How to CAD Almost Anything

(ocw.mit.edu)

375 points | by o4c 3 days ago

24 comments

  • mafm 19 hours ago
    There has been a recent explosion in open source cad options.

    fluidcad appears to be a remarkably good option now - particularly for something that only started development about a year ago.

    I have used freecad, solvespace and onshape for at least a hundred hours each to make some reasonably complex parts. fluidcad resolves pretty much everything I found painful about those options.

    I have some more limited experience with build123d which is also nice, but just not as easy and interactive as fluidcad.

    Fluidcad provides normal interactive 3d cad interface with tools like extrudes/revolves/sketches but explicitly generates code as an intermediate step to constructing geometry and the code can be edited and refactored easily.

    The linking between the code and the gui works remarkably well in practice.

    It’s far easier to manage parameters and constraints in fluidcad where they are written out explicitly and can be edited vs what happened in a gui-only interface like Onshape.

    If you’re playing around with a semi-constrained sketch to understand kinematics onshape tends to “helpfully” add random constraints which is really annoying to identify and correct. In fluidcad everything is way more explicit and visible and clear.

    I have only spent about 5 hours using fluidcad but it looks remarkably useful despite some minor rough edges. It’s pretty much exactly what I wanted onshape or build123d to be.

    Solvespace is a really beautiful piece of software but quite limited. fluidcad has the same sort of nice vibes as solvespace but is way more powerful.

    • eadz 15 hours ago
      I love open source ( although slightly less relevant now ), and so I spent many many hours using OpenSCAD to write parts that in hindsight would have taken me 1/10 time in something like Fusion. When I finally learned Fusion, I regretted not learning it earlier.

      However now everything has changed. I personally believe that the days of designing by hand are numbered. I think the current apps are a bit like the early days of AI assisted coding - autocompleting certain manual steps. Compared with full agentic "solve x" prompts.

      I've been working on my own AI-First CAD ( and CAM! ), thinking about everything from first principles for AI - is it easier to think in terms of operations? or start with the finished product? I'm pretty happy with the engine ( and kernel ) I've (+agents) built, it can do full assemblies, including interactive sliders to move them etc, and then do the 3 axis CAM ( or export to STL for printing). I'm really interested in genuine feedback if anyone has a few minutes to try it out www.zelm.ai.

      • adamweld 14 hours ago
        As a mechanical and software engineer I've been following this space closely, and similarly thinking about what interfaces will power the CAD packages of the future.

        I checked out your project but was a bit disappointed to see that it's closed source. As you're probably aware in the last 6 months there have been a ton of new OSS B-Rep kernels and CAD programs hitting the ground, many of them working towards the same goals that you and I have.

        It's my personal opinion is that in this new age of AI coding, the best way for a project like this to gain momentum and user base is to make it free and open for other users/contributors to work on and with.

        Just my 2c.

        • eadz 14 hours ago
          I totally get the sentiment, but I think software has changed quite a bit. If we're honest, AI 99.9% built this, I just provided the vision.

          I've contributed to open source, and ran my own open source projects in the past, but when everyone can have the software exactly they want it's just a different ballgame. I think the value of Zelm isn't so much the code, but the infrastructure around it to make it a seamless experience.

          If there is a feature you want, just ask for it in the chat or via MCP, and it will be added to the backlog automatically.

          Open source was never about free as in beer, but in this case Zelm is pretty close to free if you use it as a MCP server.

          • adamweld 14 hours ago
            I've been having some interesting discussions with friends about this exact topic, whether AI will be a boon to OSS, or the end of it.

            I think that many tools, services, and programs will be more personalized, and maybe less shared. But that the core building blocks of the tools will stay open and continue to be refined by the community.

            When I'm coding with AI, one of the main reasons it's so capable is the ability to build off of languages, packages, and frameworks that are battle tested across many deployments. The end result is better if my (+ my agents) focus is on what's new and different about what I want to create, not reinventing the wheel. So I'll import a proven serial library rather than writing my own, pick the right programming language for my use case rather than writing in assembly.

            With regards to your project and CAD, I think it's a clean and efficient interface, and part I generated looked good. Personally I need more traditional CAD features for my day to day (drawings, multi part assemblies, hand editing) but it looks like a good tool for the right use cases.

            To finish my open vs closed source thought - have you considered open sourcing your b-rep kernel? Your app may be the secret sauce, but the kernel is boring but vital core. Parasolid has been developed over the last 35 years and is the reference point. NVIDIA open sourced SMLib but it is buggy and incomplete. OCCT exists but doesn't handle fillets well. Zoo spent two years with tens of developers building their kernel from the ground up (also unfortunately closed). There are a dozen new kernels written in Rust on github but none are complete.

            I really think that if all of the smart people trying to roll their own kernel worked together and put effort into really completing a good open source kernel, it would be the start of a cambrian explosion of CAD tools and AI enabled hardware engineering tools. Food for thought.

            • eadz 13 hours ago
              Have you seen https://solidcraft.io/solidcore/ ? It's a rust CAD kernel and open source. It's a port from SMLib. It also seems to have an active open source community.

              I don't really know much about CAD kernels, and so I really don't think I should be running a cad kernel open source project/community ( which is what successful open source is about ) I wouldn't be able to decide if we should do feature x or not.

              > Personally I need more traditional CAD features for my day to day (drawings, multi part assemblies, hand editing)

              Zelm supports drawings ( including custom drawings, see Download button ), assemblies ( with movement interfaces, click on Features ), and while it doesn't support hand editing ( i.e. with a mouse ), designs are parametric so you can tweak them ( click on Params down the bottom ).

              This is an example: https://www.zelm.ai/p/v8nzvHBhz7?share=Kis1TuAUtwn7nR3VNx-rj...

              • imtringued 13 hours ago
                >I don't really know much about CAD kernels, and so I really don't think I should be running a cad kernel open source project/community ( which is what successful open source is about ) I wouldn't be able to decide if we should do feature x or not.

