ttouch

thank you!

there are a couple and have very good reputation - though I've never used them

https://www.blacksmith.sh/ and https://runs-on.com/

They also say that they're much cheaper than github

you can use either the cursor cli and/or zed editor with cursor as the underlying provider with ACP (agent context protocol)

That's true but some years ago that I tried to develop a very similar product/project I went mostly on the same path. as another comment states, UI is a great place NOT to re-invent the wheel

Also changing CAD programs is already hard enough and having very similar UI/UX is a feature, not a by-product of agentic coding.

Also in many cases I honestly can't think of a better way to do stuff

Thank you very much!

That is exactly my take, though I'm ramping down the agents for the time being - 90% of the functionality is there and it's time to pay back the engineering time that I borrowed :P

I have a lot of experience developing CAD (before the AI boom) and while it made 0 sense to burn 1-2 years of my life to produce a similar MVP, now that I HAVE this MVP, it makes a ton of sense to pour my time on it

Also I'm gonna tackle some more god-problems with agents, like reversing the sldprt/f3d filetypes to allow importing them without losing history and more - each of this would require years of engineering hours and even achieving the 80% that agentic coding does is something that has never been done before

I heavily used LLM agents and it took about a month - of course I'm already starting to pay back the gained time

The development as I see it for such a project is about a year so I expect to take roughly 6 months to iron out the weird slop here and there

Having said that, the project right now is at an astonishing point and, in fear of jinxing myself, it works. Yea there are the occasional UI bugs here and there (incorrect tool grouping, incorrect tool icons, etc.) but welp, it does what I mostly want it to

Side note: I've done a similar project before and have tons of experience around building similar products, this is not an "AI success/AI will take your job" story - far from it

the target audience (which I'm a part off too) is hobbyist and small shops. Competing with the big giants I think is impossible (SolidWorks/Fusion/onShape/etc.) but I think that the vast majority of people use only a very small subset of features

Interoperability is of paramount importance and it's probably going to be the best deep dive

that was the idea and this is such a huge project that I kept as much variables static as possible. I want the app to feel close to what people already know - and that applies to me as a user too (I want to use the app too after all!)

This is the pre-alpha/alpha release of a new CAD that I created. It uses OpenCascade 7.9 under the hood via WASM and a custom sketch solver written in typescript (substantially faster than I thought it would be)

Right now it can do most of the CAD stuff that one expects apart from anything that involves meshes or curved face manipulation - that'll come soon but I want to have a stable base first before I move any further

I hope you like it, I'm eager to hear your thoughts!

we'll start posting only when gh is 100% up and it'll make it to the frontpage

this is such a good take, it makes so much sense and it's a very good answer to ai related interview questions

very good project!

can you tell us more about the use cases that you have in mind? I saw that you're able to run 1-4B models (which is impressive!)

meh, internal problems mostly

No idea where truck is going, it'll take me quite some time to tinker with CAD I think, it left me quite a bitter taste...

Fornjot seems to be doing good, I'm donating to them and I get regular updates (you should too!). Still, there's a long, long road ahead

If I were to do this all over again I'd either go the OCCT route (like chili3d or zoo) or solvespace. they're both "lacking" kernels if you compare them to the commercial ones, but I think there's enough "market gap" for makers that would prefer a sustainable CAD format instead of perfect fillets (and IMO freecad is not the solution).

Keep in mind though that my efforts where laser focused on non-math stuff. From what I gathered from my time in cadmium, b-rep kernels are hard in an unsustainable level. Browser level unsustainable. I just hope that out of seer necessity we'll find another way to solve the CAD problem, instead of a b-rep kernel

Cadmium is long dead and unfortunately neither truck nor fornjot are "there" yet - "there" been anything more complex than a cube

Context: I was the main contributor/maintainer of cadmium

this is hands down the most heartwarming post I've seen in A WHILE.

Like most people said, I too started kinda like that. Coding/computers is an amazing journey with near infinite paths to choose from

I wish you the best and slow clap to the amazing dad that was there to set the stage and be there for him.

Made my week for sure!

Just feed it a git URL and it will build the project and upload any produced artifacts in a release.

Especially useful for reversers and malware analysts

It seems to be too simplistic. Just sketch & pull. No Surface/Mesh modeling, no advanced mirroring/etc. tools and no CAM (I can use something else for CAM but would be nice to have) And most of all: it doesn't run in wine

Paywall free archive: https://archive.ph/xCfBT

I did that, the very hard way (I didn’t know better at the time): https://whynot.fail/homelab/lets-encrypt-the-house/

I actually set it up and it's much better than my previous setup. Thank you! Editing the post that it's totally useless (but leaving it on the net anyway, just with a note on top).

I'm kinda disappointed as I was very proud for the setup but hey, let's not get emotional over bits n bytes :)

Damn you can do this whole thing with Cloudflare for free? That's why that's a failure blog :'D

I'll give it a try at some point...

that's a good idea, but keep in mind that the underlying file descriptor could be of a UDP stream, or a TCP stream to another host. I finally solved the problem, check out https://github.com/dzervas/mage/blob/master/src/api_ffi.rs. When you bind or connect, I return a "file descriptor" that you will pass back to me and I use it as an index to a vector of connections :)

It works fairly good - although I have other problems...

what is the difference with showterm?

http://git.dzervas.gr/c4 An attempt to make this readable :) (at least with right syntax...)