Protobench

Hardware work does not end when a shape looks right. Move a board or substitute a connector and the consequences spread: mounting, access space, enclosure openings, purchased parts, and documentation may all need attention.

Before making the next decision, the designer often has to reconstruct how those pieces relate. I started Protobench around that coordination problem: keeping design intent, editable source, and the evidence behind a decision close enough that the next revision remains understandable.

The starting point is small teams building products around a circuit board and an enclosure. It is a focused place to explore a much larger question: what should it feel like to design a physical product with an AI collaborator?

Competitive desk research and technical investigation moved the project beyond the assumption that AI-generated geometry would be enough. Many tools already address creation, editing, sourcing, or review. The unresolved opportunity was in the continuity between them.

The question became more specific: can a workspace make a consequential hardware revision easier to perform, explain, recover, and hand off? That changed the standard for product decisions. The correction effort after a first result matters as much as the speed of generating it.

These are working hypotheses being explored through research and implementation. The project is still developing the evidence needed to understand how well they translate to other people’s hardware workflows.

The model occupies the work surface. Conversation, selection, inspection, and project navigation sit around it. A person should be able to examine a part, ask about it, and understand the resulting change while staying close to the design.

That principle guided the interface experiments: compact panels, contextual controls, restrained color, keyboard interaction, and view state that can be recovered. Conversation is useful when it helps someone work with the artifact in front of them.

The real challenge is maintaining a clear relationship between what is requested, what is built, and what appears on screen. A polished viewport has to remain connected to the source that defines the product.

The Roboto Origin humanoid sample makes the relationship between parts and motion visible. Pose presets sit beside controls for the waist, crouch, arms, and elbows, so I can examine the assembly in different configurations.

This is a way to explore an articulated model inside the workbench. The robot model is RoboParty’s Roboto Origin, used here as a sample for inspecting assemblies and their controls.

The Agent Composer keeps an instruction next to the work it refers to. I can type a request, mention a part, attach pictures, and choose the agent settings while the model remains in view.

In this captured session, I asked the agent to explain the robot’s assembly and the relationship between its pose controls and joints. The sidebar shows the actual question, activity, and completed explanation.

The visual identity gives that same idea a recognizable form: four nested contours, connected layers, and the possibility of another iteration. The original Spectrum mark brings color to an otherwise quiet system of graphite, white, and precise typography.

The surrounding space matters as much as the mark. The product and its evidence need room to be inspected. This brand direction explores how a distinct identity can coexist with the clarity required of an engineering tool.

Protobench developed through several working environments rather than one uninterrupted implementation. Early industrial-design tooling explored how briefs, references, parameters, native geometry, and revisions could surround a continuous canvas. BetaBench investigated a local desktop foundation connecting mechanical and electrical source, generated artifacts, component identity, and an AI runtime.

Those experiments made a lasting distinction clear: a convincing preview and an editable engineering model serve different purposes. The product needs both, connected to the same project.

The current Electron alpha builds on the wider research while establishing an independent application foundation. Its recent work brings project lifecycle, CAD import and export, agent-driven source changes, history, drawing views, assembly communication, and a circuit-board workspace into that environment.

In the alpha, the AI works with editable project source and asks the application to build it. A geometry pipeline produces solid and display artifacts, which the application makes available for inspection. Work moves between a natural-language request, the parts on screen, and the source that generated them.

The application keeps files, runs, revisions, and exports connected, so the next iteration can build on work that is still understandable.

The broader vision includes coordinated, consequence-aware engineering across domains. The current alpha is the environment where the underlying workflows are being implemented and examined; that wider vision remains a direction for development.

Current alpha: Electron · React / TypeScript · Three.js · Python / build123d / OpenCascade · tscircuit · Codex

The environmental sensor sample gives the mechanical and electrical views a shared subject. Its enclosure can be inspected as an assembly, while the circuit-board workspace presents the board in 3D, as a technical layout, and as a schematic.

These screenshots document the current interface using a sample project. The board is marked unfinished in the application; the views show design inspection rather than evidence of manufacturing or electrical validation.

Experimentation depends on being able to return to work you understand. Project history captures source, generated artifacts, and presentation together. Conversation versions can retain different directions, and undo restores saved project work.

Recent development has focused on continuity: long tasks should remain observable, stoppable, and resumable. Authored colors should survive the transition from source to the rendered product. These details connect the promise of an AI collaborator to an experience a person can actually work with.

A product has to be understood by someone other than its author. Vector drawings turn geometry into legible technical views. Interactive assembly guides introduce parts in sequence and can be exported as a self-contained browser document.

The circuit-board workspace adds technical, 3D, schematic, and parts views to the same environment. Together, these workflows explore how a project can carry more of its own explanation from design into communication.

An assembly sequence is a digital explanation of a proposed design. Physical fit, manufacturing choices, and electrical behavior still require their own inspection and evidence.

The strongest lesson has been that useful autonomy depends on continuity: the right source, the right revision, a recoverable change, and an honest explanation of what remains unresolved. A convincing model is only one part of the experience.

Protobench is a desktop alpha in active development. My focus is to deepen the design process, improve reliable builds and fit inspection, and make the relationship between mechanical and electronic changes clearer.

The larger ambition is a tool that helps people move from an idea to an inspectable physical design while retaining control of the work.