RoboRigs/Robotics & hardware design
From a sentence to a 3D-printable robot.
An agentic design workspace that turns a plain-language robot idea into concept art, CAD, printable parts and validated 3D meshes — driven by a chat that can operate the tools.
Design pipeline
A described concept becomes parts, checks and a printable mesh.
- Sector
- Robotics & hardware design
- Year
- 2026
- Disciplines
- Agentic workflowGenerative designCAD & meshMCP tools
The problem
Turning a rough hardware idea into something you can actually print involves a chain of specialist steps — concept design, CAD, part layout, printability checks, mesh export — each in a different tool. For a non-expert, the gap between “I want a boxy robot with an AI HAT” and a validated print plate is enormous.
The approach
We built a single workspace where the whole pipeline is driven by conversation. Describe the robot; the system generates concept images, lets you breed and refine them, then hands a chosen concept to CAD.
Behind the chat, a set of MCP tools do the specialist work: generate concepts, produce CAD and views, measure parts against the printer bed, run validation checks and export meshes. The agent orchestrates them, and the workspace visualises each stage — concepts, assembled and exploded views, print plate and mesh.
Validation is explicit. Parts are checked for fit and interface tolerances, and findings come back as pass, warn or fail with fix hints — so a design converges on something printable rather than merely plausible.
What we built
- 01
Conversational design
Describe a robot in plain language and iterate on it through chat, with the agent driving the tools.
- 02
Concept generation and breeding
Generate concept images, select the promising ones and refine them across generations.
- 03
CAD handoff
Turn a chosen concept into parametric parts and rendered assembled and exploded views.
- 04
Bed-fit checking
Measure every part against the printer build volume so the design stays printable.
- 05
Validation findings
Interface and tolerance checks returned as pass, warn or fail — each with a fix hint.
- 06
Mesh export
Generate and preview 3D meshes and a print plate, with artefacts kept in the project.
Technology
- SvelteKit
- Shared chat client
- MCP tools (CAD, mesh-gen, concept-images, rig, parts)
- Python services
- Nx monorepo
Outcome
A design studio that takes a spoken idea through concepts, CAD, validation and mesh export — keeping every intermediate artefact, from concept images to meshes, in one project the agent and the user share.