                Is this some kind of joke I'm not getting? People use AI to do things so far outside their competency to the point where they have no idea about anything? I mean, who exactly would trust such a project? I'd rather do it myself.

                • eadz 13 hours ago
                  > People use AI to do things so far outside their competency

                  yes, if you want to do things far outside of your competency

                  > I mean, who exactly would trust such a project?

                  Depends how it can be verified. In this case there are potentially easy solutions, print it on a 3d printer and measure it. Is it 30mm across?

          • nomel 3 hours ago
            I don't buy it! The value in open sourcing is no different than it was before: help others save mental effort by letting them build on yours. Previously, this was meat mental effort, now it's tokens and your (not insignificant) mental effort and time it took with the guidance!

            I think open source is still very valuable, and we can still build on top of each others collective efforts (resulting in LLM).

      • reillyrider 13 hours ago
        I'm not really sure about AI CAD. The things I CAD are very specific, I can't just go "give me a generic box", it needs to be these exact dimensions, it needs to have cutouts at this position with this size, it needs holes at this spot, manufactured with these materials and tolerances, then I need this all nicely parametric and easy to change. If I need to put that many specific details in my prompt (and then double check the work afterwards) I might as well just do it myself.

        And how would iteration look like? It's a million times faster and easier to just navigate around a 3d model, click on the feature to edit, start editing. Describing which feature for the AI to change seems like a nightmare.

        • WillAdams 8 hours ago
          I tried to use an LLM for reasoning about a quite simple 3D project (a box with rabbet joinery) and it created a convincing UI and so forth, but when the DXF was examined, it was _not_ workable.

          I was able to (eventually) get a working file by laboriously describing the geometry of each part and its joinery in exhaustive detail:

          https://community.carbide3d.com/t/a-simple-cnc-box-in-depth/...

          Next up is adding a hinged lid to the existing sliding and no-lid options.

      • kabes 10 hours ago
        Thing is, I find natural language a really bad fit it describe a 3d model.
        • rurban 3 hours ago
          That's why the better tools use 3D images to generate the CAD models. And use natural language optionally to describe the model
    • NegativeLatency 17 hours ago
      Been a free tier fusion 360 user for years and recently switched to solid craft, it’s been amazing. Faster and nicer for what I do (3d printing stuff mostly) no cloud stuff and I can also use it via MCP which I find to be a nicer workflow than openscad

      https://solidcraft.io/

      Freecad never really worked for me well would get stuck frequently and not be able to figure it out.

      • omnimus 13 hours ago
        This is probably best full on AI clone we've seen. It is very bad news for all the established (Adobe, Autodesk) software.

        The authors have this custom process where humans pick yes/no on features/commits without looking at the code at all. The are abusing many claude 200usd subs and the tokens costs are apparently close to 100k usd.

        But still this is very bad news for traditional software.

      • sebazzz 5 hours ago
        FYI Fusion free tier also has an mcp, which is enabled in the settings. However, unless you give it guardrails like using constraints etc AI will make very shitty unmaintainable designs.
        • bsimpson 5 hours ago
          I used MCP to design a video game controller holder that clips onto a mic stand (e.g. for Rock Band/Stage Tour).

          Took a lot of effort. I was certainly supervising, but I don't know enough CAD to replicate an industrial designed video game controller. It made something that would have been outside of my skillset, and it did indeed hold a real controller on the first print.

        • donandon 5 hours ago
          Why would anyone use fusion these days.
          • buildsjets 4 hours ago
            Because CAD only solves half the problem, and the open source CAM packages that I have tried can only create simplest Gcode machining operations operations compared to a commercial package. If you want to cut a basic perimeter, you are golden. If you want to have automatic compensation for tool flex / spring-back, advanced stairsstep milling to minimize distortion due to internal stresses, or more than 3 axis control, you are out of luck.

            That said, I am working to transition to Solidworks as the noncommercial license is only $24 a year, while Fusion continues locking more and more basic functionality behind the paywall, and their first paid tier costs way too damn much for hobby work.

          • seesthruya 2 hours ago
            CAM
      • meelooo 4 hours ago
        have you tried https://macadamcad.com ? It's not open source but it's kernel is much more solid for complex models and organic shapes, among other things. It's also programmable with a node base graph engine similar to Rhino's grasshopper...
      • mft_ 13 hours ago
        +1

        I’m now using Solidcraft as a daily driver and it’s great. Some areas seem perhaps a bit ‘fragile’ (filleting, possibly surfacing) but overall it’s very stable and usable.

    • maouida 12 hours ago
      FluidCAD author here. Thank you for review.

      To be honest, I did not expect it to reach this level of interactivity when I first started it. I only wanted to add few interative features to make CAD by code more enjoyable. Now it has become more of UI driven CAD with the code there when you hit a wall.

      I'm going to add tools for surface modeling + tech drawings soon then I'll lock features for v1.

      • mafm 4 hours ago
        Thanks very much for creating it.

        Despite my limited experience with fluidcad, and some minor issues, it's probably already more practically useful to me than onshape and certainly more useful than freecad.

        Once I actually started using fluidcad, I was really surprised by how smoothly the the source code and the gui interactions work together. Using build123d and onshape, it seemed really obvious that something along that those lines could be really useful, but it really was surprising to experience how well and smoothly it works in your implementation.

        Is there a preferred way to discuss with other users (or you) at the moment?

        There's one key thing I want to be able to do that doesn't appear to be supported. But possibly it's already there and not documented - haven't investigated properly yet.

      • seesthruya 4 hours ago
        Awesome work so far, following this very closely! Like the OP FluidCAD is by far my favorite of the 'new' cad programs that have exploded over the last few months.
      • unimplmntd_trap 8 hours ago
        I’ve been following FluidCAD since early this year, the direction it’s going in is absolutely fantastic. It’s been a joy to use, thank you.
      • bsimpson 5 hours ago
        I'd not heard of FluidCAD until now, but this proposition:

        > Model with the mouse. Control it with code.

        sounds like what I loved about Flash when I first got into making things. Definitely wanna try FluidCAD now!

