TLDR: My friend and I built an AI assistant that models with you directly in Rhino (via plugin with native chat). No rendering painted over your model, no baked meshes imported from the outside. Check out neospline.ai to learn more and see its full capabilities.
Hi everybody, my name is Simon. I am not an architect, but my girlfriend is. I’ve noticed that she keeps complaining about the same stuff at work:
-
scattering natural vegetation to precision-aligning hundreds of furniture blocks
-
“small” adjustments that ripple across the entire file, needing the “same” click hundred times
-
spending hours moving around geometry manually to explore “what-ifs” or to see options next to each other
-
cleaning up curves, verifying the model against project or legal constraints, etc
So I talked to a bunch of her architect friends, and turns out there are many such workflows that they would happily reduce in favor of “real architecture”.
As a software dev who replaced similar “chores” in my work with AI, I was wondering if AI can do the same for people using Rhino. Keep the creative and fun, while offloading the boring repetitive parts to the AI assistant.
You can see the AI assistant in action here.
Or watch a short version here:
I know that there’s already RhinoMCP, but I also know that “just” an MCP cannot really take the full advantage of current AI capabilities on its own (full comparison here). AI is not trained to understand 3D; the spatial data is too verbose, and it bloats the context(and cost) like crazy. It’s hard to force the AI to verify everything it’s modeling, etc.
Also, it’s really hard to give the AI the right context so it does not get confused about what the task is and which parts of the file it’s supposed to edit.
So, my friend (AI engineer) and I built an AI agent that lives inside Rhino 8 as a plugin, and makes changes to your model directly, using the same tools as you would. You can seamlessly transition between AI-prompting and manual modeling without leaving the viewport or losing a single curve.
The agent:
-
allows you to @mentionmention specific layers, objects, block/block instances in the prompt, attach images/viewport screenshots, so you can give him the task and context precisely
-
understands the file without loading a ton of useless data into context
-
makes changes (that you can adjust/undo) directly in the file
-
notices the changes you make while he is working, and adjusts his plan accordingly
-
automatically checks for collisions, alignment, and self-corrects without your intervention
-
only uses the allowed set of Rhino tools, no arbitrary (unsafe) Python or C# code
-
proactively and intelligently compacts and cleans up the context (messages, spatial data, command usage, etc.) to reduce cost and likelihood of confusion
He’s still messing up sometimes, but there are already many use cases where he’s doing the work reliably and much faster than doing it manually.
We’re currently looking for more people interested in Early Access (offered at token+infra cost) - testing this out and giving us feedback that can help the agent become more reliable, faster, and more token-efficient. Every week, we make great jumps in capabilities and efficiency, and we want to keep improving 
If you are interested - ARCHITECT OR NOT - please let me know here on our site: neospline.ai