I wanted to build a bed frame recently and imagined software where I could choose lumber from a list and visually assemble the bed frame in the software how I see fit, choose stains, flourishes - then go order the wood and have it cut. At that point it's like taking home an IKEA bed, except of way higher quality materials and to your exact design spec.
So
> Real SKUs, a real BOM you can send to any lumberyard; understand exactly what you need down to the last nail.
Is especially interesting, it sounds like you could basically do that with this software.
Turning the woodworking process into code to be automated by an LLM is pretty ambitious, and is the part I am most skeptical about. Of course, if you get that right (plus adoption) it would be unbelievably huge.
And you can customize it beyond the existing design by asking an agent to modify it, e.g. "let's use walnut 4x4s and 2x4 slats", "I don't like how the corners come together, can we make it more like traditional joinery?" Also, staining is a supported procedure (it's very useful for e.g. coloring geodesic dome struts by strut type), you can just ask the agent to stain the pieces with whichever color you like, I'm sure you could even use a named stain you want to use like "Minwax Dark Oak" and get something pretty close.
Impressive stuff. Actually more impressive than I even imagined.
“Fits in my SUV” you are a mad man! Love everything about this site. I might actually do it now haha - no more excuses.
A couple of years ago I did a small woodworking project and used my vr headset do do the design. It was really fun to iterate on the piece in 3d space, but the software i used was not made for that so some actions were rather tedious. An agent, with voice support for the actions that are not fun to do with vr controllers, would make it awesome.
>A couple of years ago I did a small woodworking project and used my vr headset do do the design. It was really fun to iterate on the piece in 3d space, but the software i used was not made for that so some actions were rather tedious. An agent, with voice support for the actions that are not fun to do with vr controllers, would make it awesome.
Yes! I played with a SketchUp VR extension a few years ago and it was so close to cool, but in reality practically useless because of poor controls. I lack VR hardware right now, but am building with that surface in mind.
Feel free to reach out! Email in profile.
If this does what I think it does (and I confess I’m a goddamn idiot still when it comes to anything involving ‘agentic’) this is honestly pretty damn amazing.
I keep seeing this word. Every time, I have to look it up; I marvel at the fact that we have such a cool word describing a complex but interesting topic; and apparently forget it again. And then I never find a situation where I can use it.
I also suspect there’s a fairly finite number of sane join primitives, you can have rabbets, dados, dominos, dovetails, pocket screws, etc. These could be experimented with digitally against an existing design. And your tool availability could determine what you can do.
And of course cut order and assembly is just a DAG and can be generated from geometric constraints of the actual design.
> And of course cut order and assembly is just a DAG and can be generated from geometric constraints of the actual design.
That's exactly it! It's a well-constrained space, and some amazing utility falls out as a consequence of using real-world materials and operations instead of cad-native primitives.
This was all pre AI. Super impressive that you’ve made something awesome. I’ll try it out tonight
The MCP server seems to have a bug. The tool list flashes for a second in Claude Desktop, but then immediately states "This connector has no tools available." and the error "Couldn’t reload tools from the server." when I try to refresh it.
I spent the last 5 years working at a bn+ valuation YC fintech; I naturally started to apply to YC with this project, but kept feeling internal pressure to make it something different from my vision in order to get to that bn+ unicorn status. Not every cool idea needs to corner an industry, ya know? I'd rather build this tool for 100 people like me and not gate a bunch of features behind paywalls. I live off-grid on acerage in Hawaii and paid off my mortgage and car loan, so my income needs are low. It would be nice to replace that 6 figure SV income one day, but I have a young family. There are different kinds of wealth, ya know?
[0] https://novc.fyi
Please consider adding metric measuremts support though. My brain can't work in fractions of an inch.
If you are looking to improve the mcp, you should try adding a UI to it (see https://apps.extensions.modelcontextprotocol.io/api/). And the server happens to be in python, you should check out: https://mplemay.github.io/belgie/examples/mcp/. It is a project I have been working on which makes it easy to build MCP UIs from python.
It's been really fun to build with. Agents can create parametric procedures so e.g. stick framing becomes a simple function call for the next agent, building a flywheel. There is full agent-human parity in every surface, so there is a human UI for the procedures too.
Every operation is yaml, so agents pick it up pretty quickly. Agents can also file feature requests when they reach for a tool that doesn't exist. You can export to 3d printing files, video walk-through, step files, get a BOM and a cut plan, view your project in life-size with AR. There is a notion-like interface for authoring build guides... a lot going on. If you're a woodworker/software engineer you might enjoy it.
You can add the MCP to your agent of choice, send a picture, description, sketch or whatever; say "build this with Sawdust"
https://sawdust.diy