From Lid Cleaners to Everything
I built this because I needed it.




At Growbotics I designed 5+ production machines. A single machine took 3 days in Fusion 360. I described that same lid-cleaning system to an early prototype of OpenFrame. It designed the complete assembly in 30 minutes.
The crane
This crane is a 3-axis gantry — same kinematics as a CNC router. That’s when it clicked. Every machine I built uses the same language.
Same language, any physical thing
LID_CLEANER
cleaning automationAutomated lid sanitization station. Multi-head brush array traverses a linear rail, scrubbing greenhouse container lids between grow cycles.
Three machines I designed. The architecture doesn’t know the difference.
What's real today
- 275 parts — real McMaster-Carr catalog, STEP → GLB pipeline
- 20 port types — Shaft, Bore, Rail, Carriage, Face, BoltCircle, etc.
- Multi-model compose — Sonnet reasoning, Haiku parallelization, vision critique
- Vision-in-the-loop — 13-point quality jump, $0.015, automatic
- 3 execution surfaces — Studio, MCP (Cursor), headless API
- Event-sourced undo — 76 lines, full collaborative editing
What's not
No kinematics simulation. No FEA. No G-code. No wiring diagrams. No domain beyond mechanical assembly. But the architecture doesn't care.
The market
50 compositions/mo
Unlimited + API
White-label + BOM commissions
TAM $10.8B. SAM 5M active users. SOM 200K serious builders. 1% of SOM = $1.2M ARR at $49/mo.