      • YuechenLi 7 hours ago
        FluidCAD is pretty fantastic and probably has the best UI out of the OpenCascade based CAD imo.
    • phkahler 7 hours ago
      >> Solvespace is a really beautiful piece of software but quite limited.

      Maintainer here. We just merged the long-wanted named parameters capability, so you can try it in the "nightly" builds on github. Not only that, but relation constraints (arbitrary equations on a sketch) and fully parametric parts - regenerated per-instance each with own parameters. Also in the works but not in yet are linking STEP files in assembly (up to the capability of our NURBS kernel) and a fillet and chamfer tool. All that on top of a huge number of geometry kernel bug fixes. We are still not using OCCT, but bringing all this capability to the same few megabyte executable. Oh, and the single-window interface is in. Still looks the same but the next release will be incredibly more complete.

    • YuechenLi 16 hours ago
      Well, if you like code CAD, I think I've finally got my implementation of CAD app to a state that isn't absolutely embarrassing now, if people would like to try it out.

      https://helios-cad-wasm.yuechenli.chatgpt.site/

      Be warned though, the Firmament DSL used is designed to be LLM friendly, so it's a language that's way easier to read than it is to write, and it has a very nontraditional constraint system.

      • eadz 15 hours ago
        This is quite cool, I think you had some similar ideas to mine. With LLMs your parts can at least be named nicely! What is nontraditional about your constraint system?
        • YuechenLi 14 hours ago
          There is no traditional 2D constraint solver for parallel/perpendicular/equal, etc. Instead, constraints are requirements generalized in code as Concepts and construction geometries (planes, axis, profiles, etc.) are unified as Concept Structs and erased at compile time like TypeScript, so over/under-constraint for sketches doesn't exist because the constraints are explicit and pre-resolved.

          It's easier if I show you some sample code.

          https://github.com/yuechen-li-dev/Aetheris/blob/master/fixtu...

    • WillAdams 19 hours ago
      FluidCAD looks very interesting, but when I tried it I got:

      >The FluidCAD engine did not start within 30s.

      which is an open issue: https://github.com/Fluid-CAD/FluidCAD/issues/80

      • mafm 19 hours ago
        Not sure what the issue is. I did see that a couple of times on a slow windows vm but it stopped happening after I reopened the app.

        I tried it again a few days later on a Mac and went through a couple of the tutorials and then did a couple of hours of actual work without any major issues.

        So yeah, definitely some rough edges at the moment, but once I got past those, the overall experience was surprisingly good.

      • maouida 12 hours ago
        That issue should have been fixed but probably there is a regression in the last release. I'll check it. Do you mind telling me you environment details?
        • WillAdams 10 hours ago
          Windows 11 w/ far too many programs installed (and in not enough cases uninstalled).
      • braiamp 19 hours ago
        > which is an open issue

        > closed this as completedon Oct 5

        Hmmm?

        • WillAdams 10 hours ago
          I was looking at a couple of issues in researching this and obviously confused them.
    • kabes 10 hours ago
      But still no decent foss alternative to parasolid, which really limits foss cad software
    • kilpikaarna 12 hours ago
      Sounds interesting, but, sigh, javascript…
    • SilentM68 18 hours ago
      >> There has been a recent explosion in open source cad options.

      Yea, especially more so with ArtCraft cloning the more expensive stuff.

      It would not surprise me to see ArtCraft versions of DS CATIA/MatLab/Wolfram that can compete with the originals, in the near future.

  • statenjason 22 hours ago
    MangoJelly’s FreeCAD courses on YouTube took me from zero to dangerous enough to design my own prints. I appreciated his approach of modeling a part and then modeling it again with an improved method.

    https://youtube.com/@MangoJellySolutions/

    • WillAdams 22 hours ago
      I've watched a fair number of these, but I always get hung up on some tiny little interface difference which confuses me.

      Even bought a book which was purportedly for 1.1 (according to a bar across a corner of the cover), but when I got it, the text referenced 1.0 and the screengrabs were also for the older version.

      I am hopeful that the UI will settle down sufficiently with the nascent 26.3 that learning materials will catch up.

  • seemaze 21 hours ago
    These are great! I've been using CAD for the better part of three decades, from AutoCAD R14 to 3DS Max, SketchUP, Rhino 3D, Solidworks, Vectorworks, and now OnShape.

    Parametric solid modeling (OnShape in particular) is by far the most enjoyable (for me) of the conceptual frameworks which transform ideas in to geometry, though it seems one of the most difficult to intuit for newcomers.

    I highly recommend anyone interested give it a shot, there's never been a better time with more resources available to take advantage of the incredible capabilities of these tools.

    • dividedbyzero 19 hours ago
      I wish there was a way to have at least a small number of projects private with OnShape's free plan. With everything public I can't use existing models I have the right to use, but not to re-publish, that's very limiting. Their cheapest paid plan is way more than I could justify for occasional non-commercial hobbyist modelling.

      There doesn't seem to be anything but Fusion 360 that's viable for hobbyists, and I believe that limits the number of projects to some crazy low number.

      There is FreeCAD, and I really do like the idea of open source CAD, but I don't understand how people get anything just the slightest bit complex done on it given how rudimentary and anal its constraint solver is. Dozens of hours of practice in it's still just painful in a way no other solver I've used has ever been, sucks all of the joy right out of the process. It's really not for me.

      I'm stuck with SolveSpace for the time being, which is much much nicer to use than FreeCAD, but extremely limited feature-wise, and it has a bunch of annoying bugs.

      I get why the big commercial players do it this way, but it sucks that there isn't some affordable non-commercial option for one of the established tools.

