There are currently a lot of newer browser CAD projects that already wraps OCCT WASM, like FluidCAD and Chili3D. And as you said previously, current CAD tools are slow to improve, and part of the reason is because of using kernels first built in the 1980s, and projects like Shapr3D and Plasticity both hit limitations of their existing kernels and switched to Parasolid when their project got bigger, and OCCT's per entity tolerance issue is pretty well know by now.
On another topic, getting a plate with thousands of holes to mesh in under a second is some serious performance, and I do agree with your conclusion that most CAD apps are slow not because of programming language but because accurate NURBS math is inherently slow, but their is also the issues with the 2D constraint solver and fillet/boolean algorithms too. Happy to talk more about the geometry kernel stuff too.
That's awesome. I'm also working on a new B-Rep kernel for my own modelling app.
It's small, ~2MB, but already handles a lot of complex modelling and has constraint-based sketching - I'm focusing on watertight shapes for 3D printing, not exact accuracy. All operations map to KCL (KittyCad - now Zoo) so it's possible to edit the code directly.
Not to brag too much, but I believe that my own code CAD language, Aetheris' Firmament, is a bit ahead of KCL in terms of language design, so maybe take a look and see if Firmament is a better target for you to map to?
I had a brief look at Aetheris and correct me if I'm wrong but it can't do any CSG between real world assemblies. All code is for special cases like cylinders, cones, planes.
I do not see any code for general NURBS/trimmed-surface SSI. And after SSI classification (what to keep and what to throw away) is just as difficult.
It seems a lot of people are working on CAD kernels these days, in Rust or otherwise, and clanker-written READMEs and messages over-promise on abilities because the authors have never worked with real world assemblies and trust what their agents tell them they "implemented".
I tried importing two simpler STEPs you can find online, boxy-with-surfacetex.step (fails import entirely) and raspberry-pi-4-model-b.step ( Step242Importer.ImportBody fails, Step242Importer.ImportRidgidRoots succeeds, Step242AssemblyImporter.Import crashes).
On the latter a simple BrepBoolean.Subtract(ridgid_roots, box) returns: NotImplemented: "bounded boolean family only supports recognized axis-aligned boxes"
This is a simple as it gets with a real world assembly. A box is just six plane intersections. And Aetheris can't do it.
I also tried a furniture STEP from a designer friend, Sascha Ulber the simple version (w/o rolling ball fillet surfaces) of his Coffy chaiselongue. The STEP came from Rhino3D. It succeeded the Step242AssemblyImporter.Import but all booleans I tried gave the same error as the PI model above.
Truly nice stuff. I've been looking into getting into 3-d printing recently and wondering could this output the kind of files I could use for that? Truly possibly a noob question, but just wondering if this could do that.
Onshape, which is also free and runs in the browser, is probably the better choice. This looks more similar to an autocad style drafting application, which isn't an ideal fit for most 3D printing applications. This app is ideal for doing things like architectural work, or traditional machining where the 2D drafts are vital for sharing with the engineers who will actually make the part on a lathe or whatever.
3D printing doesn't need 2D drafts so you're usually better with a solid modelling CAD app. Fusion360, Solidworks and Onshape are all reasonable choices for this - I use onshape because I dislike fusion and solidworks doesn't work on Linux.
Fusion is arguably the easiest and best free option to pick up, yet I would urge people to avoid it. No linux support, won't work with any workarounds, mostly online, with the non-CAD part of the UI being absolutely horrible and slow.
Oh you want to add a new project? Let me animated from the bottom of a hundred of other projects laggily, then wait 5 seconds to be able to name the project, only for it to animate all the way down again!
The project title is too long? Guess the only option is to view it online!
Search is slow as butt too.
There are issues with some of the CAD features too, chamfer/fillet sometimes just don't work for no obvious reason (where solidworks has no trouble filleting the same geometry) and patterns are extremely, painfully slow.
This is reasonable criticism of Fusion, but I stick with it for the integrated CAM. Can't really find any other easy to use "professional" caliber software with CAD & CAM at that price.
