CAD Agent

**Judgement:** High-value tool for agents that need visual, parametric CAD iteration in a controlled environment. Best suited to containerized workflows where Docker is available and HTTP calls are easy to orchestrate. **What it does well:** - Runs all CAD, meshing, and rendering inside a Docker container via build123d + VTK, so the agent only sends code and receives JSON/PNG results. - Supports a rich workflow: model creation/modification, dimension queries, 3D and 2D renders (including multiview composites), export to STL/STEP/3MF, and printability checks. - Documentation clearly explains endpoints and provides concrete curl examples plus a build123d cheatsheet, which is very helpful for agent prompting. **Key risks / limitations:** - Requires Docker and proper networking; in nested/containerized environments, `localhost` access may fail and you must use `docker exec`, adding operational complexity. - Performance and robustness for large or complex models are not characterized; heavy geometry or many rapid iterations may cause slow responses or higher resource use. - Exposes a code-execution surface (build123d Python scripts) inside the container: safe in isolated setups, but should not be publicly exposed without standard network security controls. **Recommended scenarios:** - An OpenClaw or similar agent that needs to design and tweak 3D-printable parts based on user requirements, with tight visual feedback loops. - Workflows where you want to keep CAD tooling and dependencies out of the agent runtime and instead rely on a standardized HTTP CAD microservice. - Educational or prototyping environments where the agent can learn/teach build123d-style parametric modeling by iteratively modifying code and inspecting rendered results.