Build, run and check finite element models by talking to your AI.
feaX connects the AI assistant you already use to the finite element solver you already own. Describe the problem or hand over an existing model; feaX builds it, writes it for your solver, reads the results back, and checks them against an independent second solve. Nothing leaves your machine.
A pre- and post-processor built for an AI to drive, not a person to click.
Finite element work has always had three parts: building the model, solving it, and making sense of the results. The solver is mature software you already own and trust. The other two parts are where engineers spend their days, and where the errors live.
feaX takes those two parts and makes them something an AI assistant can do well. It holds one neutral, typed description of the model, validates it before anything is written, writes the input file your solver expects, reads your solver's results back, and compares them with a second, independent solve.
It does not solve anything itself. It does not send your models anywhere. It gives you back every input file it writes, so nothing depends on feaX to reproduce.
A small local server between your assistant and your solver.
Hours back, a second opinion on every run, and your data where it is.
What feaX does not do: it does not replace the engineer's judgement or signature. A second solver agreeing with the first means the model was solved as written, not that it was the right model. feaX makes the evidence for that judgement fast and complete.
One planted error, found in under three minutes.
A bolted end-plate connection, modelled in a commercial solver, with one node set left incomplete on purpose. The original solver runs it without complaint. feaX rebuilds the model, solves it a second time, and the reactions disagree by 40%. The report points at line 212.
read the full sample report →FIGURES ILLUSTRATIVE UNTIL THE BENCHMARK SET IS PUBLISHEDInstall once, connect your assistant, keep working the way you already do.
$ [install command]$ feax connect [assistant]› check this model against a second solverSeven years of core library. The product on top of it is being built now, in the open about what works.
WE PUBLISH WHAT WE MISS. ONE BENCHMARK, A CONTACT TEST, IS CURRENTLY OPEN ON OUR LIST AND STAYS THERE UNTIL IT IS FIXED.
Run early builds on real models.
- Early builds, direct line to the founder
- Your models never leave your machines
- Fixed seat price for two years after launch
- In return: real models, a monthly call, a quote when it earns one
Every release, every benchmark, every miss.
We publish the benchmark set and its results with each release, including the cases we miss, and write up what we learn. A short note when something ships, nothing else.
Get release notes