Cartesian: AI-Powered Precision 3D Modeling

Cartesian by Formas is an AI tool that converts text, sketches, and images into precise, editable 3D CAD models. It focuses on professional-grade geometry like NURBS solids, ensuring that models are ready for manufacturing, BIM workflows, and architectural design. The platform allows users to iterate quickly while maintaining exact control over specific project components.

Key Points

• Generates precise, editable solids and NURBS geometry instead of simple polygon-mesh approximations.
• Supports multi-modal inputs including text prompts, sketches, photographs, and existing CAD models.
• Maintains 'exact' elements defined by the user while allowing the AI to modify or generate the surrounding environment.
• Provides structured data for professional workflows, including named elements and compatibility with BIM and major CAD software.
• Scales from small product design and manufacturing to large-scale architecture and urban infrastructure.

Sentiment

The overall sentiment is mixed and technically skeptical. Commenters broadly agree that AI-assisted CAD is an important and promising direction, and some are enthusiastic about adjacent workflows that already save time. However, the dominant tone is cautious: Hacker News wants clearer evidence that Cartesian can produce precise, editable, manufacturable models and can respect the broader physical constraints of engineering beyond surface-level geometry.

In Agreement

• AI-assisted modeling can dramatically speed up tedious CAD work, especially through fast iteration on practical hobbyist parts and prototypes.
• Generating precise editable geometry such as Rhino/openNURBS-style models is more compelling than ordinary AI mesh generation because it fits professional design workflows better.
• The field is clearly moving toward AI-augmented mechanical, architectural, and product-design tooling, much as AI has already affected software work.
• AI can become more useful when paired with model-inspection, measurement, simulation, and constraint-checking tools instead of being treated as a plain text-to-model generator.
• Natural-language and reference-driven modeling could lower the friction of creating parts and free users to focus on fasteners, assembly choices, and other physical design concerns.

Opposed

• CAD geometry is only part of engineering; materials, manufacturability, tolerances, durability, assembly, testing, and debugging remain hard physical-world problems.
• Using AI in civil or mechanical engineering could create false confidence for non-experts, especially in safety-critical work.
• The product page reads more like aspiration than a demonstrated tool, and commenters want concrete demos showing correct, usable models.
• Some generated examples appear more suitable for games or visualization than architecture, product design, or manufacturing.
• Text prompting may be a poor interface for spatial design because many designers think through sketches, shapes, and direct manipulation rather than verbal descriptions.
• The website's moving background, scroll behavior, contrast, and marketing copy make it harder for commenters to evaluate the product seriously.