      • WillAdams 19 hours ago
        If you want something in-between FreeCAD and SolveSpace there is Dune 3D:

        https://dune3d.org/

        which has been discussed here previously:

        https://news.ycombinator.com/item?id=37979758

        https://news.ycombinator.com/item?id=40228068

        https://news.ycombinator.com/item?id=47475417

      • aliher1911 6 hours ago
        With fusion limitation is 10 editable documents, not total. I usually have half of the limit of things that are wip and for one offs I just set it to read only as soon as they are done. I can always set it to editable, make changes, print and set back to RO. I'm sure there are other workflows where you want to have many files edited at the same time.
      • adamweld 14 hours ago
        Realistically, for non-commercial hobbyist work, your best option is to just use OnShape and leave your models public. With a random name, no one is likely to find your work.

        Fusion lets you have private models, 10 of which can be active at any time. I personally prefer OnShape as a software.

        You can try some of the recent OSS clones hitting the web, e.g. SolidCraft as a Fusion replacement or CadRs for onshape. But expect some rough edges as the geometric kernel implementations are still very young compared to Parasolid which is the kernel used by most of the big dogs.

      • wlesieutre 14 hours ago
        Alibre Atom3D is currently 40% off for November sale, puts the one-time purchase at $120 and you can buy a year of updates later for $50.

        Currently Windows only, but the last major release cut out a lot of old .NET stuff and porting to other platforms is more likely now.

      • NegativeLatency 17 hours ago
        Would recommend trying solidcraft
      • FullStack_Byte 7 hours ago
        [flagged]
    • tootie 20 hours ago
      I've done some CAD training online (also onshape) and have gotten pretty handy and designed some things for my 3d printer. I can make whatever shapes but it all looks like crap and wilts quickly under load. I'd love any advice or resources for actually understanding functional and aesthetic principles. Or does it just come with experience?
      • jononor 13 hours ago
        For FDM 3d printing print/layer orientation is key to understand, as well as the concept of stress concentration. NeedItMakeIt I think has some nice videos around making strong functional parts, afair.

        For astetics I recommend doing an exercise where you make N=3+ variations of a simple part, with only differences in look. By simple think something like a clothes hanger, even angle bracket for shelves, a cup, etc. Experiment with different shapes and details. Repeat regularly with new parts, and over time you will develop a sense for different styles are achieved, and also develop your own style (or styles).

      • darksim905 17 hours ago
        I guess I never looked at the definition of wilt in relation to an item, but interesting.

        Telling you things you probably already know, but, there's designing an object and the representation of that (CAD) and then the creation of that in a real space and printing it (3D printing). The slicer that does the translation of how to go about printing that, is doing a lot of your interpretation and will define your quality of the item you are printing.

        My understanding is, different slicers have different feature sets and work good/better/best with different printers. Inherently, the same CAD file printed by several different printers, with different infil and different configurations, beds, heat, filament will all have different feel and function in terms of quality.

        I'm drawing a blank on the channel , but I'm positive there are a few YouTube channels that cover infil and how it affects the quality of a finished 3D print. Some of these concepts are no different than design principles for buildings and such that actual engineers™ learn and utilize.

        Once you get comfortable with those concepts (e.g. using honeycomb infil in some parts that need to be more rigid vs solid for something you want to survive -- also note the infil percentages) that's when you can combine that with how something pieces together and how to survive in different operating conditions.

        To your question: experience is the hard, costly and pain in the butt way to do these things. Finding a YouTuber you like and watching their content and experiences of the outcome of something they designed and why they made the choices they did, is free and implementing those ideas should* be painless.

        I know this isn't as much of a thing anymore, but I guess the quality for printing overall with certain filaments so good quality people don't mess with acetone baths for products anymore (I think it depends on filament). That finish can also make things look a lot better quality as well.

        Another variable in all of this that changes things is bed adhesion. If you don't have a good starting ground, there's nothing for everything else to connect to.

        • tootie 6 hours ago
          Coming from a software background, my analog is software performance. We are taught in CS that algorithm choice is paramount. You can always make your program run faster by adding more and better compute horsepower, but you'll get more bang by optimizing the algorithm.

          I'm looking at this one thing I made that has a base and a platform and the platform is supported by a single column and the column is slowly bending. Surely I can make the column thicker or denser or use a different filament but it also just kinda looks stupid. Like I should have a cross brace or taper the column or idk what else a real engineer would do.

          • darksim905 5 hours ago
            it all depends on the structure of that column internally, and the height of whatever it is you're building. If this is something personal and you don't care, then cool, leave it be! but if it's breaking or causing issues, obviously you'll want to fix it.

            I'd be curious what filament you're using. the algorithm here is very much your slicer and how it's printed. The only other real variable I guess is the extruder on the printer and how fine it can print.

            You mentioned a base/platform and a column.

            I wonder how the column is designed and if that platform is supposed by a wall or any reinforcement.

            maybe some of these links help:

            You could cheat and do things like inserting other hardware while building something: https://www.youtube.com/watch?v=b1JfzW8GPZo

            rounded corners and other design tips may improve quality of certain struts and orientation makes a difference: https://www.youtube.com/watch?v=wLBsO1UzIL0

            also: https://www.youtube.com/watch?v=7aJ4dDyyf20

            If you actually wanted to nerd out about material: https://www.youtube.com/watch?v=-ZPV-nVIIX4

            https://www.youtube.com/watch?v=Lq-SoGgKOcQ

      • NegativeLatency 17 hours ago
        Try looking at stuff you see and think about how you’d make it or how they made it

        I don’t really know how to make stuff beautiful but I don’t know if 3d printing has really been around long enough to develop its own signature styles. There are existing design styles you can copy though.

