Welcome to the Onshape forum! Ask questions and join in the discussions about everything Onshape.
First time visiting? Here are some places to start:- Looking for a certain topic? Check out the categories filter or use Search (upper right).
- Need support? Ask a question to our Community Support category.
- Please submit support tickets for bugs but you can request improvements in the Product Feedback category.
- Be respectful, on topic and if you see a problem, Flag it.
If you would like to contact our Community Manager personally, feel free to send a private message or an email.
Best Of
Re: [Chatbot Copilot] - We Built a FeatureScript AI Assistant That Outperforms ChatGPT
I do AI at my "real" job. You will always have issues trying to train FeatureScript on a very large LLM.
Think of how these language models work: they learn relationships between words by representing them as points (vectors) in a multi-dimensional space. Words that often appear together or have similar meanings are placed closer to each other. When predicting the next word, the model looks for words whose vectors are "close" (short vector distance) to the current context. However, because training data specifically for FeatureScript is relatively limited or "sparse" compared to general language, the relevant FeatureScript concepts might be represented by vectors that are farther apart. When the model has to "jump" a larger distance in this vector space to find the next likely word, there's a higher chance it might land on something plausible-sounding but technically incorrect or nonsensical in the specific FeatureScript context – essentially, a hallucination. So, sparser data can indeed lead to larger effective distances between relevant concepts, increasing the risk of inaccurate predictions.
Professionally, I specialize in smaller local models, typically no larger than 32 gigabytes. I train them for specific use case in a variety of ways.
I am working on a FeatureScript AI -but I am using smaller models that run locally to decrease the noise.
All LLMs I have tried halucinate badly in FeatureScript, GEMINI is the worst - I do mean worst, it has the idea that the current FeatureScript API is outdated and whatever it is using is the correct one and will repeatedly halucinate even when told it is wrong with souces, links etc. I gave up.
My FeatureScript workflow
I use Claude to do daily work and to help with my local LLM project.
Sometimes I will "lean in" to the hallucination. If I get an hallicination that I think SHOULD exist, then I work with the LLM to create it and add it to my utilities.
When I need a utility function I ask Claude and ChatGPT to create it with the same prompt and pick the better candidate, some time I will ask DeepSeek if I am not happy with the other answers.
Also I will ask one AI to "Fix" another AIs output - this works suprisingly well.
The OnShape AI assistnat will haluicinate as well, but it does a good job at finding examples and documentatin for a method.
Ping me and I will try to help anyway I can.
I wish I could use VS Code for this work! Onshape HINT HINT.
Gary
Re: How do I move an object to a specific coord
Have you tried the learning center for the sketching tutorial?
Mostly draw things where you want them to be. You could just drag a corner or point to the origin and it would snap there. If your sketch was constrained properly it would just move.
Re: Investing in Onshape's Foundation: Focus on Stability and Performance for the Next Two Releases
This sounds fantastic. And what an excellent example of good communication with the Onshape community. Ty for the update!
Re: How to get a derived part to automatically change based on existing variables
You cannot do this with standard variables - create a configuration variable in the source part studio, then in the derived dialog you'll see an option to set the diameter (which can be linked to a variable in the target part studio, but it doesn't have to be).
Re: Custom Feature: Cable/Wire Routing
Changelog for release 1.197.1
⭐ Added wire color to wire data table
⭐ Added 4/0, 3/0 and 2/0 nomenclature to AWG size in the Wire data CSV parsing function
Re: Imported an STL part but cannot design on it
create a plane with 3 point method. Sketch on the plane
MDesign
Re: Multi-Part Part Studios no correct answer
Hi all,
There is the link to the doc https://cad.onshape.com/documents/566594ae9320ff8568f26a47/w/5d18b9dd7e3a85bf4559f4db/e/a6d66a8eeea55b13fd23ec46?renderMode=0&uiState=68131b72e7ee437bac11edd9
(By the way, I just wanted to point out, that the exercise could have wrong answers.)
Fix text bug with unicode symbols like "✱" or "⬣"
Strangely, the text preview for these works fine, but when placed in a sketch, they no longer show up, instead just being rendered as boxes:
I mean, it ""should"" work, as Noto Sans definitely has these characters, but it seems they just can't be converted to sketch elements?
Re: FeatureScript versions
The versions that you can use are the names of the versions in the standard library. The imports of standard library items are then done with
import(path : "onshape/std/box.fs", version : "2641.0");, where box.fs is replaced with the name of the tab in the standard library and the version matches. Non-standard-library imports can be done with the import button at the top. They import with various 24-character ids.
https://cad.onshape.com/documents/12312312345abcabcabcdeff
Note that the main FeatureScript 1234; at the top is actually a language version. Generally it should match the current version of the standard library (A quick way to find out is to use FeatureScript 9999 and then inspect the parser error, which gives you a maximum version).
For example, this is what the start of onshape/std/debug.fs looks like:
