Browser-native simulation
CAD import, CSG edits, meshing, boundary conditions, solver setup, and post-processing happen in a responsive browser workspace without forcing a local workstation install.
Torso is building the connected operating layer for engineering simulation and scientific compute.
The Torso ecosystem unifies browser-native CAE, high-performance remote linear algebra, distributed solvers, replayable engineering workflows, and AI-assisted setup into one coherent simulation stack. Geometry, meshing, physics, and compute stop being fragmented tools and become one programmable system.
One ecosystem from model to machine.
Torso is designed around a simple idea: simulation should be interactive, reproducible, scalable, and compute-aware by default. Engineers should not have to stitch together CAD translators, meshers, solver wrappers, remote compute scripts, and post-processing utilities before they can answer a design question.
Replayable engineering state
Every workflow can become a journal: readable, replayable, and parameterizable. A click-driven model and an automated Python run can describe the same simulation state.
Compute-aware by design
Dense kernels, sparse solves, MPI-distributed matrices, remote CPU/GPU execution, and reusable content-addressed data are first-class parts of the platform.
Two products, one simulation stack.
TorsoCAE gives engineers the modeling and physics cockpit. TorsoHPC gives applications the remote numerical engine. Together they close the loop from geometry to scalable compute.
TorsoCAE
Open TorsoCAE
AI-driven CAD, meshing, FEM, CFD, and post-processing.
A browser-native CAE workspace for geometry operations, robust meshing, structural analysis, thermal simulation, CFD, FSI, contact, nonlinear dynamics, modal workflows, and replayable automation.
- STEP/IGES/SAT import with CSG and assembly-aware workflows
- DOLFINx, MFEM, Kratos, native assembly, and TorsoHPC-backed solvers
- Python journals, AI-assisted setup, live results, and browser post-processing
TorsoHPC
Open TorsoHPC
Remote linear algebra and solver infrastructure for demanding workloads.
A high-performance compute service for scientific applications that need fast, secure, reusable dense and sparse numerical kernels without rebuilding remote infrastructure from scratch.
- Remote BLAS, cuBLAS, MKL, Blaze, AMGCL, PARDISO, and MPI solver paths
- Content-addressable storage to avoid repeated multi-GB data transfers
- Native C++ client, Python integration, encrypted transport, and scalable deployment
Connected Solver Path
TorsoCAE models the physics. TorsoHPC scales the numerical solve.
The integration boundary stays explicit: CAE owns engineering state and assembly; HPC receives numerical operators and solver controls, executes the selected backend, and returns results to CAE visualization and reporting.
TorsoCAE
CAD, mesh, materials, BCs, physics, and assembly.
TorsoHPC
Remote dense, sparse, DCSR, direct, iterative, and modal execution.
TorsoCAE
Fields, eigenmodes, post-processing, reports, and replayable journals.
From design intent to distributed solve.
Torso treats simulation state, mesh identity, solver inputs, and compute artifacts as connected data. That is the foundation for fast iteration, exact replay, and scalable execution.
Unified Flow
Geometry
CAD import, CSG, edits, topology, stable body and surface identity.
Physics
Materials, boundary conditions, contacts, fluids, dynamics, and solver controls.
Compute
Local, parallel, remote, CPU, GPU, direct, iterative, and DCSR solver execution.
1
Project state contract across UI, journals, reconnect, and saved simulation files.
0 MB
Repeat transfer target for reused compute artifacts through content-addressed storage.
MPI
Distributed sparse workflows designed to avoid serial matrix round-trips.
Revolutionizing simulation means removing the seams.
The industry has treated simulation as a chain of isolated tools: CAD here, meshing there, solvers somewhere else, and compute as a separate operational problem. Torso collapses those seams into a unified ecosystem where state is durable, compute is elastic, and every result can be traced back to the exact model that produced it.