Loading video...

Video Failed to Load

Go Home

🔥We’re open-sourcing Multi-Agent CAD, a text-to-CAD pipeline that generates 3D assets for about $0.15 each—at roughly 1/13 the cost of CAD Skills, with faster inference and higher accuracy 🚀 This project began when our lab recruited Guanxing for his CAD expertise 🛠️ to build connected parts for robot learning...

150,324 views • 1 month ago •via X (Twitter)

57 Comments

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Code:

AI Mastery Guide's profile picture
AI Mastery Guide1 month ago

15 cents per 3D asset is a massive drop in cost.

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Yeap : ) Feel free to try it out and raise any issue!

AI Mastery Guide's profile picture
AI Mastery Guide1 month ago

Definitely will, open sourcing this could push the whole space forward.

Numan's profile picture
Numan1 month ago

Thanks for sharing !!!

AdventureNLearn's profile picture
AdventureNLearn1 month ago

This is WONDERFUL!!!

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

Arrow's profile picture
Arrow1 month ago

Interesting! Following along 🔖

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

Dylan Oh's profile picture
Dylan Oh1 month ago

Exactly what I need at the moment. Starred

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

Vinod Thale 乐辉's profile picture
Vinod Thale 乐辉1 month ago

thank you how its different from @

Xueyan Zou's profile picture
Xueyan Zou1 month ago

We are using multi-agent, it is cheaper with better success rate. The code is very different if you do a comparison. We also politely cite and compare with them at the very beginning of the github paper. If you see my twitter, I just share their code base yesterday.

Vinod Thale 乐辉's profile picture
Vinod Thale 乐辉1 month ago

好的 谢谢

skhlgnev's profile picture
skhlgnev1 month ago

nice job!

Cagatay Cali's profile picture
Cagatay Cali1 month ago

I am curious to compare the notes on this for text to print directly…

bdca's profile picture
bdca1 month ago

Thank you!

Paul's profile picture
Paul1 month ago

Wow! Will try it out! gave you a follow as well✅

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

Praveen Kumar Verma's profile picture
Praveen Kumar Verma1 month ago

Awesome to see. Can this be integrated with kicad?

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Sorry, not so familiar with kicad, the output is stl file, raise an issue in the github to ask my student 🥲

Praveen Kumar Verma's profile picture
Praveen Kumar Verma1 month ago

Got it. It is amazing though. Thanks for the work.

Jenny Zheng's profile picture
Jenny Zheng1 month ago

@J0hn_bee check this out!

Xueyan Zou's profile picture
Xueyan Zou1 month ago

@J0hn_bee Thanks!

Matthew Thomas's profile picture
Matthew Thomas1 month ago

Incredible, that's a beautiful piece of art. Added to my bucket list. So peaceful.

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

Shadow Defense's profile picture
Shadow Defense1 month ago

add actuator planetary and cycloidal parts along with the exact motors that can be used to create actuators with pancake drone motors and nema, etc. then please give me 5% of sales for the idea.

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

Everett Hope - DadCanMakeIT3D's profile picture
Everett Hope - DadCanMakeIT3D1 month ago

This is beyond cool.

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

Markus J. Buehler's profile picture
Markus J. Buehler1 month ago

👀

Xueyan Zou's profile picture
Xueyan Zou1 month ago

👀

Jonathan Korstad's profile picture
Jonathan Korstad1 month ago

Nice!!

Hubert Thieblot's profile picture
Hubert Thieblot1 month ago

@Do_owne

Rizwan Syed's profile picture
Rizwan Syed1 month ago

This looks very cool, Professor. Would love to try it.

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

Knowix's profile picture
Knowix1 month ago

good to see guys,I’ll test it out and give feedback

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Give me a feedback if you happens to play with it.

alber 🌊's profile picture
alber 🌊1 month ago

Nice! These are the kind of projects that will lead to computers building useful things 🚀

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

HumanoidRobots.com Tom's profile picture
HumanoidRobots.com Tom1 month ago

the $0.15 ball-in-cage is going to ruin a lot of afternoons.

Xueyan Zou's profile picture
Xueyan Zou1 month ago

hhahah!

MP's profile picture
MP1 month ago

Awesome stuff

BadGuyTy 🅁🅅🄽 (Tyler Hess)'s profile picture
BadGuyTy 🅁🅅🄽 (Tyler Hess)1 month ago

THANK YOU! been looking for a good text to cad repo

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks!

Gregor's profile picture
Gregor1 month ago

The $0.15 figure makes sense if the output is clean geometry, but genuinely asking, how are you measuring 'higher accuracy' here? Prompt adherence and structural validity feel like very different bars.

Keifer Lee's profile picture
Keifer Lee1 month ago

Awesome work, congratulations on the release! That aside, I personally know of a few founders in the text -> CAD space and they'd be stoke to see releases like this. Perhaps a super impactful continuation would be to create a set of useful "harness" on top of this. Cheers!