      • numpad0 11 hours ago
        Increase thickness? IME, 3D prints often have to be a lot thicker than their industrially made equivalents.
      • Palomides 8 hours ago
        are you using PLA? don't use PLA for anything that experiences sustained force unless you make it way chunkier than it would otherwise need to be
    • vermilingua 20 hours ago
      I’ve used FreeCAD and now mainly use Fusion360, but am absolutely fed up with Fusion’s performance and terrible cloud setup. Is OnShape comparable on its free plan?
      • NegativeLatency 17 hours ago
        Was in the same boat until last weekend, try https://solidcraft.io/
        • seemaze 17 hours ago
          will check it out!
          • NegativeLatency 16 hours ago
            I’m sure it’s got lots of bugs being so new but the community looks really promising and it’s worked great for my simple to mid complexity 3d printing needs
      • seemaze 17 hours ago
        There are two big caveats with OnShape when using the free tier:

        1. OnShape is browser based only. This has benefits and drawbacks.

        2. OnShape free tier models are all open, as in the equivalent of a public github repo.

        For me, the quality of the product is worth those trade-offs.

        • WillAdams 7 hours ago
          It's worth mentioning that the OnShape folks arrived at this after a licensing kerfuffle similar to that of Fusion 360 ceasing to be fully "free" (as in beer), so this status quo seems to be good for the long haul with little worry about being rug-pulled.
  • sinker 3 hours ago
    FreeCAD shill here. FC actually isn't actually difficult to use despite its reputation. For most parts you'll be using the same two workbenches and the same workflow:

    1. Draw a 2D sketch.

    2. Apply an operation to the 2D sketch to create a 3D feature.

    3. Repeat steps 1-2 until you have the finished object.

    The UI isn't bad. It's just poorly explained and overwhelming for new users. This is its main drawback.

    • Ccecil 3 hours ago
      Pre-1.0 it was a bit cumbersome.

      Nowadays I can barely tell the difference between using Freecad and Fusion. Most things I use/need are there.

      Mangojelly tutorials are extremely helpful for new users.

  • vogtb 22 hours ago
    I learned CAD back in 2016 from a guy on YouTube with a thick german accent. Ever couple of years I've needed to do something w/ some sort of CAD, mostly FreeCAD recently. It's nice to know I've not forgotten everything, but it's also interesting to play with the variety of MCPs and agent tools for CAD out there. Dictating that I want something to have holes at such and such a size, and spread 20 degrees apart and so on is interesting, and feels like essentially the same thing, but with a very specific image of what I want in my minds eye.

    The parts of this course seem generally the same to what I learned (although without the thick german accent) and probably translate equally well to the agent/mcp UX for CAD.

  • beering 22 hours ago
    Despite AI being able to do all this already, I still think it’s a very interesting skill to learn because almost all products around you are designed this way. For better or for worse, you will be able to see how CAD affected the design of products you touch or see.
    • manbash 20 hours ago
      Do you have an example? Last I checked, AI did bad meshes and didn't use common techniques. I believe it was back when stable diffusion was employed, though.
      • dagmx 18 hours ago
        CAD isn’t polygonal geometry be definition.

        Polygonal geometry is harder for GenAI to do because it’s still largely diffusion based to look like an object and topology is just a polygonal projection over it. It’s roughly equivalent to the diffusion model generating the image inputs to photogrammetry (not quite but a trivialized explanation)

        Meanwhile CAD is represented as a series of B-rep surfaces and operations. Both are largely more procedural and more akin to a programming language for models. Therefore it’s easier for an LLM to drive something towards an end result.

        Then the CAD engines tessellator generates the polygons, but topology is not super important for CAD.

        The best LLM CAD results I’ve seen even in academia are largely simplistic still though. The more data you put in , the better it’ll be able to match that but it’s sort of like giving it a very clear set of functions it can do to make something.

      • adamlrhodes 19 hours ago
        Over this past weekend I had Claude Opus 5.5 medium design some components using openscad, combined with some hobby servos and an arduino to make a robot arm, and then I printed them on my 3d printer and assembled them.

        Granted, it wasn't perfect, but it got pretty close, and with some light iterating, the finished result is not bad. Giving it access to prusaslicer as a tool helped it catch some common mistakes like overhang early too.

        https://github.com/AdamRhodes/robot-arm-public/tree/main/par...

        Checkout some of the stls in the repo if you are curious the kinds of stuff it can already do fairly well.

        I'd recommend trying it again, it has gotten significantly better since the last time I tried it about a year ago.

      • averynicepen 17 hours ago
        The most recent examples of the state of the art are from the initial release of GPT Astra and later Opus 5.5, maybe one or two months ago. Blender has already had an MCP since early this year, but it wasn't until these models came out that they are now capable of spatial reasoning and spatial consistency to be able to create something in 3D.

        The way it works is the same way all other agentic AI works - feedback loops, and an AI driving a computer. The AI creates a shape in Blender, then takes orthographic view screenshots so it can "see" what it did, double checks its work against what the goal is, fixes errors if there are any, and then repeats until done.

        It is worth noting that using Blender creates mesh representation 3D models, and does not create the boundary representation 3D models used in CAD programs. But I don't see anything stopping the industry from creating a similar MCP or other AI-native interface for b-rep CAD programs.

        You can find the latest examples on social media or Youtube if you search for keywords Astra, Opus 5.5, and Blender.

        This is a notably different approach than many of the existing "text to 3D" or "image to 3D" generators, which train their own machine learning algorithm to output models. Those have also been getting better, but the latest in this field was Astra driving Blender.

  • brudgers 1 day ago
  • userbinator 16 hours ago
    Students in this workshop learned basic CAD skills and reverse-engineering of an object into a 3D model.

    I wonder if this skill is one that comes naturally for some people, but is difficult and/or nonintuitive for others; perhaps it's related to the ability to imagine and mentally manipulate 3D objects in one's own mind, and translate between that and a 2D view. In the last century, when AutoCAD was one of the most pirated warez along with Photoshop, I had no trouble using the former to model various things I happened to have around, with just a bit of exploratory play and some reading of the online help.

    • martin_a 12 hours ago
      > I wonder if this skill is one that comes naturally for some people, but is difficult and/or nonintuitive for others

      I think so, yes. I've seen the same in students tasked with learning programming. For some the handling of loops, recursion and other concepts is easy, they have an image of how that works in their head, for others it's complicated and frustrating because they lack that image.

