CADlingo 2.0 creates editable parametric models in a browser from natural-language descriptions. It can ask follow-up questions, handle multiple components and export STL or STEP. According to the developer, the result is not merely a triangle mesh but a model retaining dimensions and editable relationships. No independent reproducible accuracy benchmark is public.

Why parametric representation matters

A mesh generator can create a shape without preserving that a hole is 42 mm, concentric or positioned from a datum. A parametric model keeps dimensions and relationships editable. That can be a better starting point for fixtures, enclosures or simple assemblies where the next iteration changes known parameters.

STEP export is not an automatic quality seal. A file may contain a solid yet still include faulty edges, tiny surfaces, wrong units or topology repaired differently by the destination CAD or CAM system. STL discards parametric history and transfers only a triangulated surface.

Follow-up questions as a safety function

Natural language is often ambiguous. 'Two holes in the centre' does not define diameter, axis, through condition or datum. A good system should ask for missing dimensions and relationships rather than guess. Evaluate whether it detects ambiguity or confidently inserts arbitrary values.

Risk grows with multiple components. Beyond part geometry, an assembly needs mates, degrees of freedom, collision, tolerance and sequence. Public product material does not demonstrate complex constraints, configurations, drawings or full assembly behaviour.

Data protection and versioning

For browser CAD, establish where prompts, models and exports are stored, whether they train models, how deletion works and which access logs exist. Trade secrets, customer geometry or export-controlled data should enter only an approved contractual and technical environment.

A prompt is not a manufacturing source of truth. Retain input, system and model version, follow-up dialogue, generated CAD revision, human edits and approved export. Otherwise a later regeneration is not demonstrably the same design.

Pre-production verification

Compare the model with a requirement list: units, primary dimensions, holes, walls, radii, fits and prohibited geometry. Run solid and topology checks in the destination CAD, then measure critical features. Re-import STEP and STL separately because export is another failure point.

For assemblies, run collision and motion checks. Loaded parts require appropriate engineering analysis and material models; additive production also needs orientation, minimum wall, support, tolerance and finishing review. Begin with a low-risk prototype or inspection coupon, not a critical end-use part.

Build an internal benchmark

Choose ten predefined tasks from a simple bracket to a multi-component fixture. Run each specification three times and measure correct dimensions, complete constraints, useful questions, regeneration failures, export success and human correction time. Requirement-level conformance is more informative than an attractive render.

Fact, vendor claim and inference

It is verifiable that the browser service communicates STL/STEP export, parametric editing and multiple components. Reliable language interpretation and productivity remain vendor claims without a public independent benchmark. Our inference is that the strongest early use is low-risk concept generation and first CAD drafts under mandatory engineering review.

Sources and transparency

Capabilities come from CADlingo's primary product page; launch context comes from 3Druck. This is not a hands-on test. Validation steps are editorial methodology recommendations.