From drafting to virtual testing, AI for design offices

The Prism platform brings together part design, physical simulations, and reports, with agents capable of orchestrating these steps.

In a Navier AI demonstration, an agent designs a cooling module for a drone, runs simulations, and writes a report on the result. This work usually spans several software tools, with files to prepare and parameters to carry over between each step. The Prism platform brings these operations together in a desktop application: the engineer describes the expected work, then the agent uses the tools needed to carry it out.

The public launch announced on November 3 concerns Windows, macOS, and Linux. The application combines computer-aided design, 3D visualization, physical computation, and code editing. A free starter option is offered, while the pricing page shows a plan at $500 per user per month. The promise concerns automating operations between the initial request and the results of a study. Carrying the same request through the tools

The issue being targeted is the repeated handling between two versions of an object. According to Navier AI, changing a part often requires cleaning up its geometry, remeshing it, restoring the calculation conditions, and transferring the charts into a presentation. Meshing consists of dividing the geometry into small elements on which the software can perform its calculations. A design change can therefore require a new analysis setup.

The Prism presentation describes an agent capable of producing an editable geometry from the dimensions, loads, and constraints provided by the user. It then prepares the studies, launches the calculations, and turns the outputs into charts or tables. The request may concern a part to design just as much as a study to carry out from an existing project.

In the case of the cooling module, the founders describe a flow simulation with OpenFOAM and a vibration analysis with CalculiX, followed by a manufacturability review and a report. This is an example presented by the company. It illustrates the sequence of operations, without establishing a measured time savings across all engineering projects. Simulations run through identified calculation engines

The agents control specialized tools to study fluids, structures, heat transfer, or electromagnetic phenomena. The catalog includes twelve families of external solvers, including OpenFOAM, CalculiX, codeaster, and Elmer. Added to this set is Ferro, Navier AI’s integrated structural calculation engine. The announced coverage therefore depends on several software tools, each associated with types of analyses.

The solver execution documentation states that the external tools are installed on demand in containers. A container engine, such as Docker or an equivalent, remains necessary. The application oversees their launch and gathers the calculation logs, in order to track what was executed and identify any failure.

In this local configuration, calculations continue even if the Prism window is closed. A remote calculation option also exists for organizations to which Navier has enabled it. File transfers and launches on these machines are then subject to specific permissions, with the resources and hourly rate shown before execution. Keeping the files on your computer does not make the agent offline

The FAQ distinguishes project storage from the operation of the AI. Geometries, meshes, and results reside on the user’s disk. However, the content the agent reads to respond is sent to the model through Navier AI’s service. An analysis performed with local files therefore involves exchanges with a remote service.

A connection is also required to sign in at launch and to use the agent. Once access is established, the editor and visualization tools can work without a network. The supported versions are macOS on Apple Silicon, as well as Windows and Linux on x86-64 processors. These conditions clarify what the desktop application covers, beyond the phrase “your files stay on your machine.”

Control over actions depends on the chosen permissions. Manual mode asks for validation for the relevant modifications; the initial setting automatically accepts file creations and edits, while still keeping prompts for certain commands. The ability to review a step therefore exists, but it does not mean that every operation is systematically subject to approval. Comparing versions and keeping assumptions

The agent’s work leaves behind viewable files: scripts, solver inputs, meshes, and results. On a project tracked with Git, a view shows the changes made to the files. The review documentation emphasizes the numbers, units, and configurations used by the calculation engine. An incorrect boundary condition can produce a seemingly plausible result, even if the software runs correctly.

Several agents can be launched simultaneously to explore variants, prepare a thermal study, or draft a design review. The company presents this organization as a way to examine more solutions. The validity of a simulation remains tied to the physical model and the assumptions retained; keeping the files gives the engineer the elements needed to inspect them.

The results can then be assembled into a PDF or a PowerPoint presentation, using the team’s document templates. The proposed connections also cover the tools already in use, including Onshape for design, GitHub and GitLab for code, or documentation and messaging services. The Altium Designer integration, intended for electronics, is marked as beta.