Empowering OpenSCAD with Local Agentic AI Assistance

This proposal introduces an optional, privacy-first AI assistant for OpenSCAD to flatten the learning curve of SCAD syntax and geometric transformations. Integrated as a togglable dock panel, the assistant remains completely inert when closed and requires no cloud subscriptions or external API keys, instead leveraging locally running models via Ollama or any OpenAI-compatible server. Beyond standard Q&A, the project implements agentic behavior where the AI can analyze the active script, propose modifications through an inline diff, and apply approved changes directly to the editor as undoable actions that trigger an automatic preview. The architecture is designed to be vision-model-ready from the start, ensuring future viewport context integration requires no structural refactoring. This creates a transparent, local-only loop that empowers users without compromising the project’s core philosophy of user-controlled development.