      I think it's the same for CAD and looking at objects and reverse engineering how things were constructed or put together.

    • arthur_steel 7 hours ago
      [flagged]
  • Snibz 14 hours ago
    As a CAD (BIM) developer, I think the learning curve caused by the complexity of CAD software will remain a pain point for users, regardless of open source or not. I could see open source CAD's gain significant popularity if they were easy to get into, and perhaps more importantly, easy to extend and modify. The CAD and BIM software business is in large part about creating solutions to niche industry needs.
  • pyuser314 8 hours ago
    FreeCAD. Not sure if anyone knows but you can write macros in Python that can touch any cad function on the program. I then use AI (or myself) to help create macros of features for my desired sketch. It’s like openscad in a way, but way way better because you can control the visual tools.
    • nisegami 7 hours ago
      I installed an MCP server plugin in FreeCAD and between that and the python scripting Claude/Fable has been killing it at 3D modelling for me now.
  • nightbun 21 hours ago
    This is weird timing.

    Just yesterday I setup Claude and Claude duet so I can just snap photos of a caliper in multiple ways around an object and let opus auto generate the model. Slice it in orcaslicer. Send to printer on LAN.

    Anything else I should know as I don’t want to have to do anything more than this to print items I need around the house

    • hackshack 20 hours ago
      Photogrammetry is the way. Many open and closed implementations. Higher resolution than FaceID and LiDAR sensors on iPhone. Keep the light and object in the same place; move camera around. Much higher fidelity.
      • coderenegade 19 hours ago
        That gives you a mesh, but for some applications (not the OP's) you might still want a BRep model, which is where the AI is still useful.
        • boaztheostrich 8 hours ago
          Cnc kitchens new mech to step tool is great. You can connect your favorite robot and it generates the file for you. My first pass was only ok but when I asked 6.1 sol to look at it again it was basically 1:1.
    • BirdieNZ 21 hours ago
      I want to get a flatbed scanner to take orthographic photos of items that I want precise measurements for to eliminate the caliper process, maybe that?
      • nightbun 20 hours ago
        Apparently iPhone lidar isn’t precise enough or else that would be neat to scan quick and instant print. Maybe it’s close enough for certain things.
    • Palomides 8 hours ago
      can you show the result?
  • corn-bread 4 hours ago
    I took this class a couple years ago!! It was excellent. Highly recommended!
  • WillAdams 19 hours ago
    So this has versions for:

    - Solidworks

    - Creo

    - Fusion

    - Inventor

    - OnShape

    - NX

    Is there any possibility of it getting updated for opensource options? FreeCAD would seem to be a good fit --- I'd love to see Dune 3D get some traction as well, and I'm very hopeful for the next version of SolveSpace which will add parameterization.

    • jlarocco 17 hours ago
      It pains me to say it, but CAD is one area where the open source options are light years behind the commercial offerings.

      Every open source CAD system seems to focus on geometric modelling or scripting geometric modelling because those are fun to work on. Those are important but small parts of what commercial CAD systems offer.

      FreeCAD, last I checked, didn't even have PMI.

      • jstanley 12 hours ago
        Some of us refuse to use proprietary software, so the universe of viable CAD software consists only of open source software.

        From that perspective, the commercial offerings are the ones that are light-years behind, due to not even trying to compete.

        I've been using FreeCAD for 9 years, it is amazing software and has changed my life. Don't fear FreeCAD.

        I've also tried out Solidcraft recently, it is impressive software but very slow and I keep running into bugs, so FreeCAD still wins for now.

        > FreeCAD, last I checked, didn't even have PMI.

        I can count on no hands the number of times I have wanted this. It's not part of my workflow at all. It's hardly a dealbreaker.

        • breakfast_burri 11 hours ago
          Good for you… All he is saying is there are many features open source CAD is missing that non-hobbyists take seriously.

          I’m a mechanical engineer and live in CAD. I can promise you, PMI is very much something that is taken seriously in industry.

          • matchstickman23 3 hours ago
            100%, I would LOVE an open source CAD suite that wasn't miserable to use. Hell, I'll even take something that offers a perpetual license that is up to snuff to professional offerings. Alas, Fusion is the only one that really comes close at a decent price point.

            Alibre is getting there, it's also not open source, but it at least offers perpetual licenses.

      • adamweld 14 hours ago
        Six months ago I would have agreed with you, but recently some FOSS CAD programs have landed that have a level of polish that's caused me to change my thinking here. They are, for better or worse, mostly clones of commercial software.

        You can check out solidcraft as one example. Easily good enough now for hobbyist 3d printing workflows, and a user can follow tutorials for Fusion to get started.

        As a professional it'll be a while before I give up Creo, but maybe not more than a year or two before the open source projects are ready for broad adoption.

      • YuechenLi 16 hours ago
        I wouldn't call geometric modeling "fun" to work on, you get to 80% of the way there super quickly, and the remaining 20%(boolean robustness, fillet/chamfer) is a never ending swamp if done with conventional approach.

        I think my implementation is the only open-source kernel with actual AP242 PMI support though and has the ability to label STEP files and generate 2D drawings directly from PMI. But it's code CAD and still fairly janky, if you want to try it out in the browser:

        https://helios-cad-wasm.yuechenli.chatgpt.site/

        Repo:

        https://github.com/yuechen-li-dev/HeliosCAD

        • jlarocco 16 hours ago
          Maybe "exciting" would be a better word than "fun"?

          People get excited about a cool looking 3D model. They don't get excited about implementing different PMI standards, or reading the ISO and AWS standards for weld symbols, or making sure a model saved 5 versions and 20 years ago still loads and behaves identical to how it did back then, or making sure hide/show in views from imported third-party formats works correctly.

          I took a look, but I didn't have any luck loading any of the NIST test STEP files from https://www.nist.gov/ctl/smart-connected-systems-division/sm...

