ForgeCAD launches Agentic CAD: a workbench for
physical innovation and SimReady assets
ForgeCAD launches Agentic CAD, a workbench where AI agents turn requirements into editable parametric designs and refine them through checks and feedback.
Until now, ForgeCAD put programmable CAD in the browser and provided skills to use in Codex or Claude Code. You could write a model, change its parameters, inspect it, and export it.
But the product stopped at the workbench door. You still had to do the modeling work yourself.
Today, that changes.
ForgeCAD is launching Agentic CAD. It is a workbench where AI agents turn requirements into designs they can build, evaluate, and refine.
Describe the result. The agent does the CAD work.
Open a ForgeCAD Project and tell the agent what you want to build. Describe the part, product, mechanism, or simulation asset that you need. Add drawings, images, specifications, or existing model files when they matter.
The agent works inside the Project. It reads the source and references. It builds an editable, parametric ForgeCAD model. It runs the model, studies errors, renders views, inspects measurements, and revises the design.
You can answer questions while the agent works. You can pause it, continue it, or ask for another change. When the work is ready, you review the source and geometry before accepting it.
## Agentic CAD is a loop
Physical design rarely has one correct answer. A set of requirements can admit many useful solutions.
ForgeCAD joins both sides of that process:
1. State the goal and the constraints.
2. Let the agent build a candidate.
3. Run the model and inspect concrete evidence.
4. Give feedback or change a parameter.
5. Compare the result with the requirements.
6. Export, manufacture, simulate, or continue the search.
The first result is not the promise. The loop is the promise.
## Physical AI needs design at scale
Physical AI is central to why we are building ForgeCAD. Robotics teams need large and diverse asset sets for training, evaluation, and simulation. Producing and maintaining those assets by hand does not scale.
Physical AI systems learn by interacting with physically accurate simulated environments before deployment. They require 3D content with physical and semantic data, not only visuals. [SimReady](https://www.nvidia.com/en-us/glossary/simready/), built on [OpenUSD](https://www.nvidia.com/en-us/glossary/openusd/), defines shared requirements so assets are simulation-ready across supported tools and pipelines.
ForgeCAD turns one set of requirements into editable asset families with controlled variation and useful physical data. AI agents can generate and refine the assets, while checks keep the results tied to their intended use.
An agent can author a parametric robotics prop and its simulation data. ForgeCAD can export the selected variant as an OpenUSD package.
The launch uses the [NVIDIA SimReady Foundation](https://docs.omniverse.nvidia.com/simready/latest/index.html) Prop-Robotics-Neutral profile. ForgeCAD checks each exported package against the selected, pinned profile version. The saved result records the profile, validator contract, package identity, and complete rule outcome.
The verifier is also part of the agent's design loop. The agent can run it, read each finding, repair the parametric model, and check the new package. It fixes the source instead of patching one exported file. This feedback helps the agent produce higher-quality simulation assets.
## The same workbench can serve more applications
The launch also includes ForgeCAD's hosted render and export API. Applications and automated pipelines can request outputs from a saved Project revision without opening the editor.
One request can vary model parameters and produce PNG, STL, 3MF, STEP, GLB, or USD outputs. Each durable run records its inputs, status, files, and checksums. The ForgeCAD Orchestrator CLI uses the same API and run records.
This turns a ForgeCAD model into a reusable production capability. An application can request the next variant without operating a CAD server or rebuilding the workflow.
## This is version one
Agentic CAD starts with the work that is already running across ForgeCAD Projects, AI agents, checks, and exports. We will keep expanding what agents can build and what the workbench can evaluate.
Richer virtual tests will let us reject weak designs before downstream use. Physical results will help us correct those tests. The feedback can then help future agents propose better candidates with fewer attempts.
Our goal is not to automate the old CAD workflow one click at a time. Our goal is to make more physical ideas practical to explore.
If you build products, machines, or robotics assets, bring ForgeCAD a real requirement.
Start building at [forgecad.io](https://forgecad.io).