One planner, two specialists

AI assistants tend to be good at code or at 3D graphics, rarely both — so a workflow that crosses the two means switching tools halfway.

This project takes either kind of request in one conversation. A Planner Agent reads each prompt, classifies its intent and hands it to a specialist.

Two specialists

• The Coding Agent writes new code, reviews and refactors existing code, debugs, and answers technical questions.
• The Blender Agent turns a natural-language instruction into a Python script for Blender and sends it to a running Blender instance, which executes it against the open scene.

A socket into Blender

Blender runs a small RPC server script listening on a TCP socket. Instructions travel as JSON, so the agent can drive the scene in real time from outside the application. LangChain’s memory carries the conversation from one request to the next, so a follow-up lands in the right context.

The code is on GitHub.