          • YuechenLi 15 hours ago
            That's odd, considering I run round-trip import/export tests on the NIST test step files every commit.

            It works for me on both Chrome and Firefox.

            https://i.imgur.com/lWU5uP6.png https://i.imgur.com/2pyZ49X.png

            Do you have the error message when you try to import CTC-01? And did you use Open File instead of Import STEP and it didn't work? because if you did, that's probably something I should fix in the UI.

            • jlarocco 14 hours ago
              Sure: https://photos.smugmug.com/photos/i-qptfbk9/1/KMRfC7G6Trxfcx...

              I went through "Open File". When I tried just now some of the files opened, but several still failed.

              • YuechenLi 13 hours ago
                Should be hotfixed now, and the NIST AP242 parts should import/export at least. The kernel was importing with default tolerance of 0.001mm instead of the part's declared tolerance of 0.01mm. Thanks.

                And I think you understand now why I said build a geometry kernel is really a slog sometimes.

                • jlarocco 4 hours ago
                  I work on CAD software. I know the kernel is a slog, but I think the other parts are more so.

                  Usually the external format readers and writers are built on top of the kernel.

                  • YuechenLi 1 hour ago
                    Aetheris kernel for Helios is very different in that STEP AP242 read/write is part of the kernel itself, hence the term "STEP AP242 Native-ish", because the internal representation of the geometry maps directly to a mostly analytical subset of STEP instead of a different internal representation. This was critical because Aetheris doesn't mutate geometry from repeated STEP import/export, so any Aetheris exported geometry always reimports/reexports the exact same way, even though it is slower, which I don't think any other kernel on the market can claim and eliminate a class of error altogether.
  • Kuyawa 4 hours ago
    I don't need a course, I need an MCP
  • vhiremath4 22 hours ago
    If you are using Solidworks, I highly recommend Vertanux's SW course on YouTube. It's free, has example problems, and helped me really grasp the broad basics from bottoms up to tops-down assembly, drawings, etc.

    https://youtube.com/playlist?list=PLROUP1bV8REQmZgDTTJ0JCanX...

  • binsquare 18 hours ago
    For phyaucal engineering, Cad wasn't particularly the bottleneck ffrom my perspective as we were able to teach young kids to cad at a pretty competent level.

    The rigor behind a functional design needs to consider physics, manufacturability, and real engineering was what is considered good or great CAD

    • WillAdams 7 hours ago
      Interestingly, one of the better video tutorials for Moment of Inspiration was made by a group which used it to teach troubled youth while counseling them.
    • horsawlarway 18 hours ago
      CAD is still a decent bottleneck here. It's simple to use basic tools to create something, it's hard to create precise copies or very specific models without a lot of learning.

      And I agree on the need to understand engineering. Neat thing is that 3d printers have gotten incredibly cheap and capable, and they are wonderful immediate feedback on how CAD translates to real manufacturing.

      Is that the same as more traditional manufacturing techniques? No. Is it close enough a lot of the learnings carry over? Yes.

    • adamweld 14 hours ago
      Yes, exactly. Making shapes is picked up quickly; good hardware engineering is so much more.
  • Buttons840 18 hours ago
    I'm interested in 3D modelling, and I'm wondering if something like this would be adjacent enough to offer a new perspective but still be useful in art rather than just technical CAD?
    • ImPleadThe5th 10 hours ago
      I took a few CAD courses and feel like I'm a pretty proficient user for my needs.

      I found parametric CAD easy to intuit, but Blender very difficult.

      Most parametric CAD has a "2D first" workflow. You pick a plane draw a 2D sketch, constrain it to some dimensions and extrude 3d shapes from it. I like that it requires me to provide dimensions because it anchors my model in some kind of "reality". I feel this makes it both faster for me to get proportions right and allows me to play around with a tape measure to get an idea for what I'm making.

      Later, if I decide to modify the measurements the effects ripple through the entire design. And as a developer, I find using variables to also be intuitive and allows for some level of "generative" experimentation.

      Blender has a "3D first" approach where you are directly modifying mesh points/normals/faces. It feels less like working with rulers and pencils and more like working with clay.

      Many cad softwares have Mesh manipulation as well, but I feel like a lot of "getting started" tutorials will focus on the Sketch -> 3D workflow.

      Maybe there is a way to do this "unit based" drawing with Blender. If so I'd love someone with more experience to share!

      All that being said, playing in 3D and getting a feel for proportions with exact measurements might help you with your artistic modeling as well. It likely won't help you understand the tools better though.

    • JK-Swizzle 18 hours ago
      [dead]
  • xnx 22 hours ago
    Probably a useful intro tutorial for basic traditional CAD, but I don't see any mention of photo to 3D or AI assisted CAD. Workflows must be very different now.
    • matchstickman23 22 hours ago
      Not really all that different so far, even in companies where AI is extremely encouraged. From what I've seen, AI can help with tangential tasks to the actual design (exploded views, interference checks, probably basic drafting layout, etc) but it's still not that great at designing mechanisms or things that will be "used". By which I mean that AI doesn't have the same understandings as a human designer for how an item is intended to be used or operated.
      • radish_spirit 22 hours ago
        I can second this. I'm at an extremely AI-forward company, and though there are very aggressive internal efforts to figure out AI-first CAD workflows, it's the general consensus that the technology just isn't quite there yet.
        • nitroedge 17 hours ago
          kind of like when people shrugged off AI's ability to do image generation because it always created 6 fingers. Same story, very soon it will generate exactly what takes us hours or weeks to model now in minutes.
          • Silamoth 6 hours ago
            Your reply seems premised on the notion that “AI” image generation is good. I think most people would disagree with this. “AI” image generation is still laughably bad and painfully obvious. Most of us regularly laugh at “AI” generated flyers and such.
            • nitroedge 4 hours ago
              Your reply is premised on the fact (which I agree with) 95% of AI image generation is horrible, plastic, and artifical without any substance. Its the other 5% which when prompted and carefully thoughtfully crafted with perhaps a LoRA and fined tuned to an art style is awesome and can't readily be dismissed as AI generated.
        • YuechenLi 21 hours ago
          Just curious, what do you think is lacking in the current AI-first CAD workflow that makes it less than production ready?