Solidworks have a "for makers" product if you're not doing this professionally that has integrated CAM (assuming you don't need 4/5 axis support). It's not free but it's very cheap. If it worked on Linux I'd probably be using it.
Thanks. Yes, Aircad can be used for 3d printing workflows! You can import and export stl and obj file formats, which are used for 3d printing. We plan to add more formats.
Aircad is a general solid modeler. So, you can model anything w the solid modeling tools, and then convert to a mesh. They participate in boolean ops like union and subtract. There's a collection of commands specifically for manipulating meshes: https://aircad.com/docs/commands#meshes. The agent can also write scripts to generate and process meshes completely from scratch: https://aircad.com/docs/ask
Thanks for asking. Core ops are OCCT. It's got great primitives and tolerant modeling. For performance purposes, I've implemented my own algorithms for hidden line rendering, tessellation, and others. I've spent my whole career (15 years+) working on and around kernels.
More detail for the curious: it’s C++ compiled to WebAssembly and runs entirely client-side with a WebGL2 renderer, so there’s no server round-trip for geometry. OCCT does booleans, fillets, and STEP. I replaced its mesher with my own tessellator because BRepMesh was the bottleneck at scale, and hidden-line/Make2D are custom too. Happy to go deeper on any of it.
Browser-based CAD with real professional-grade capabilities is a big deal. Especially the scripting API + AI integration that's a much more interesting direction than simply adding an AI chatbot to existing tools. Great work!
Thanks so much! That’s basically my favorite thing to hear. :) We spent a ton of time ironing out the bottlenecks for huge projects. CAD users are notorious for pushing products as hard as possible. A few things we did:
The C++/WASM engine owns the model, selection and WebGL2 renderer on one thread, so nothing is copied back and forth on every edit. We use multithreading, SIMD, the works...
For instance, we wrote our own tessellator to replace OpenCASCADE’s mesher, and it only re-meshes the faces that changed. A plate with 2,500 holes meshes in 127 ms instead of about 10 s.
Solids are batched into a few draw calls, and repeated geometry is instanced automatically.
We hunt down anything quadratic. If I had to guess, that is what makes most CAD software interaction slow.
i actually recently stood up my own end to end step reader writer as part of my own cad shenanigans, where i can have time to first geometry be in milliseconds even if the overal file is half a gig and physically needs a few seconds to parse. is there some reason why you dont have step export?
Fair concern, and it’s the right question to ask of any new tool. A few concrete answers:
Your work isn’t trapped. Drawings export to DWG, DXF, PDF, and STEP at any time, and you can save the native file locally. If Aircad vanished tomorrow, your drawings open in AutoCAD.
The engine runs in your browser, not on our servers. Modeling and drafting don’t round-trip to a server, so it’s fast and isn’t fragile the way server-side CAD is.
On liability: the professional of record carries it whatever the tool. AutoCAD’s license disclaims it too. What a firm needs from software is accuracy, standard formats, and an exit path, and that last one is why DWG round-tripping got so much of my time.
The price is low because a small team building on modern infrastructure has much lower costs, not because corners were cut.
AutoCAD can charge a high price exactly because they have the moat GP is talking about. I wonder what price they started at.
(The argument put forth is similar to "no one was ever fired for hiring IBM." I think that's true in some circles. To me it's a red flag that you hire by name and not ability.)
That’s fair, and it’s the right thing to weigh with any new tool. Right now it’s a small company and I’ve written most of the engine. What I’ve tried to make sure is that your work never depends on Aircad surviving. Everything exports to DWG, DXF, PDF, and STEP whenever you want, and you can keep native files locally, so the exit path is always open. The engine also runs entirely in your browser, not on our servers. I’d rather earn trust over time than ask for it up front.
Or, capture the hobbyist and DIY market. Lots of people have 3D printers now. Lots of people hate the existing CAD offerings and especially the fact that the affordable offerings tie you to a cloud service. A product like this is very welcome.
