After a long period of testing and iteration, I’m ready to share a first look of something new.
ARIMA: Artificial Intelligence Modeling Assistant is my experimental plugin that explores how AI can generate simple 3D geometry from text or images, directly inside Vectorworks.
The idea of integrating artificial intelligence into Vectorworks started about a year after the release of ChatGPT. At the time, I had to put it on hold:
- models weren’t reliable enough
- APIs were too expensive
- building something sustainable felt unrealistic
Even today, things are evolving fast — new models appear constantly, and what works today may be outdated in a few weeks. What made ARIMA possible now is OpenRouter: a single API key that gives access to many different AI models, both free and paid.
Through OpenRouter, you can access multiple AI models from one place, using free models with daily limits. You can try more advanced models by adding credits — even $5 can go a long way and let you run a surprising number of experiments.
At the moment:
- Free usage allows around 50 requests per day
- If you purchase at least 10 credits, your daily limit is increased to 1000 free model requests per day
Not all free requests succeed (server load can cause errors), and performance and cost vary significantly between models.
Model selection: you can select and use AI models directly inside Vectorworks.
- You can filter for free and image-capable models
- You can enable and show reasoning models
- The plugin shows only recent models
- Models are sorted by context (high to low)
- Models are limited to a manageable list (max 25)
To reduce unnecessary API calls, model lists are cached locally and updated every 30 minutes.
After you selected a model, the workflow is very simple:
- If you selected a model that supports images, you can paste an image URL
- Use the “Write Prompt” button and adjust the prompt according to your needs
- Wait for the response from the selected model. Response times can vary and may be slow, especially with complex prompts or image
ARIMA is built around a simple idea: generating 3D geometry from descriptions or images by leveraging Vectorworks tools.
Current capabilities include:
Create simple 3D geometry from text descriptions and images (depending on the model’s capabilities)
Use materials from the Resource Manager
Use a single guide geometry (you can assign the name “marked” to a polyline)
Read all named objects in the active layer (circles, rectangles, polylines, polygons, lines)
Insert symbols by specifying the exact symbol name in the prompt
Token usage and cost of the last prompt
Results depend heavily on the model you choose. Some free models already produce interesting results. More advanced paid models can significantly improve output quality — but also increase cost. This balance is constantly changing. A model that is “best” today might be surpassed very soon.
ARIMA is still experimental, and it’s important to be clear about what it cannot do.
- It cannot fully read your entire drawing
- It cannot modify or delete existing objects
- It works within strict context limits imposed by AI models
Sending too much data (e.g. entire layers) would:
- Increase costs
- Reduce output quality
- Increase the likelihood of errors
Because of this, ARIMA focuses on generation, not full scene understanding. ARIMA is not a finished product: it’s an ongoing exploration. Some results are surprisingly good. Others are inconsistent or fail entirely. That’s part of the process!
- How quickly ideas can be prototyped
- How new workflows might emerge inside Vectorworks