Executive Summary

We’re building fast, reliable robot workers for warehouses and factories.

Initial market:

Independent sneaker and apparel warehouses.

Fundraise:

$15M

Our approach

What we have so far

Foundation models

Execute each task quickly and accurately

What we have so farBreakthrough in reinforcement learning. Our models execute tasks up to 7X faster than π_0.5. On unseen tasks, they achieve 3X the success rate (27% → 84%).

Agentic harness

Complete the whole job reliably

What we have so farTool use raised our success rate for folding a scrunched-up towel from 0% to 100%.

Jev-style models monitor and verify each warehouse pick in real time. With multi-robot coordination, we get 95+% reliability in simulation.

Deployments

Fast, reliable work in real life

What we have so farThree warehouse owners committed to development space and pilot deployments for picking and packing (an industry average of 80% of warehouse labor spend).

We’re building fast, reliable robot workers at a price people will pay. We train our own models for fast, accurate movement while retaining earlier skills. In separate early simulation tests against PI’s π0.5, we saw up to seven times faster execution on a drawer task and 84% versus 27% success on unseen tasks. With tool use, our success rate on folding a scrunched-up towel went from zero to 100 percent. Jev-style models monitor and verify each warehouse pick in real time. With multi-robot coordination, we get over 95 percent reliability in simulation. Three warehouse owners have committed to development space and pilot deployment. Their workflows guide what we build. We’re starting with picking and packing at independent sneaker and apparel warehouses. Across the warehouse industry, picking and packing account for 80% of labor spend on average. The broader cross-border market has more than 3,200 U.S. warehouses, and we estimate roughly a billion dollars in annual revenue opportunity there. We’re raising a $15 million seed to prove the economics over a full shift, convert the first pilot into a paid deployment and repeat at the next warehouse. Model benchmarks are founder-provided. The speed figure is from a simulated drawer task. The unseen-task result is from a separate zero-shot MolmoSpaces evaluation. Earlier-task performance loss was under 5% in a separate retention test. In the MuJoCo warehouse demo, the planner called the pick policy and the harness visually checked the box on the cart. See partner-meeting-prep/review/slide-6-demo-review.json. Three warehouse owners have verbally committed to development space and pilot deployment. Customer revenue, full-shift performance and repeatability are milestones for this seed round. Market estimate from deck slide 8. The broader U.S. cross-border ecosystem includes 3,200+ warehouses. Luke’s midpoint model assumes 3,200 sites × seven robots per site × $48,000 per robot per year, or $1.075B annually at full penetration. Sneakers and apparel are the starting category within this market. Sources: partner-meeting-prep/shared/question-prep-plan.md; partner-meeting-prep/shared/slide-by-slide-talk-track.md (slides 5, 8, 11, 12 and 16); partner-meeting-prep/review/pitch-deck-content.json.