid offering files it cannot clear. A buyer can scope diligence and sample only the groups that might be included in a license. The seller's CAD data inventory guide covers the technical side of that first conversation.
Put the rights story next to the dataset story
The DWGAtlas-194K card describes its source as engineering project material from Brazilian companies under exclusive rights and gives the available drawing count and sample. That is more useful than a bare file count. A prospective buyer should still ask to review the specific license scope and supporting records for its intended use.
If you are assessing a CAD archive, share its file inventory and the rights questions you have already resolved with Orla. A narrow, well documented group is a better starting point than an unreviewed drive.