          Full disclosure, I'm currently trying to build something to solve this and it would be nice to get some market feedback.

          • zonkerdonker 15 hours ago
            Ill take a shot at answering. Im also an ME at a very AI forward company.

            The hard part of mechanical design isn't really the CAD. Ask any long time professional mechanical design engineer, and I can almost guaruntee you that they will tell you that they can CAD up a design faster than they can think it. There are AI tools that can absolutely help with some of the more tedious tasks in the process (2D specification and tolerancing, part families and variants, texture tables etc). The difficult part of the job is synthesizing the massive number of variables and requirements and finding an optimal solution. Current models have been trained on all sorts of 3D, not just functional mechanical designs. AI doesnt know that a MIM steel part from vendor X can have a thinner wall thickness compared to a diecast steel part from vendor y, which would mean x grams of weight saving which cpuld be used to shore up a weak area on the other side of the part. It doesnt know that a sharp corner near an injection gate can cause porosity in some materials. It's not good at designing strong parts in general. It doesnt know that the ID team wants a specific surface on an user-facing surface which requires draft angle minimums relative to the mold parting direction. And, if you take the time to gather all that data and then type it all out for a model, you would spend twice as long writing everything down in plain language as it would take to just CAD it up yourself.

            It can take months to gather all of these requirements piece by piece from all teams across the org (rf design, thermal, structural and FEA, aerodynamics, environmental ratings, safety, etc etc), and as the design matures things change as well.

            For professionals with training in the tools, CAD as it is now is already the most optimal method of getting a design into the computer, and that was never really the hardest part of the process anyway.

            • YuechenLi 13 hours ago
              So... many of these are just DFM checks, right? Requirement checks like draft angle minimum and refusal of sharp angles near gates can be enforced programmatically during build time, and in terms of strength of material those can be checked via fast FEA. You are correct that modeling is never the hard part and everything else that happens after modeling is, and I do agree with you that the current AI CAD is insufficient on that front because they overfocus on the modeling portion.
              • breakfast_burri 11 hours ago
                I’d also add that a truly parametric cad model requires either a mystical level of precognition or constantly evaluating the function of the part and what is likely to change in each design step. AI struggles with functional evaluation (especially with moving components) and guessing what might be changed in the future.
              • zonkerdonker 5 hours ago
                My examples were there, sure. But they are "just DFM checks" if you know what manufacturing process is required for the part. Knowing that means knowing volumes, cost targets, part requirments, etc. Which again, is the hard part.
      • criemen 21 hours ago
        I'm very curious about this. I don't know anything about this space, but saw that OpenAI was advertising their mechanical engineering capabilities for the 6.2 release. How large of a part of the job is what AI can do at the moment? Do you feel a big jump in capabilities on each model release?
        • matchstickman23 3 hours ago
          I haven't been paying THAT close attention to it, but it can be good for getting a basic shape out and some suites are allowing for it to do direct parametric features. The problem comes in where deciphering the feature tree can take as much time as actually just building it yourself.
    • numpad0 21 hours ago
      that's because those things are not yet viable
  • villgax 11 hours ago
    The fact that this course isnt updated for Blender, OpenSCAD, FreeCAD or SolidCraft is just sad
  • analog8374 20 hours ago
    I really do prefer using a programming language for describing 3D models. Like openscad.

    Yes, the openscad language is weird. Considering switching to python.

    Do you relate?

    • GreyStache 17 hours ago
      I think you will be pleased by build123d. Especially the math mode feels pythonic to me.

      Also works well in combination of your favourite AI: "using python and build123d in math mode generate a banana". I don't oneshot it, just plan the object in advance and lead it there step by step.

    • WillAdams 19 hours ago
      If you like OpenSCAD, but want to use Python, then PythonSCAD, which just updated to version 1.2 is what you are looking for:

      https://pythonscad.org/

    • somat 14 hours ago
      Cadquery https://cadquery.github.io might be worth trying.
  • sneak 21 hours ago
    I have been building objects in Satisfactory lately and it’s the first interface I have found that seems really intuitive for making 3d shapes. Is there any modeling software that works this way (like an fps with build/delete gun) or must I just bite the bullet and adapt to blender or fusion or something? A long time ago I figured out sketchup but it turned into some sort of cloud garbage now and even back then I recall it being slightly clunky.
    • WillAdams 7 hours ago
      The one traditional 3D CAD tool which I was able to make it through the tutorial of is Dune 3D:

      https://docs.dune3d.org/en/latest/simple-part.html

    • rspeele 21 hours ago
      Not exactly what you are looking for, but I recommend following the Solvespace tutorial. It's short and simple and SolveSpace is a small download with a non-overwhelming number of features. It is harder to get lost in SolveSpace than in FreeCAD which is an absolute maze of "workbenches" and glued together features.

      https://solvespace.com/bracket.pl

      It is worth getting comfy with parametric sketch-based CAD because it's so powerful. It feels a lot like programming, really, in that by taking a good approach to modeling a part you can reduce or eliminate the amount of repetitive labor and make-this-line-up-with-that housekeeping you have to do.

      And the fundamentals transfer over to FreeCAD, Fusion, Solidworks, etc. pretty easily in the same way that once you know Java you can learn Python, C#, etc.

    • JustFinishedBSG 6 hours ago
      The best "CAD" to make assets is Plasticity (https://plasticity.xyz).
  • SilentM68 18 hours ago
    I don't see DS CATIA on the site? :(
  • bitwize 19 hours ago
    The same answer to everything else: ask Claude to do it
  • mrdootdoot 20 hours ago
    Hey Claude you got freecad and Isaac sim, shame the establishment. Build me a digital twin worthy of Mordor.