The product must be both amazing and extremely cheap to have a long term chance against the free and opensource vibed cads like solidcraft (and lets not forget freecad although it has lower vibe levels). And fusion for example is free for hobby use if you are ok with the limitations. I really wonder how you can compete in the hobbyist circles against stuff you can fork and tell ai to create the exact feature you want without asking permission or convincing anyone about your use case. The serious business is a different story since you need accountability and stability, but also there the attitudes may change eventually since some companies are already ditching autodesk products for cheaper yet very capable stuff like bricscad that offers also a fairly priced perpetual license. I am never again buying a subscription to cad software. If I cannot actually buy your product and own it I am simply not interested anymore.
Figure I try pitching at least, but would you consider evaluating my Aetheris kernel to use as Aircad's geometry kernel over OCCT?
https://github.com/yuechen-li-dev/Aetheris
There are currently a lot of newer browser CAD projects that already wraps OCCT WASM, like FluidCAD and Chili3D. And as you said previously, current CAD tools are slow to improve, and part of the reason is because of using kernels first built in the 1980s, and projects like Shapr3D and Plasticity both hit limitations of their existing kernels and switched to Parasolid when their project got bigger, and OCCT's per entity tolerance issue is pretty well know by now.
On another topic, getting a plate with thousands of holes to mesh in under a second is some serious performance, and I do agree with your conclusion that most CAD apps are slow not because of programming language but because accurate NURBS math is inherently slow, but their is also the issues with the 2D constraint solver and fillet/boolean algorithms too. Happy to talk more about the geometry kernel stuff too.
That's awesome. I'm also working on a new B-Rep kernel for my own modelling app.
It's small, ~2MB, but already handles a lot of complex modelling and has constraint-based sketching - I'm focusing on watertight shapes for 3D printing, not exact accuracy. All operations map to KCL (KittyCad - now Zoo) so it's possible to edit the code directly.
Might be worth exchanging notes at some point :)
Not to brag too much, but I believe that my own code CAD language, Aetheris' Firmament, is a bit ahead of KCL in terms of language design, so maybe take a look and see if Firmament is a better target for you to map to?
https://github.com/yuechen-li-dev/Aetheris/blob/master/fixtu...
I had a brief look at Aetheris and correct me if I'm wrong but it can't do any CSG between real world assemblies. All code is for special cases like cylinders, cones, planes.
I do not see any code for general NURBS/trimmed-surface SSI. And after SSI classification (what to keep and what to throw away) is just as difficult.
It seems a lot of people are working on CAD kernels these days, in Rust or otherwise, and clanker-written READMEs and messages over-promise on abilities because the authors have never worked with real world assemblies and trust what their agents tell them they "implemented".
I tried importing two simpler STEPs you can find online, boxy-with-surfacetex.step (fails import entirely) and raspberry-pi-4-model-b.step ( Step242Importer.ImportBody fails, Step242Importer.ImportRidgidRoots succeeds, Step242AssemblyImporter.Import crashes).
On the latter a simple BrepBoolean.Subtract(ridgid_roots, box) returns: NotImplemented: "bounded boolean family only supports recognized axis-aligned boxes"
This is a simple as it gets with a real world assembly. A box is just six plane intersections. And Aetheris can't do it.
I also tried a furniture STEP from a designer friend, Sascha Ulber the simple version (w/o rolling ball fillet surfaces) of his Coffy chaiselongue. The STEP came from Rhino3D. It succeeded the Step242AssemblyImporter.Import but all booleans I tried gave the same error as the PI model above.
Truly nice stuff. I've been looking into getting into 3-d printing recently and wondering could this output the kind of files I could use for that? Truly possibly a noob question, but just wondering if this could do that.
Onshape, which is also free and runs in the browser, is probably the better choice. This looks more similar to an autocad style drafting application, which isn't an ideal fit for most 3D printing applications. This app is ideal for doing things like architectural work, or traditional machining where the 2D drafts are vital for sharing with the engineers who will actually make the part on a lathe or whatever.
3D printing doesn't need 2D drafts so you're usually better with a solid modelling CAD app. Fusion360, Solidworks and Onshape are all reasonable choices for this - I use onshape because I dislike fusion and solidworks doesn't work on Linux.
