datamore+
The data lab for
physical AI.
Robots and world models are learning to work. We give them the best teachers: the skilled people who already do the job.
A single format for real-world work data, shared by experts, builders and labs.
Make human skill the most valuable training data in AI, and make sure the people behind it share in what it earns.
Ten fields. Thousands of hands.
One shared library.
in motion.
We are a growing network of experts, tradespeople, operators and researchers capturing how the world gets done, in the open, for every lab.
Join us [+]Ranked by expected demand and pay from robotics and AI labs (est.).
Labs post what their models are missing. Experts record it.
Every brief names a task, a field and the hours needed. Contributors pick a brief, record it on a real job, and get paid when each hour is approved.
- Apply [+]Fit a front door seal and trimAutomotive · 30 hrs · Head + wrist cam
- Apply [+]Replace a front wheel bearingAutomotive · 45 hrs · Head cam
- Apply [+]Route an engine wiring loomAutomotive · 40 hrs · Head + wrist cam
- Apply [+]Assemble a gearbox sub-assemblyManufacturing · 60 hrs · Head + wrist cam
- Apply [+]Inspect and rework a machined partManufacturing · 35 hrs · Head cam
- Apply [+]Build a mixed-case palletLogistics · 80 hrs · Head cam
From a real job to a verified hour, in four steps.
Choose a task from the board that matches your skill and your next job.
Clip your phone to your chest and talk through what you do, start to finish.
Expert check, job outcome, consent and quality, signed and recorded.
Paid for every approved hour, plus a royalty each time a lab licenses it.
1 from datamore import Capture, Structure, Verify, Library 2 3 capture = Capture(device="phone", mount="chest", fps=60) 4 structure = Structure(layers=["steps", "hands", "tools", "depth"]) 5 verify = Verify(checks=["expert", "outcome", "consent", "quality"]) 6 7 def run_pipeline(job: str) -> Hour: 8 hour = capture.record(job) 9 hour = structure.extract(hour)10 verdict = verify.check(hour)11 if verdict.passed:12 return Library.publish(hour, buyers=["robotics", "world_models", "video_ai"])13 return verdict.flag(hour)14 15 run_pipeline("Tile a wet-area wall to spec, start to finish.")
Every hour arrives structured, verified and ready to train on.
Not raw footage. Each recorded hour is broken into the layers robotics and world-model teams actually use.
Each task split into labelled steps, start to finish, in the worker’s own words.
Hand poses, grips and tool contact tracked through every move.
Scene depth and camera motion, so models learn the room as well as the task.
Many workers, sites and conditions per task, not one person on repeat.
Building robots for the real world? Tell us what your models are missing.
Talk to us [+]Built with the people who do the work, not taken from them.
The skill is theirs. Every use of it is tracked, credited and paid.
Nothing is recorded, shared or sold without a clear, signed yes from the worker.
No exclusive owner. Qualified labs license the library on the same terms.
Every dataset carries the trade, the region and the hands that made it.
Every pour, every tie, every fix.
The library grows
with every shift.
1 # For labs: find and license verified hours 2 from datamore import Library 3 4 hours = Library.search( 5 field="construction", 6 task="tile a wet-area wall", 7 layers=["steps", "hands", "tools", "depth"], 8 verified=True, 9 )10 11 for hour in hours:12 print(hour.trade, hour.region, hour.reviewer)13 14 licence = Library.license(hours, use="robot_training")