Implicit-first computational design
Computational design that refuses to guess.
kForge turns implicit geometry into printable parts through voxel FEA, topology optimization, and a verification layer that never reports a number it didn't measure. Built for FDM 3D printing — runs on your machine.
Desktop app + command-line tool · your designs stay local
The tool, not a mockup
Change the geometry. See the stress.
Von Mises stress on a real part — measured, with a live colorbar and pass/fail read against your own criteria.
Honest by design
Every result is measured, or it isn't reported.
Most design tools are happy to hand you a confident number. kForge won't. Requirements are derived from your project's structure, checked against on-disk evidence, and a value the tool can't actually measure comes back as pending — never invented to fill a gap.
The analysis is real, too: voxel FEA cross-validated against CalculiX to Δ = 0.00% and standard NAFEMS benchmarks. When a part passes, it passed for a reason you can trace.
AI-native
Bring your own model. It still can't fake the physics.
kForge exposes its whole toolset as a command line and an MCP server. Any agent that can drive a terminal — Claude, Gemini, GPT, Grok — can model the geometry, run the analysis and read the verdict, on its own.
The part that makes it trustworthy: the sandbox and the certify layer apply to the agent too. It can propose anything — but it can't report a number it didn't measure, and a fabricated result is caught, not shipped.
Or summon one right inside the desktop app — kForge has a built-in terminal, so your agent works next to the viewport. Bring your own key.
you ▸ design a 40 mm bracket — 200 N tip load, SF ≥ 2 agent ▸ kforge project new bracket kforge evaluate max_von_mises 38.2 MPa PASS kforge simp mass 68 g → 41.7 g kforge certify certified agent ▸ Done — 41.7 g, SF 2.1×. Fatigue is pending: you gave no load spectrum, so I can't claim it.
The pipeline
From a field to a certified part.
One continuous model — the geometry, the analysis, and the optimization all live on the same grid.
Model as an SDF
Signed distance fields, CSG booleans, and TPMS lattices. Geometry is a field you sample and drive — not a fragile mesh.
Voxel FEA
Finite-element stress straight off the field grid, with materials calibrated for FDM prints.
Topology
SIMP optimization finds the lightest shape that still carries your loads — printable, on the same grid.
Verify
Requirements checked against measured evidence, with a report you can hand to someone and trust.
Made in kForge
Real parts, from field to print.
Each of these is an SDF model — analyzed, optimized, and exported to STL by the same tool. No CAD import, no mesh repair.
Whole machines, not just parts
Assemblies you can take apart.
Multi-part projects with joints, motors and cabling — checked for fit and collision, and explodable to see exactly how every piece goes together.
What's inside
An engineering tool, not a shape editor.
Implicit fields & lattices
SDF primitives, smooth blends, shells and offsets, and gyroid / Schwarz TPMS lattices — driven by parameters or by other fields.
Voxel FEA, verified
Hex-element stress analysis on the SDF grid, cross-checked against an industrial solver and NAFEMS benchmarks.
CalculiX Δ = 0.00%Topology (SIMP)
Density-based topology optimization with print-direction and overhang awareness, so the result is one you can actually make.
The certify layer
Multi-level certification with waivers and accepted risks — the anti-fabrication core that reports pending instead of guessing.
Real dynamics
Assemblies, joints and cable-driven mechanisms simulated in MuJoCo, plus fatigue, buckling, modal and bearing-life checks.
An agent that designs
An optional AI agent proposes geometry, simulates it, critiques the result and revises — under the same honesty rules as the rest.
Get the preview
Try kForge on your machine.
kForge ships as a desktop app and a command-line tool. This is an early-access build for a small group of first users — expect rough edges, and honest ones.
Download for Windows (64-bit)