Fusion is arguably the easiest and best free option to pick up, yet I would urge people to avoid it. No linux support, won't work with any workarounds, mostly online, with the non-CAD part of the UI being absolutely horrible and slow. Oh you want to add a new project? Let me animated from the bottom of a hundred of other projects laggily, then wait 5 seconds to be able to name the project, only for it to animate all the way down again! The project title is too long? Guess the only option is to view it online! Search is slow as butt too. There are issues with some of the CAD features too, chamfer/fillet sometimes just don't work for no obvious reason (where solidworks has no trouble filleting the same geometry) and patterns are extremely, painfully slow.
This is reasonable criticism of Fusion, but I stick with it for the integrated CAM. Can't really find any other easy to use "professional" caliber software with CAD & CAM at that price.
Solidworks have a "for makers" product if you're not doing this professionally that has integrated CAM (assuming you don't need 4/5 axis support). It's not free but it's very cheap. If it worked on Linux I'd probably be using it.
Fair, I only do 3D printing, so my experience is limited to that
Thanks. Yes, Aircad can be used for 3d printing workflows! You can import and export stl and obj file formats, which are used for 3d printing. We plan to add more formats.
Aircad is a general solid modeler. So, you can model anything w the solid modeling tools, and then convert to a mesh. They participate in boolean ops like union and subtract. There's a collection of commands specifically for manipulating meshes: https://aircad.com/docs/commands#meshes. The agent can also write scripts to generate and process meshes completely from scratch: https://aircad.com/docs/ask
What is the geometry kernel? That's always the first question I have.
Thanks for asking. Core ops are OCCT. It's got great primitives and tolerant modeling. For performance purposes, I've implemented my own algorithms for hidden line rendering, tessellation, and others. I've spent my whole career (15 years+) working on and around kernels.
More detail for the curious: it’s C++ compiled to WebAssembly and runs entirely client-side with a WebGL2 renderer, so there’s no server round-trip for geometry. OCCT does booleans, fillets, and STEP. I replaced its mesher with my own tessellator because BRepMesh was the bottleneck at scale, and hidden-line/Make2D are custom too. Happy to go deeper on any of it.
Product aside, it's nice to see what seem like real answers over equivocality. Wishing you the best of luck!
Thanks so much! Appreciate it. It's a long road from here in CAD land. :)
OCCT doesn’t handle “complete” fillet/chamfer, right? Where the operation replaces the top line?
It does in the next (8.1) release: https://github.com/Open-Cascade-SAS/OCCT/issues/172
It still generates invalid solids apparently: https://github.com/Open-Cascade-SAS/OCCT/pull/1625
Browser-based CAD with real professional-grade capabilities is a big deal. Especially the scripting API + AI integration that's a much more interesting direction than simply adding an AI chatbot to existing tools. Great work!
I don't want cad. I want CAM. Where are the AI generated CAM clones?
Here is one recently released (commercial) full 3D 3-axis (+4th axis) with automatic feature recognition https://takshakcam.com/
Solidcraft (fusion look-alike) have a CAM-module in the works but not released yet. https://github.com/tangency-labs/solidcraft
I bet Mitcad (also fusion look-alike) will also have a CAM-module soonish. https://github.com/mitcad/Mitcad
Here is one 2.5D CAM software (developed by me) although not a direct clone of anything. https://github.com/chipload/chipload
Interesting. What kind of workflows and toolchains are you looking for? MasterCAM?
What's the current state of foss in regards to replacing parasolid?
Not even close. But with AI maybe 3 to 5 years.
Just checked it out; the whole experience is incredibly smooth.
Thanks so much! That’s basically my favorite thing to hear. :) We spent a ton of time ironing out the bottlenecks for huge projects. CAD users are notorious for pushing products as hard as possible. A few things we did:
The C++/WASM engine owns the model, selection and WebGL2 renderer on one thread, so nothing is copied back and forth on every edit. We use multithreading, SIMD, the works...
For instance, we wrote our own tessellator to replace OpenCASCADE’s mesher, and it only re-meshes the faces that changed. A plate with 2,500 holes meshes in 127 ms instead of about 10 s.
Solids are batched into a few draw calls, and repeated geometry is instanced automatically.
