95 karma · joined June 11, 2025
Used about 200k tokens in 30 minutes. You can view the generated implicit model here: https://www.cad.fun/?file=implicits%2Fpelican-bicycle.implic...
This is: 1. A small library of SDFs with a browser-native renderer: https://github.com/earthtojake/implicit.js 2. A demo that renders 3D models from code in realtime: https://www.implicit.sh/ 3. A set of skills to write code to generate 3D models with agents: https://github.com/earthtojake/text-to-cad
Some awesome things about implicit CAD: • No CAD kernels (Parasolid, OpenCascade, etc.): the math + code is the model. • They are FAST: the https://www.implicit.sh/ demo compiles and renders models from GLSL in the browser in realtime. • Representations are small: implicit code can be 100x smaller than STEP files.
Implicits are great at: • Manipulating complex geometry • Producing high resolution, scaled CAD models • Boolean / offset operations
But they are really hard for humans to intuit, which is why we prefer vanilla CAD that works more like LEGO: faces, edges, holes, cuts, fillets, etc., with human-readable topology.
Agents, on the other hand, are very, very good at coding and math — the two key ingredients for implicits.
The third ingredient is visual / spatial reasoning. While lacking today, I believe for many reasons (including robotics) this will be the next frontier that labs pursue.
To caveat: agents aren’t great at implicits today. Frontier models are still pretty blind, so implicits break down on complex parts/assemblies with visual feedback (code → model → screenshot → repeat). There is also no easy toolchain to inspect implicits because they lack topology.
That said, 6 months ago the models couldn’t write reliable code. Things are changing fast, and I think 2026 will be the year the models solve 3D design.
I also think it’s possible (and even probable) that implicit CAD becomes the preferred way for models to do 3D design.
Special thanks to Blake Courter and the Gradient Control team for teaching me about implicits and sharing the Python primitives used to build this library. Blake is the GOAT of implicit CAD and gave an excellent talk on this exact topic at Kinetic a few weeks ago.
Many CAD tools like Autodesk have huge closed libraries of standard mechanical and electrical parts.
The goal is to build an equivalent catalog for the open source CAD community (that’s easy for humans and agents to use).
The directory is seeded from dozens of existing open source catalogs and generators, catalogued and organised by family, standard, size, etc.
The directory also includes an API, llms.txt and skill to make it easy for agents to download relevant STEP files in generative CAD tools. I’ve added the skill to my text-to-cad repository:
I’m hoping many can contribute to the repo and expand the directory to hundreds of thousands of standard parts!
Happy building
I actually used GPT 5.5 Pro to generate the prompts from simpler one sentence prompts, so hypothetically it’s just an extra step in the harness for an agent to unpack / add detail to a prompt based on the user’s goal.
I'm brushing up on robotics after spending the last 10 years working in software land. After being humbled by modern CAD tools like Onshape, I built this harness / skill to help me generate some basic CAD models for a 7dof robot arm I'm designing.
It ended up working much better than I expected, particularly on the latest GPT 5.5 and Opus 4.7 models. It's been a lot of fun to work on. I've learned a lot about how STEP files work (opencascade, breps, etc) as well as 3d rendering tools like threejs.
I don't have much intention of turning this into a business, it's really just a fun open source tool that I'll continue to maintain as long as myself and others find it useful. Very open to ideas and contributions.
P.S. I just pushed a major update that improves the workflow and scripts/tools for the CAD skill. I also added some basic benchmarks to start measuring performance over time.
There's also a bunch of work going into the SKILL.md to plan for more complex parts (this is mostly a stop gap while the models don't have amazing spatial reasoning).
Use it to prompt and edit complex 3D models. Export STEP, STL, GLB, DXF and URDF files. Built for CAD newbies. Link to github below.
CAD is hard. As a software engineer getting back into robotics, I’ve been humbled by new tools like Onshape. Struggling to kick old habits, I started prompting Codex to generate 3D models and had some limited succes. After a few iterations I found a recipe that actually works:
1. Generate a python script for every STEP file. The agent can easily edit each part’s source without touching the raw STEP file. Use build123d > cadquery. 2. Reference specific faces and edges in prompts for precise edits. I built a basic local ui to inspect / cache STEP B-reps to make this easier. 3. Maintain markdown explaining important part features in plain English so the model can index on project context quickly. 4. Verify results with screenshots and geometry. Models don’t have great spatial awareness, but they can interpret images and verify constraints very well.
For the best results I’ve been using GPT 5.4 xhigh / Opus 4.6+. Fair warning, this will burn through tokens, I recommend the Pro/Max plans if you’re planning on building anything serious. PRs welcome!
I tried a few different implementations (e.g. using background video/audio), but ultimately the device’s unpredictable management of background apps made it impossible to distinguish between navigating away from the app and locking the phone. I ended up going with the “dumb” solution.
It actually works quite nicely (especially as a PWA on iOS) because it makes the stretch feel more intentional, and it discourages “gaming” the system by recording long stretches while sleeping.
I’ve found the battery drain is also pretty negligible over 2-3 hour stretches, the dark screen / pixels seem to help a lot.
For the first week we were practically glued to our phones, contacting family members, insurance companies and air ambulance services. I found myself obsessively checking my phone for updates, sending empty messages and mindlessly scrolling feeds. My screen time reached all time highs. I was spending 12 of my 16 waking hours staring at a screen instead of being there for my wife and her parents. It felt like I was hiding on my phone.
I don’t have a particularly addictive personality, but I am undeniably addicted to my phone. And this was the week I finally needed to deal with it.
I tried Apple Screen Time and a few popular screen time management apps, but found the blocks were too easy to bypass. I also realised that most apps (e.g. YouTube) were as useful as they were distracting depending on the context. I didn’t necessarily want to use my phone less: it’s an incredibly useful tool, and the distractions were sometimes helpful.
What I really needed was intentional stretches of time spent away from my phone. I built touchgrass.fm as a simple way to record and incentivize those stretches of time. It’s not quite finished (built in a few hours of downtime), but it helped me stay present during hospital visits, meals and important conversations.
I decided to share it on the off chance it helps others get some control back and be a bit more present in their day to day lives!
Link: https://www.touchgrass.fm
What Are You Working On (June 2025): https://news.ycombinator.com/item?id=44416093#44427955
Brief backstory: While visiting us overseas, my in-laws were in a very bad car accident. Everyone involved is alive and going to be okay. But what followed was a series of emotional, physical and logistical challenges that pushed my wife and her parents to their limits.
During this time I found myself (shamefully) hiding on my phone. I was obsessively refreshing for updates from insurance/hospital teams, sending empty messages, and mindlessly scrolling feeds. My screen time was averaging 12 hours a day. Time I could have spent being fully present with my wife and her parents.
I finally accepted I have a serious phone addiction. I tried Apple Screen Time and a few popular screen time management apps, but found the blocks were too easy to bypass, and some apps were as useful as they were distracting depending on the context (e.g. YouTube). I didn’t necessarily want to use my phone less: it’s an incredibly useful tool, and the distractions were sometimes helpful.
What I really needed was intentional stretches of time spent away from my phone. I built touchgrass.fm as a simple way to record and incentivize those stretches of time. It’s not quite finished, but it’s been helping me stay present for hospital visits, meals and important conversations.