Adam is a significant player in the 'vertical agent' category, specifically targeting high-precision mechanical engineering. Within the AI agent stack, they occupy the application layer but distinguish themselves by building deep integrations into specialized software (CAD) rather than operating in a general browser or desktop environment. Their work on geometry indexing is a notable contribution to the problem of 'grounding' agents in non-textual, spatial environments.
For the broader ecosystem, Adam demonstrates how agents can solve long-horizon tasks that require multi-step planning and error recovery. By moving from simple text-to-3D generation to parametric CAD manipulation, they are pushing the boundaries of what agents can achieve in physical product design. Their approach suggests a future where professional tools are controlled primarily through intent-based agents rather than manual UI manipulation.
Adam is an AI agent designed to operate within Computer-Aided Design (CAD) environments. While the first wave of AI in 3D focused on generating meshes for gaming or visual arts, Adam targets the more rigid world of mechanical engineering. Traditional CAD software like SolidWorks or Onshape requires engineers to manually manage thousands of granular operations—fillets, chamfers, extrusions—organized in a sequential 'feature tree.' Adam uses a specialized agent to execute these tasks through natural language, acting as a collaborative partner that understands the underlying geometry rather than just the visual representation.
To move beyond the hallucinations common in general-purpose large language models (LLMs), Adam relies on a custom approach to spatial reasoning. The company maintains a geometry index that records vertices, edges, faces, and bodies in a structured JSON format. This allows the agent to identify specific entities within a 3D model with precision. By pairing LLMs with these search and selection tools, the system can perform complex operations like merging duplicate features to stabilize a design tree or cascading parametric variables throughout a design. The model is trained on real CAD sessions to learn how to plan and recover from errors, mimicking the workflow of a human engineer.
Founded in late 2024 by Zach Dive and Aaron Li, Adam emerged from the UC Berkeley Master of Design program. The founders recognized that hardware engineering was being held back by software interfaces that had not fundamentally evolved since the late 1990s. The company gained significant traction during the Y Combinator Winter 2025 batch, subsequently raising a $4.1 million seed round led by TQ Ventures. The team is primarily based in the San Francisco Bay Area, focusing on recruiting engineering talent capable of bridging the gap between machine learning and computational geometry.
Rather than attempting to force users to abandon their existing tools entirely, Adam launched as an extension for Onshape, a popular browser-based CAD platform. This allows engineers to use the agent within their current workflows for specific high-value tasks like part editing and selection context. The company also offers a standalone web application for more creative generations. Pricing is structured as a SaaS model, with a standard tier for individual makers and a pro tier for professional engineers that provides access to more advanced models and unlimited generations. By targeting hardware startups and 3D printing enthusiasts first, Adam is building the data and feedback loops necessary to eventually challenge the professional engineering software incumbents.
An AI agent for computer-aided design that automates part editing and feature tree optimization.
Adam is hiring.