Custom collection

You write the spec. We collect it.

Commission demonstration data to your exact specification — from 100 to 5,000 hours. You define the tasks, environments and rejection criteria; we commission the collector community and deliver to spec, in batches, with a QC report on every drop.

How it works

From spec to delivery, in five steps.

Scoping call

We map your task taxonomy, target environments and output tier together, so the spec reflects what your model actually needs to learn.

Written spec & rejection criteria

Tasks, environments, distributions and per-clip rejection criteria are fixed in writing, so acceptance is unambiguous on both sides.

Commission the community

The spec goes to the collector community as accepted tasks. Collection runs in parallel across contributors and locations.

Pilot batch

You review a pilot batch and sign off before the full run, so any spec adjustments land early rather than at delivery.

Delivery in batches with QC report

Hours ship in batches in your format, each with a QC report: acceptance rate, rejection reasons and reviewer distribution.

What you control

The spec is yours. Every dimension below is set by you and enforced through QC.

  • ✓Task list — the specific manipulations and activities to capture
  • ✓Environments — kitchens, workshops, retail floors, offices, and more
  • ✓Demographic and geographic distribution across the collector community
  • ✓Rejection criteria applied per clip at the QC gate
  • ✓Output tier — Raw + filtered, Structured, or VLA-ready
  • ✓Exclusivity — keep the resulting set to yourself where required

Timeline: weeks, not quarters

Because collection runs in parallel across the community, a commission delivers in weeks rather than the quarters a teleoperation build takes. How fast depends on how narrow the spec is.

Pricing: quoted per project

Custom commissions are scoped and quoted per project. Volume and exclusivity discounts apply. You only pay for hours that clear QC.

Ready to scope a commission?

Book a 30-minute call. We'll map your task taxonomy, environments and output tier, and turn it into a written spec.

Book a scoping call