How it works

How an orderruns.

The same six stages whether we’re licensing one codebase or sourcing a thousand CAD models.

Start with a brief →

The process

Six stages. No surprises.

  1. 01

    Brief

    A 30-minute call to pin down modality, volume, intended use and quality bar. We tell you straight away whether it’s on the shelf or needs sourcing.

    You: the gapUs: the plan
  2. 02

    Source

    Shelf data is licensed from the company that holds it. On-demand data is sourced from the firms and experts we find for your project.

    You: the specUs: owners & sources
  3. 03

    Clean and check

    Personal data is removed or de-identified, duplicates are stripped, and sets are screened against public benchmarks. A second reviewer checks the result.

    Us: QA pipeline
  4. 04

    Test sample

    You get representative data to run on your own workloads. The full set only moves once you’re happy.

    You: your evalsUs: the sample
  5. 05

    Contract

    License terms built around the use: pre-training, fine-tuning, evals, exclusive or not. Written plainly so legal can move.

    You: sign-offUs: drafting
  6. 06

    Delivery

    Encrypted transfer in your formats, in batches or all at once, with documentation for every file. We stay on for questions after it lands.

    Us: delivery & support

Lines we don’t cross

What we won’t do.

The fastest way to trust a data vendor is to know where they stop.

  • No scraping.

    Nothing we sell was pulled from the open web.

  • No data of unknown origin.

    If we can’t trace it back to an owner who agreed, we don’t touch it.

  • No raw personal data in transit.

    De-identification happens before transfer, not after.

  • No license beyond what the owner agreed.

    Buyers get exactly the rights the owner signed off on.

  • No benchmark leakage.

    We screen sets against public evals so your scores stay honest.

Security

Locked down end to end.

NDA

From the first call

Owners and buyers are covered before anything sensitive is shared.

ACL

Restricted access

Only the people working on a dataset can reach it, and access is logged.

TLS

Encrypted transfer

Data moves encrypted, straight to the buyer’s storage.

Questions

The practical stuff.

Is there a minimum order?
It depends on the line. Ask on the call. Small pilots are often how on-demand sourcing starts.
Which formats do you deliver?
Each line ships in the format it’s actually used in, mapped to your schema:
  • CodeGit bundles · JSONL
  • Company operationsJSONL · Parquet
  • Egocentric videoMP4 + annotations · WebDataset
  • MedicalParquet · JSONL
  • AudioWAV / FLAC + transcripts
  • CADSTEP · native files
  • Presentation decksPPTX · PDF + JSON
  • Financial modelsXLSX
  • Expert feedbackJSONL
Where does on-demand data come from?
From companies and professionals in the relevant field (engineering firms, advisory teams, clinics, studios, domain experts), found and vetted for each request.
What if the sample isn’t good enough?
Then we either fix it or we part ways before anything is signed. You aren’t committed until the sample works for you.

Ready when
you are.

Bring a rough brief. We’ll turn it into a plan on the call.

Contact us →