We hunt down anything quadratic. If I had to guess, that is what makes most CAD software interaction slow.
i noticed you dont have step export! why is that?
i actually recently stood up my own end to end step reader writer as part of my own cad shenanigans, where i can have time to first geometry be in milliseconds even if the overal file is half a gig and physically needs a few seconds to parse. is there some reason why you dont have step export?
STEP export is on our public roadmap: https://aircad.com/roadmap#building
I'll ping this thread when it's ready.
STEP export is now available in Aircad.
the landing page and copy is top quality
Aw thanks! I spent longer than I wish to mention on this.
This is super, just checked it out.
Thanks so much. Let me know if there's anything you think is missing. There's still a lot to build on our roadmap: https://aircad.com/roadmap
Congratulations on launching.
I would not bet a business that depends on CAD on something that only costs $19/month. Even if it didn't require taking a dependency on a website.
Or to put it AutoCad is cheaper than being sued. Good luck.
Fair concern, and it’s the right question to ask of any new tool. A few concrete answers:
Your work isn’t trapped. Drawings export to DWG, DXF, PDF, and STEP at any time, and you can save the native file locally. If Aircad vanished tomorrow, your drawings open in AutoCAD. The engine runs in your browser, not on our servers. Modeling and drafting don’t round-trip to a server, so it’s fast and isn’t fragile the way server-side CAD is. On liability: the professional of record carries it whatever the tool. AutoCAD’s license disclaims it too. What a firm needs from software is accuracy, standard formats, and an exit path, and that last one is why DWG round-tripping got so much of my time.
The price is low because a small team building on modern infrastructure has much lower costs, not because corners were cut.
Problem is even if you can export as STEP you loose all the rest of your data like the edit history and so on.
AutoCAD can charge a high price exactly because they have the moat GP is talking about. I wonder what price they started at.
(The argument put forth is similar to "no one was ever fired for hiring IBM." I think that's true in some circles. To me it's a red flag that you hire by name and not ability.)
Adjusted for inflation AutoCAD probably costs less today. And a system including hardware is much much less.
The moat has always been stability (except for r13 which was a complete rewrite).
a small team building on modern infrastructure has much lower costs
Support, training, documentation, UX design, testing, etc. don’t.
In a B2B setting, those matter a lot.
And it is a good strategy to not compete with zero dollars.
As a related note, reading lots of "I built" here is not confidence inspiring, no matter how good you are. There's always that bus factor thing.
That’s fair, and it’s the right thing to weigh with any new tool. Right now it’s a small company and I’ve written most of the engine. What I’ve tried to make sure is that your work never depends on Aircad surviving. Everything exports to DWG, DXF, PDF, and STEP whenever you want, and you can keep native files locally, so the exit path is always open. The engine also runs entirely in your browser, not on our servers. I’d rather earn trust over time than ask for it up front.
If it is a B2B product, price it like a B2B product so you canbuild a B2B company with B2B service.
Or, capture the hobbyist and DIY market. Lots of people have 3D printers now. Lots of people hate the existing CAD offerings and especially the fact that the affordable offerings tie you to a cloud service. A product like this is very welcome.
The product must be both amazing and extremely cheap to have a long term chance against the free and opensource vibed cads like solidcraft (and lets not forget freecad although it has lower vibe levels). And fusion for example is free for hobby use if you are ok with the limitations. I really wonder how you can compete in the hobbyist circles against stuff you can fork and tell ai to create the exact feature you want without asking permission or convincing anyone about your use case. The serious business is a different story since you need accountability and stability, but also there the attitudes may change eventually since some companies are already ditching autodesk products for cheaper yet very capable stuff like bricscad that offers also a fairly priced perpetual license. I am never again buying a subscription to cad software. If I cannot actually buy your product and own it I am simply not interested anymore.
The odds that the people who hate everything in the space will love your product are close to zero.
Particularly when what they hate is paying.
If you want to stay in business, customers who want you to stay in business are much much much better than competing on price.
Consumer CAD has a long history and the absence of good products in the space is a consequence of its economic viability.
For context, I’ve been involved with CAD since 1989. I’ve seen many things come and go.
Can test it w/o signing in. Pass.