Prism by Navier AI

• Turning a CSV of requirements into candidate designs for a drone cooling module, with parametric CAD, an OpenFOAM airflow and cooling check, a CalculiX modal analysis and a written report
• Running a structural design study that iterates a parametric bracket to the lightest design that passes its static and modal margins
• Pointing the agent at an existing OpenFOAM, SU2 or code_aster case directory to install the runtime, run it, debug a failed solve and plot the results
• Comparing a batch of runs, such as cooling layouts or composite layups, in a run-explorer dashboard and summarizing them as a trade-study PDF
• Writing data-reduction scripts for test data, or embedded firmware for a prototype, alongside the CAD and simulation files in one project
• Reviewing a KiCad or Altium schematic or board with the agent before fabrication, checking nets, stackup and design rules

Prism takes the coding-agent pattern and gives it the tools a hardware engineer actually uses. Navier moved to this approach after its earlier physics-ML surrogates proved too narrow. The agent drives solvers engineers already trust, and leaves every mesh, input deck and script on disk where a reviewer can check it. The breadth is real: one window covers code-based CAD, linear FEA, CFD, electromagnetics, circuits and report writing, and skills and MCP let a team encode its own conventions. The limits are just as clear from Navier's own docs. CAD is generated in build123d rather than in the SolidWorks or CATIA files most mechanical teams work in. The built-in Ferro solver is linear, solid-only and unforgiving about units. The agent needs Navier's cloud backend even though solves stay local. And at $500 a seat a month it has to save real engineering hours. It fits best where a team already runs OpenFOAM or CalculiX by hand and wants to run far more variants. It is a weaker fit where the work is driven by a commercial CAD/CAE suite or must stay air-gapped.

About Prism by Navier AI

Navier AI, a San Francisco startup founded by former SpaceX engineers Cameron Flannery and Evan Kay, launched Prism publicly on November 3, 2026, pitching it as for hardware what coding agents have been for software. In concrete terms Prism is a native desktop application (macOS on Apple Silicon, Windows and Linux on x86-64) that puts an agent chat, a code editor and terminal, a 3D viewport and a local solver runtime in one window, attached to a project folder on your own disk. You describe the task, such as a part with its loads and constraints, a CSV of design requirements or an existing OpenFOAM case directory, and the agent plans and does the work with tools specific to engineering. It builds geometry as build123d Python scripts on the Open CASCADE kernel and exports STEP files you can open elsewhere. It meshes with Prism's own OCCT-based mesher (or gmsh, which Prism can install). Structural cases go to Ferro, Navier's in-house FEM solver, which is compiled into the app and handles linear static, modal, buckling and prestressed modal analysis of 3D solids, switching to the GPU on large models when one is available. Everything else runs through the Solver Desk: a local daemon installs twelve open-source solver families as containers and tracks every run. They cover CFD (OpenFOAM, SU2, Basilisk), structures (code_aster, CalculiX), multiphysics (Elmer), discrete elements (Yade), multibody dynamics (Project Chrono), Modelica (OpenModelica), electromagnetics (Palace, openEMS) and circuits (ngspice). Results open in Prism's VTU and post-processing viewers, and an engineering-reports skill writes structural, CFD, trade-study or memo reports to PDF from Typst templates, while a presentations skill builds editable PowerPoint review decks. The agent can split big jobs across sub-agents, ask you questions mid-task, and run on a schedule. Every script, mesh and solver input stays a file in the project folder, and in a git repository every edit appears in a thread diff. Plugins add Onshape (STEP export into the project), Altium Designer (beta, Windows only), KiCad (beta), MATLAB and Simulink, GitHub, GitLab, Jira and Confluence, Linear, Notion, Slack, Microsoft 365 and more, and you can add your own MCP servers or SKILL.md skills to encode team conventions. Prism also reads KiCad and Altium schematics natively. The tradeoffs are what an engineering buyer will care about. Solves run on your machine, but the agent itself runs against Navier's backend, so the files it reads are sent to the model provider, sign-in is required to launch, and the agent does not work offline. You pick the model, from Anthropic and OpenAI models to DeepSeek, Kimi, GLM and Grok. CAD is generated in code rather than driven inside SolidWorks, CATIA, NX or Creo, with Onshape the only CAD system that has a documented plugin. The open-source solvers need Docker, Podman or Colima. Ferro is linear only and strictly SI with no unit conversion, and the docs say plainly that the agent will solve a wrongly specified case without complaint. Price is $500 per seat per month after a free account with a starter credit allowance, so Prism is a professional tool, not a hobbyist one. Navier engineers also sell scoped simulation engagements and forward-deployed engineering, which typically runs about three months.