CadQuery

CadQuery is an open-source Python library for creating parametric CAD geometry programmatically. In AI-driven engineering workflows, it serves as the code generation target for CAD generation agents: the model writes CadQuery Python code, and a deterministic controller executes it to produce a STEP artifact.

Why CadQuery for agent pipelines

• Scriptable — AI models generate Python code rather than binary CAD files, making generation tractable
• Parametric — geometry is defined by parameters that can be systematically modified during repair loops
• STEP export — produces standard STEP files that can be ingested by FEA solvers, PLM systems, and downstream CAD tools
• Inspectable — the code is readable, so errors in the generated geometry can be traced back to specific lines

Role in the FEA feedback pipeline

Engineering Brief → Model generates CadQuery code
                 → Controller executes code → STEP artifact
                 → FEA validation → typed feedback
                 → Model repairs CadQuery code → repeat

CadQuery is where the model’s reasoning becomes physical geometry. The controller handles everything after: execution, rendering, meshing, and validation. This separation keeps the model focused on design decisions and the controller focused on measurement.

Limitations for production use

• Generated CadQuery code may fail to execute (syntax errors, invalid geometry, assembly constraint violations)
• Even valid geometry may fail engineering constraints — visual plausibility ≠ engineering validity
• The Hephaestus-CCX benchmark found that frontier models rarely produce strict-passing artifacts on the first attempt
• CadQuery has its own API versioning; generated code may reference removed or changed methods

Related

• FEAFeedbackLoop — the agent pattern CadQuery enables
• ManufacturingAIAdoption — where CAD generation agents sit in the adoption landscape