What is Quarry
A broker and exchange that licenses hard-to-get data from the businesses that hold it to frontier labs.
Quarry is the data supply company for frontier AI. It sources, licenses and verifies data that labs can't get from the open web, from the people and companies who own it, and delivers it rights-cleared and ready to train on.
A broker, not a data owner
Quarry doesn't own the data it licenses. It matches lab demand to owners (BPOs, call centres, hospitals, diagnostic chains and platforms), then runs everything between the two:
- Sourcing. Finding the owners who hold what a lab asked for, and negotiating the licence.
- De-identification at the source. Personal data is removed inside the owner's environment, so it never leaves their systems.
- Licensing. One supplier agreement with the owner, one licence with the lab, and a manifest that ties every record to both.
- QA and delivery. Checks on every record, a datasheet, and delivery to the lab's storage.
Launch verticals
| Vertical | What it is | Spec |
|---|---|---|
| Conversational audio | Call-centre recordings across healthcare, finance and e-commerce / food delivery, in Indian languages and Indian English | Conversational audio |
| Medical imaging | CT, MRI, mammography and X-ray studies as DICOM with the radiology report, and EMR where the source can link it | Medical imaging |
For anything else, labs post a request on the demand board and Quarry looks for owners who hold it.
What every dataset comes with
- A datasheet: what the data is, how it was made and its limits. See the template.
- A per-record provenance manifest: one JSON record per item. See the schema.
- A de-identification report: what was removed and how it was checked. See reports you receive.
- A sample under NDA, so you can judge fit before you license.
Where to go next
- Buying data: How licensing works.
- Holding data: How listing works.