Challenge
CAD vendors usually demo single parts. Real programs are trees: subsystems, modules, purchased assemblies, hardware, and raw stock, all of which have to stay consistent while hundreds of parts change daily.
solidSF needed proof that a browser-native kernel holds up at program scale - not a mesh import, but native parametric geometry with assembly structure, mates, and fabrication outputs derived from the same source of truth.
Solution
The team modeled a complete desert dune buggy as 200 native parts across 13 subsystems and 47 modules: a 37-tube AISI 4130 space frame, double-wishbone front and trailing-arm rear suspension, VW Type 1 drivetrain packaging, cockpit controls, and 14 formed aluminum body panels.
Everything downstream is computed, not redrawn. The BOM decomposes to 407 nodes with a clean make/buy split and 120 raw stock lines carrying real material specs. The assembly mirrors the same tree with 559 components and 115 mates. Tube endpoints follow one trim rule, so the cut list, the notch sheets, and the physical fishmouth Boolean cuts all derive from the same 3D sketch centerlines.
“If the kernel can take one vehicle from top assembly to raw materials with no dangling branches, it can take your product there too. That is the whole point of the program.”
Andy Norris, Founder, solidSF
Results
The fabrication pack is real enough to take to a shop: 41.2 meters of 4130 tube across 37 frame members with per-tube cope geometry, 14 sheet-metal flat patterns covering 2.6 square meters of developed blank, and hardware resolved to 296 placed instances from 24 seed families.
The dossier is regenerated deterministically from the model sidecars - every number is computed from a named artifact and every quality gate passes. The program now seeds operator-trajectory training for solidSF's native AI agents.