Suzanne documents a 3D generation API for Claude Code agents

Why it matters

Suzanne gives coding agents a documented route from text or photos to downloadable geometry, testing whether AI-assisted 3D generation can advance from visual assets into product-development workflows.

Samantha J. Brown and Yehor Ivanenko have spent the summer turning Suzanne, a student hackathon project, into a browser-based 3D editor that generates models from text, sketches and photos. The product's Claude Code documentation describes an API workflow through which coding agents can request 3D models, poll generation jobs and download the resulting files.

The available primary sources establish Suzanne as a named product. Its Devpost entry, product page and API documentation all use the Suzanne name, while Founders, Inc. lists Suzanne among its Canopy teams. Those sources do not identify a registered legal entity, and no authoritative evidence reviewed for this article links Suzanne to Prometheus, the industrial AI company co-founded by Jeff Bezos and Vikram Bajaj. Prometheus' financing, founders and operations therefore should not be attributed to Suzanne.

The API documentation does not provide a public availability date. Suzanne's agent capability is presented here as an undated, documented feature rather than a new launch.

Four hackathon creators, then a two-person push

Devpost identifies Daniel Yi, Ivanenko, Brown and Ruslan Akmyradov as the four creators of Suzanne. It records a November 8, 2025 project start and says the team submitted the project to HackPrinceton Fall 2025. The entry does not establish a formal company founder roster.

Brown and Ivanenko later moved to San Francisco after being accepted into Canopy, the Founders, Inc. program, according to Ivanenko's account of the program. Founders, Inc.'s Canopy directory lists Suzanne at Booth 33 and describes it as a tool that turns an idea into a 3D model in minutes. The San Francisco claim reflects the creators' account of where they worked during Canopy; it does not establish a corporate headquarters.

Ivanenko studied at the University of Indianapolis and joined Carnegie Mellon University's Security and Privacy Undergraduate Research program, according to his LinkedIn profile. Brown's public account centers on moving the project from a hackathon prototype toward a paid product.

Self-reported users, customers and revenue

Ivanenko said Suzanne released a beta during its first week at Canopy and attracted 120 users in its first 24 hours. He later reported that the product had reached 250 users, signed its first paying customers, closed an angel round and begun work toward an API contract within five weeks, according to his Canopy retrospective. These figures and milestones are creator-reported and have not been independently corroborated.

Brown separately described Suzanne's progression as going "from a hackathon project to monthly revenue in 6 weeks". Her accessible account says the team reached hundreds of users and signed paying customers, but it does not give a revenue figure. Suzanne has also not disclosed the angel round's amount, valuation or investors. Public materials reviewed for this article do not establish API pricing.

From prompts to editable geometry

Suzanne's browser product accepts text, photographs and sketches, then lets users request changes through chat. Its editor includes primitives, Boolean operations, mesh tools, materials, parametric sliders and reversible history.

The product page lists STEP, STL, GLB, OBJ, OpenSCAD, 3MF, 3DM and USDZ among its exports. It also advertises printing and shipping of physical prototypes. Those production services are company claims; Suzanne has not named manufacturing partners or published evidence of production deployments.

The Atelier2 documentation describes three models. Sculptor is positioned for faster asset generation, Atelier for higher-detail textured models and Capture for reconstruction from four photographs. Developers can request meshes with 200,000, 500,000, 1 million or 2 million polygon faces. Quad topology is capped at 150,000 faces, and documented API outputs include GLB, OBJ, STL and FBX.

The same documentation gives a typical generation time of 30 seconds to four minutes. That range is Suzanne's stated latency, not an independently measured benchmark. The company has not published third-party comparisons covering geometry quality, dimensional accuracy or manufacturability.

Giving coding agents a 3D workflow

Suzanne's Claude Code integration guide explains how Claude Code and other large-language-model coding agents can inspect a repository and build a typed client around text-to-3D and photo-to-3D jobs. The documented workflow covers job submission, status polling and downloads in GLB, OBJ and STL formats. The guide does not establish specific support for Codex or Hermes.

Separate Python and Node.js recipes show text-to-3D and four-photo generation using bearer-key authentication. This gives developers a way to place Suzanne's generation service behind games, design tools, configurators or prototyping applications without requiring users to work inside its browser editor.

The physical-product test

Suzanne presents its editor as a path from an initial idea to an editable model that can enter product development. Physical production imposes requirements that a polished render can conceal. Dimensions, topology, tolerances, materials and manufacturing methods determine whether generated geometry can become a usable object.

Adjacent companies address other parts of that process. MakePhysical markets geometry checks, product decomposition and production through 3D printing, CNC machining, injection molding and electronics assembly. EaselAI combines AI-assisted design with pricing and supplier selection, while ThingSpin focuses on editable parametric CAD and manufacturability checks.

Suzanne's documented approach combines conversational 3D creation, browser editing and API access. Its creator-reported paying customers and monthly revenue provide early commercial evidence, though no amounts or customer names are public. The harder evidence remains absent: independent output benchmarks, named manufacturing partners and case studies showing generated models moving through production.