Navier launches Prism: AI agents drive CAD, meshing and twelve solver families from one desktop environment

Hardware engineers gain a single environment driven by AI agents, so modeling, simulation, post-processing and reporting no longer sit in tools that cannot talk to each other. A single design study previously crossed more than half a dozen tools, including a CAD package, a mesher, a solver, a post-processor, Python or MATLAB and PowerPoint, and every handoff was manual export, re-import and repair. Navier's example is a drone cooling module: given only a CSV of design constraints, the agent proposed candidate configurations, built parametric CAD, ran CFD in OpenFOAM to check airflow and cooling, ran modal analysis in CalculiX to find natural vibration frequencies, and reviewed the design for manufacturability before writing the engineering report.

Because one workspace holds CAD, meshing, multi-physics solvers, code and reporting, the agent can execute a full workflow end to end from requirements to final report. The solver was rarely the bottleneck; the human loop around it was, and Navier's earlier physics-ML surrogate models were fast but not general, since every new problem needed new training data that cost more to generate than running the simulations. Prism runs twelve open-source solver families locally, including OpenFOAM, CalculiX, code_aster, SU2, Elmer and Project Chrono, alongside the in-house GPU finite-element solver Ferro; CAD is built on the Open CASCADE kernel and every run is tracked in the integrated Solver Desk.

The agent reports what it ran, what it assumed and how it checked the answer, and every script, mesh and solver input remains a file in the project folder. Engineers are right to distrust numbers they cannot verify, and no agent will be perfect. Any number in the report can be traced back to the run that produced it, and when the agent is wrong the location is visible; users can ask for a plan before the agent touches anything and review its work as they would a colleague's.

The demonstrated scope is the Prism desktop app itself: twelve open-source solver families and the in-house GPU finite-element solver Ferro run locally, CAD is built on the Open CASCADE kernel, and plugins for Onshape, GitHub, Jira, Confluence and Notion ship with it, while users can add their own MCP servers or write skills. The one complete example given is a drone cooling module, taken from a CSV of design constraints to an engineering report.

The release states that Prism will still make mistakes and that there is plenty it does not do yet, and that it is being improved continuously. No quantitative comparison against manual workflows and no independent evaluation results are reported.