When Adam AI Labs dropped the source code to its text-to-CAD platform onto GitHub on June 17th, the response was swift—and revealing. By mid-June, CADAM v0.3.0 had racked up 4,500 stars and more than 560 forks, the kind of velocity that suggests developers had been waiting for exactly this sort of release. Or perhaps they were just curious what the Y Combinator Winter 2025 cohort company had actually built behind the slick demos.
Either way, the GPL-3.0 licensed repository transformed what had been a closed prototype into something anyone with an internet connection could poke at, modify, or run on their own servers. For a two-person startup trying to crack into a CAD market dominated by Autodesk and PTC—companies that have spent decades building moats around professional design workflows—open-sourcing your core technology is a gamble. Zach Dive and Aaron Li, the UC Berkeley design grads behind Adam, are betting it pays off.
CADAM isn't trying to be AutoCAD. It's a parametric CAD editor that ingests natural language prompts and spits out OpenSCAD code, which then renders as browser-based 3D models you can actually manipulate. Type "planetary gearbox" or "V8 engine block," and the system generates editable code complete with adjustable parameters: bore diameter, gear tooth count, wall thickness. The output files export to STL, SCAD, and DXF formats—standard fare for anyone who's ever sent a model to a 3D printer or machine shop.
What makes the timing interesting is that Adam isn't alone in chasing the text-to-CAD concept. PTC rolled out something called AI Advisor for its cloud platform Onshape earlier this year. Autodesk has been threading AI Assistant features through Fusion 360. A startup called Zoo.dev shipped Zookeeper—a conversational CAD agent—several months back. The industry, in other words, has decided that natural language interfaces are the next frontier, and everyone's scrambling to ship something before the window closes.
The Open-Source Play and the Paid Product
Adam runs a two-track strategy, and the distinction matters. CADAM, the thing they released in mid-June, is free. GPL-licensed, browser-based, modify-it-however-you-want free. The actual business lives in Adam Copilot, a paid AI assistant that integrates directly into professional CAD platforms. The copilot runs inside Onshape currently, with Fusion 360 and SolidWorks integrations listed on the company's roadmap—though how much of that functionality exists in production versus on a slide deck is harder to pin down.
Pricing for the commercial product starts at $40 a month for a Starter tier and climbs to $1,000 monthly for a Max plan. What do you get for that? Bill-of-materials cleanup, engineering change order drafting, supplier comparisons, context-aware part selection. Adam frames it as "an AI workspace for hardware teams," which is either aspirational or accurate depending on which features have actually shipped. The company put out a call for Onshape beta testers three months back, and forum chatter confirms the extension is real. Feature parity across platforms, though? That remains an open question.
The open-source release, then, is part marketing play, part community-building exercise. Give away the browser toy, build trust with engineers who reflexively distrust black-box AI systems, and hope some percentage of them convert to paying customers when they need functionality that actually integrates with their existing CAD stack. It's a strategy borrowed from the developer tools playbook: Vercel CEO Guillermo Rauch compared Adam to "the v0 of CAD" last fall, a nod to Vercel's own text-to-UI product that generates code instead of opaque design files.
Under the Hood: WebAssembly, React, and LLM Routing
CADAM's architecture is relatively straightforward for anyone who's built a modern web app. The platform compiles OpenSCAD to WebAssembly and handles rendering through Three.js and React Three Fiber. The stack leans on React 19, TypeScript, and Supabase for authentication and storage. Earlier release notes mention Anthropic's Claude API routed through OpenRouter, which suggests the large language model layer is modular—swap in a different LLM if Claude's pricing gets unworkable or a better model ships next quarter.
Keeping operations client-side wherever possible means faster iteration and no cloud rendering bottleneck for basic geometry. It also means the heaviest computational work happens on the user's machine, a practical choice for a startup trying to keep infrastructure costs manageable while scaling.
The GitHub README showcases example outputs—turbofans, gear assemblies, engine blocks—with orbiting 3D previews and exposed parameters. It's conceptual design territory. OpenSCAD, for all its strengths in parametric modeling, lacks native STEP export, the file format most manufacturers actually require. So CADAM's outputs won't slot directly into typical product development workflows without conversion steps, which limits its utility for anything beyond early-stage prototyping or hobbyist projects.
Whether that's a fundamental limitation or just a current gap depends on Adam's willingness to support additional CAD kernels down the line.
Traction, Or At Least GitHub Metrics
Measuring early-stage traction for an open-source project is messy. Adam's Hacker News launch generated attention, though vote counts fluctuate and older posts fade. A "Show HN" from roughly nine months prior—pitching the text-to-CAD concept—earned 179 points. A third-party review on May 24th pegged the repository at 3,523 stars; by mid-June that figure had jumped to 4,500.
The company claimed tens of thousands of individual users as of late October 2025, along with paying customers on its commercial plans, though those numbers came when Adam announced a $4.1 million seed round led by TQ Ventures, with participation from 468 Capital, Pioneer Fund, and angels including PostHog's Tim Glaser. Seed-stage metrics are notoriously slippery—user counts might include anyone who opened the browser app once, and "paying customers" could mean anything from five enterprise contracts to fifty individual subscriptions.
What's clear is that the open-source release generated momentum. Whether that translates into sustained community contributions or just a spike of forks that sit untouched in GitHub profiles will become evident over the next several months.
The Broader Industry Shift—and Its Limits
The CAD world is midway through an awkward pivot. Cloud-native platforms like Onshape have APIs and data architectures designed for extensibility, which makes them natural hosts for AI agents. Legacy desktop software like SolidWorks, which still commands huge market share, requires plugin workarounds and offers limited access to internal feature trees. Retrofitting AI onto decades-old codebases is harder than building for it from scratch.
Academic papers published over the past year argue—sometimes bluntly—that current text-to-CAD systems remain far from production-ready, particularly when designs need revision histories, tolerance chains, or manufacturing constraints baked in. A parametric model generated from a prompt might look impressive in a demo, but what happens when a design engineer needs to adjust it three months later? Does the code stay maintainable, or does it collapse into an unreadable mess?
Adam's argument is that exposing the parametric layer—making the code visible and editable—gives engineers more control than black-box geometry generation ever could. It also puts the company on the hook for producing clean, maintainable OpenSCAD scripts, not spaghetti code that works once and breaks on the second edit. The open-source release invites scrutiny on exactly that point, and developers who fork the repository will discover quickly whether Adam's generation quality holds up under real-world use.
What Happens Next
For now, CADAM represents something specific: proof that a small, venture-backed team can ship a functional text-to-CAD system and release it under an open license without immediately collapsing under the weight of enterprise expectations. Whether it becomes a genuine wedge into professional workflows or remains a clever marketing layer for the paid copilot depends on what happens beyond the initial GitHub burst.
The repository is live. The code is open. And judging by the forks, developers are at least curious enough to look under the hood. In an industry where "AI-powered" has become shorthand for "we added a chatbot," Adam's willingness to show its work might matter more than the technology itself. Then again, it might not. The CAD market has a long history of promising revolutions that quietly fizzle once engineers try to use them for actual work.
Time, as they say, will tell. For now, the source code is out there—and anyone with opinions about parametric modeling and large language models can dive in and decide for themselves.
