An AI-powered CAD harness called Adam has launched directly inside Autodesk Fusion and PTC Onshape, offering mechanical engineers a new way to edit feature trees with full control. The tool reads existing parts, understands the feature history, and makes agentic edits—without a black-box STL output.
“Serious mechanical engineers don’t want a black box that spits out an STL,” said Zach, co-founder of Adam. “They want help inside the CAD tool they already use, with full visibility and control over the feature tree.”
Background
Adam started as a text-to-CAD experiment on Hacker News that drew interest but revealed a gap. Users wanted integration with professional CAD software, not standalone web apps. The team pivoted to building a harness that plugs into existing tools.
The harness leverages Onshape’s FeatureScript and Python scripting in Fusion. Internally, Adam runs a benchmark across frontier models, noting significant improvements in spatial reasoning, especially with GPT-5.5 and Opus 4.7.
“We open-sourced our earlier text-to-CAD work to give back to the community,” Zach added. The company also praised Anthropic’s recent Autodesk connector as validation of the direction, but emphasized its own model-agnostic approach.
What This Means
Adam lets engineers automate mundane but critical tasks: merging redundant features, renaming features for readability, applying uniform fillets, and parametrizing models. It can also generate CAD from scratch.
By operating inside the native CAD environment, Adam keeps engineers in their workflow without switching to a separate platform. The company says it actively integrates with multiple CAD programs and remains model-agnostic, choosing the best frontier model per task from its internal benchmark.
“We pick whichever frontier model is winning on each task type instead of being tied to one lab,” Zach said. This differentiates Adam from Anthropic’s recent connector, which is locked to Claude.
The harness is now in beta for both Fusion and Onshape, with install links available. Engineers can try it today to see the feature tree transformed in real time.