Geometry that holds together.
Real production parts and assemblies, not synthetic primitives or scraped mesh dumps. Every file carries geometry classification, domain tagging and native format metadata.
Available as- 01 · Dataset
- 5M+ typed, native-format CAD files
- 02 · Environment
- Part classification and rebuild checks
Licensed, stocked, ready to ship.
Plus custom collection for what doesn't exist yet.
Drawings, parts, assemblies and 3D assets
SolidWorks, STEP, Inventor, CATIA and more
From aerospace and automotive to jewellery and BIM
Combined file size across the collection
Across seven core geometry classes
Of active production design work
CAD metadata, not raw files.
Every part carries the same fields, so the dataset is filterable at scale by geometry, domain or format.
- Geometry classification
- Turned, prismatic, sheet metal, freeform, conical, toroidal or spherical.
- Domain & industry tagging
- 40+ real-world domains, from aerospace and automotive to jewellery and BIM.
- Native formats
- Original authoring formats, not just converted meshes.
- Size & complexity
- File size tracked per part and category for scale-weighted sampling.
- Assemblies & drawings
- Multi-part assemblies and 2D drawings alongside individual parts.
- Validated geometry
- Files checked to open, rebuild and match their declared structure.
Part-level metadata. Field names are illustrative.
From licence to valid geometry.
One part from assembly gearbox_A12, through the five stages every CAD file goes through before it ships.
License
Parts and assemblies come from manufacturers and design firms, under a licence that permits training. Synthetic primitives and scraped mesh dumps are excluded.
- Who
- Manufacturers
- Gate
- No licence, no ingest
Scrub
Names, customers and project codes are removed from file properties, drawing title blocks and embedded notes.
- Who
- Scanner + engineer
- Gate
- No names in any property or title block
Validate geometry
Every file is opened in its native tool and rebuilt. Units, bodies and mates are checked, and broken files are fixed or dropped, never shipped.
- Who
- CAD engineers
- Gate
- Zero rebuild errors
Classify
Engineers tag geometry class, domain and manufacturing process, so the dataset filters by what the part is, not just its file type.
- Who
- Mechanical engineers
- Gate
- Every part classified
Package
Native files ship with neutral exports and a metadata record per part, so they open with or without the original CAD tool.
- Who
- Delivery
- Gate
- Metadata record per part
Built for frontier model tasks.
Scope, review criteria and delivery format are agreed per engagement, from the first sample to the final eval.
- Text-to-CAD
- Generative models that produce editable, native geometry.
- Part classification & search
- Retrieval and tagging across large parts libraries.
- Design copilots
- Assistants trained on how real assemblies are built.
- Manufacturability
- Geometry-aware models for DFM and costing.
- 3D understanding
- Meshes and B-reps for spatial reasoning models.
What buyers ask us first.
Something else? Ask a specialist →
01Which CAD formats do you support?
30+ native formats including SolidWorks, STEP, Inventor, CATIA, DWG, DXF and IGES. Availability depends on the dataset and licensing scope.
02How do you handle proprietary design data?
Permitted use and handling requirements are agreed as part of the licensing scope. We document usage rights and IP compliance for the data provided.
03Is the CAD data sanitized?
CAD files can contain project, author or proprietary details. Metadata handling or removal requirements are agreed and validated for each project.
04How does this compare with public CAD datasets?
Public CAD datasets mostly gather parts from openly shared online models, under terms you need to check before commercial use. Ours are licensed production parts, assemblies and drawings in their original native formats, tagged by geometry class and industry domain, with rights documented for commercial training.
05Do you have B-rep data, or only meshes?
Both. B-rep formats such as STEP, Parasolid and IGES ship alongside native SolidWorks, Inventor and CATIA files, with STL or OBJ meshes where you need them.
06Can you build evaluations from CAD data?
Yes. Part classification tasks and rebuild checks test whether a model's output opens, rebuilds and matches its declared structure, using the same validation the dataset itself goes through.
Same standard in every domain.
Audio & speech
Synchronized audio with transcripts and event labels.
- 01 Dataset
- 5.7M+ transcribed, diarized assets
- 02 Environment
- Transcription evals per language and accent
Egocentric video
First-person footage with action and object labels for physical AI.
- 01 Dataset
- 200K+ hrs synced video, IMU, action labels
- 02 Environment
- Action and manipulation benchmarks
Company operations
Workflows, approval trails and legal documents for enterprise agents.
- 01 Dataset
- 52+ companies' ordered approval trails
- 02 Environment
- Agent tasks over approvals, operator-graded
Conversational
Multi-turn conversations, ranked and reviewed for assistant training.
- 01 Dataset
- Multi-turn dialogue, intent-labelled
- 02 Environment
- Multi-turn evals, rubric-graded
Private code
Repositories, commits, pull requests and review history.
- 01 Dataset
- 3B+ lines with commit, PR and review history
- 02 Environment
- Agentic engineering tasks, engineer-graded
Healthcare
Clinical notes, records and imaging, de-identified and expert-reviewed.
- 01 Dataset
- 10M+ de-identified, patient-linked records
- 02 Environment
- Clinical case tasks, clinician-graded
See real cad & 3d data first.
A representative sample under NDA, in the format your pipeline reads. Run it through your own evals first.
- Sample records, as delivered
- Datasheet and schema
- Review criteria
- SLDPRT / SLDASM · STEP / STP · IPT / IAM
{"file":"assembly_08.step","rev":4,
"format":"STEP","parts":3,
"tree":["Frame·01","Joint·04","Surface·02"],
"geometry":"prismatic","domain":"industrial",
"size_mb":12.4,"valid":true,
"rights":"documented"}