Real-world data for robots that use their hands.
License verified first-person data from the catalog, or tell us what your policy needs and we will collect it. Every item is reviewed by a person, consented and attested on OP Mainnet.
Designed for robot learning
Catalog- first-person video at 30 fps or better
- 1080p+
- of delivered items reviewed by a person
- 100%
- consent recorded and attested
- Per item
Data that holds up when a robot has to use it.
Manipulation policies need first-person footage of real hands doing real work. That footage barely exists on the open web, and scraped video comes with no consent and no provenance.
From brief to bucket in four steps.
When the catalog does not have it, we collect it. Same capture standard, same review, same provenance.
- Step 1
You post a bounty
We scope the brief with you: task, capture standard, target volume and the pay range per item.
- Step 2
Contributors capture
People who already do the work film it on their phones, against your checklist, in their own space.
- Step 3
Every item is reviewed
A person checks each upload. Rejected items never reach you, and you can audit what was approved.
- Step 4
Delivered with attestations
Approved items land in your bucket with per-item JSON and an EAS attestation UID on OP Mainnet.
Delivered the way your pipeline reads.
No portal to scrape and no bespoke loader. Items arrive in formats your training stack already speaks.
Your S3 or R2 bucket
We write straight to a bucket you own, or you pull from ours with scoped, expiring credentials.
WebDataset and Parquet manifests
Tar shards for streaming training, plus a Parquet index of every item and label for filtering.
Per-item JSON
One record per item with capture metadata, annotations, consent and the attestation UID.
Priced by license, not by seat.
Each dataset lists the lowest tier it is offered at. These are the starting points across the catalog today. Volume, refresh cadence and exclusivity move the final number.
Research
Evaluate and publish. Not for products.
- Non-commercial research and benchmarks
- Full items, annotations and per-item JSON
- Attestation UID for every item
- 12-month term, citation required
Commercial
Train models you ship.
- Everything in Research
- Train and deploy commercial models
- Weights you train are yours to keep
- Quarterly refreshes as the set grows
Exclusive
The set is yours alone for the term.
- Everything in Commercial
- Sole license for 24 months
- Withdrawn from the public catalog
- Custom bounties to extend it
Custom collection
Priced per brief: the per-item pay contributors earn times your target volume, plus review and delivery. Send the task and the target and we will quote it.
Tell us what your model needs.
A person on our data team reads every request. Describe the task, modality and volume and we will come back with sample records and a quote.
- 01We read itNo bots, no sequence. The data team replies directly.
- 02You get samplesReal records from the matching set, in the delivered format.
- 03We scope or licenseA catalog license, or a custom bounty with your checklist.