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks a lot!

Timothy Siopan's profile picture
Timothy Siopan1 month ago

So cool, look forward to testing it!

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Thanks! Give us a feedback!

clankr's profile picture
clankr1 month ago

Amazing work. The results in the video look great, but could this be used to manufacture mechanical parts for something like a robotic arm?

Xueyan Zou's profile picture
Xueyan Zou1 month ago

Yes, we are extending on that direction.

Moez Zhioua's profile picture
Moez Zhioua1 month ago

Cutting CAD generation to $0.15 shows the power of tightly coupled agents. The Qwen backbone explains the speed boost you demoed. Keep the budget tight, let the agents handle error correction

Bilal Khan's profile picture
Bilal Khan1 month ago

$0.15 per 3D asset is impressive. The real breakthrough is making CAD generation faster, cheaper, and accessible through multi-agent workflows.

Sebastian Buzdugan's profile picture
Sebastian Buzdugan1 month ago

does the $0.15 include geometry validation retries, where these pipelines usually lose margin

Aiden's profile picture
Aiden1 month ago

Text-to-CAD agents are finally moving past demos. The real test isn't the pipeline, it's whether a designer can iterate on the output without fighting the harness.

Pewpy's profile picture
Pewpy1 month ago

these are such basic shapes, a novice CAD user could do these in a short amount of time

Related Videos

so I started with FreeCAD, since i am a linux guy, haven't touched windows since the past 8 years. Also I did this online course from Uni of California Irvine on "building your hardware for your embedded and IoT projects" which had FreeCAD as the main CAD software. FreeCAD was free, and fun, yeah a bit buggy sometimes but gets the job done, since initially all the CAD work i had was just builidng cases, or mounts, or some light hobby stuff. So FreeCAD did just fine. Then I joined a company as a product engineer where I worked with LiFePo4 home energy storage battery packs and was responsible for all the modules of the pack (Cells, BMS, MCUs, Touch Display ) mainly the embedded system. Since I was the product engineer, I'd have back and forth convos with the CAD designers off the battery pack box/casing. I didn't really do any CAD work on that project, tho it seemed really interesting to me. They used Solidworks, and I installed windows for it(swallowing that was difficult); to do Solidworks in my free time which I hardly got. I tried Solidworks and I kinda liked it. But then I quit the job and started working on my SAAS products. A few months into it, it's not fun anymore, slowly getting started with CAD and eventually to electronics to hardware products. Now, I want to go about CAD. I'll be doing a 30 days CAD challenge increasing in complexity and will be sharing daily progress here! It was day 2; I designer a bolt and a nut. I'm going with Fusion 360 for now, coz i like the UI/UX and the hobbyist version is free. So yeah. This is so much fun, i completely lost track of time while doing it. Gemini on the side is alot helpful.

Mustafa

20,127 views • 8 months ago

Introducing LobeHub: Agent teammates that grow with you. LobeHub is the ultimate space for work and life: to find, build, and collaborate with agent teammates that grow with you. We’re building the world’s first and largest human–agent co-evolving network. Two years ago, we built LobeChat, an open-source interface for using different AI models. Today, LobeChat has 70k+ GitHub stars and serves 6M+ users worldwide. How to fully unlock the power of models has always been a shared mission between us and the community. We started with interaction — a fundamentally new, agent-first experience. Agents are no longer passive tools invoked in a single conversation. They should be proactive, always-on units of work. Treating agents as the minimal atomic unit is also the core of our agent harness infra. Today’s agents are mostly one-off executors. Even with memory, it’s often global — and hallucinates. We build long-term agent teammates that evolve with users. Each agent has its own dedicated memory space, editable by users, allowing humans and agents to co-evolve over time. This, in turn, allows us to design clearer rewards for reinforcement learning and create cleaner environments for continual learning. Agent teammates can work in groups. Through a multi-agent system, agent groups operate faster, more cost-effective, and go beyond what single-agent systems can achieve. For example, a single agent often requires heavy user involvement to proceed step by step, whereas LobeHub can execute the same work from a single instruction, with a supervisor orchestrating agents that run in parallel or debate to produce better results. We are building the collaboration network among agent teammates — and between humans and agent teammates as well. Ease of use matters. AI intelligence and shared human intelligence are equally important. With simple instructions and tool selection, you can effortlessly build and team up with agent coworkers to deliver complex, systematic work — even assembling a quant team to execute trades. Through the LobeHub community, anyone can discover, reuse, and remix agents and agent groups, customizing them to fit their own workflows, preferences, and needs. Last but not least, our vision started with LobeChat: multi-model support is the most efficient approach for users. We believe different models excel in different scenarios. By routing across multiple models, LobeHub improves cost efficiency and unlocks capabilities that a single-model setup cannot easily support.

LobeHub

185,401 views • 8 months ago