Owner-licensed company data for AI training and evaluation.
What we supply
The everyday records of how companies work, from single firms and from industry pools of 10 or more firms. Real decisions, handoffs, mistakes and fixes, over years.
- Email threads with staff, clients and vendors
- Team chat from company channels
- CRM activity and notes
- Support tickets with their resolutions
- Code with commit history, reviews and issues
- Project boards, specs and wiki pages
- Field service, dispatch and operations records
- Call and meeting transcripts
Formats
- JSONL
- Per message, thread, ticket or document, with the original folder path kept as metadata.
- Parquet
- For tables.
- Code
- De-identified source files in their original folder tree.
Why license from us
Clean rights.
Every company dataset comes from the company that created it, under a signed license with us. Owners warrant their rights and export with their systems' own admin tools. We buy and license as principal.
Nothing scraped or resold.
We never take scraped corpora, lists from other brokers or data from personal accounts. Public-record sets come only from official government publications.
Measured de-identification.
Release requires at least 99% recall for each identifier type, measured separately in each source, plus a re-identification test by someone outside the deal team.
Checked before it ships.
Duplicates, padding, media filler, secrets, third-party works and excluded categories are removed. Verified volume is stated against claimed volume. We check for AI-generated content and disclose what we find.
Your purchases stay private.
We never show one buyer what another paid. We tell you whether a dataset is already licensed and on what exclusivity, never to whom.
How licensing works
Browse or brief us.
Pick datasets from the catalog, or tell us what you need.
Buyer check.
A short questionnaire on your organization, ownership and intended use.
Sample.
A de-identified sample under evaluation terms, with the datasheet and the provenance file.
License.
On our paper, or on yours with a short rider for any terms it lacks.
Delivery.
Your own copy, by cross-account access or signed links that expire. Every download is logged. You have 10 business days to accept.
License types
Exclusive.
One buyer. Nobody else can license the dataset during the exclusive term, usually 2 years.
Non-exclusive.
Licensed to more than one buyer.
First look.
One buyer gets 2 to 4 weeks to make an offer before anyone else sees the dataset. Teams that ask us for an industry get first look at new data in it.
Data-company license.
For AI data companies. Build environments and task packs from the data and license them to your own customers. You may not resell the dataset as received, and you bind each customer to the same duties we bind you to.
A license can be perpetual and irrevocable, can extend to your affiliates and customers, and models you train never have to be deleted or retrained.
Pricing
Set per deal, by license type, exclusivity, volume and term. Ask for pricing on any dataset.
Samples under evaluation terms
Evaluate before you license. A sample is for evaluation only: no training on it, deletion after your review, and the same duties as a full license, including no re-identification and no contact with anyone in the data.
Provenance you can hand to counsel
Datasheet.
What the dataset is, what it is good for, verified size, sources, years, format and limits.
Provenance file.
The owner shown by code only, staff notice and consent status, and what was excluded and why. Mapped to the EU AI Act Article 53 training-data summary template and California AB 2013.
Chain of title.
We keep the signed documents and show them to your counsel under NDA on request, with the owner's name and identifying details redacted.
Manifest.
A hash for every file.
QC report and de-identification report.
Every check, its result, and the recall measured for each identifier type in each source.
Canary records.
Each buyer's copy carries unique canary records, so a leak traces back to its source.
Our standard terms
Our warranty
We hold the rights, the data is lawful to license, and company data is de-identified to the standard stated in its report.
Your duties
- Use the data to train and evaluate your own models (a data-company license adds your customers' environments and task packs).
- Never re-identify anyone, and never contact anyone in the data.
- Do not resell the dataset as received.
- Do not approach the data's owner about buying data for 12 months after your license starts, except through us.
- Make no onward transfer to a buyer barred by the US Department of Justice rule on bulk sensitive data (28 CFR Part 202).
- Delete specific records when we ask, and report any personal data we missed within 48 hours.
Acceptance within 10 business days of delivery.
Payment 15 to 30 days after acceptance.
Buyer screening
We check every buyer before a sample goes out. We license to buyers in the United States, and to buyers elsewhere only when they are outside the countries named in the Department of Justice bulk data rule and no covered person owns or controls them. Every buyer gets an ownership check, and every sale outside the US carries an onward-transfer clause. We follow the Protecting Americans' Data from Foreign Adversaries Act as well.
Tell us what you need
We also collect data we do not list, from US companies for your order, with scope, volume and timing confirmed in writing before you sign. One of us replies by email. Every buyer passes a short check before any sample goes out.
Questions AI teams ask
Where does the data come from?
From US companies that created it in their own business systems, plus public-record sets from official government publications, marked as such. Each owner signs a license with us, warrants its rights, and exports the data with its systems' own admin tools. We buy and license as principal. Owners appear only as codes.
How do you de-identify it, and how do we know it worked?
Default exclusions first, then text extraction with metadata stripped, then three layers of detection, then consistent placeholders. Release requires at least 99% recall for each identifier type, measured separately in each source on at least 500 hand-labeled cases, plus a re-identification test by someone outside the deal team. The de-identification report ships with every company dataset. The full method is on our Trust page.
What does a license let us do?
Train and evaluate your own models. A license can be perpetual and irrevocable, can extend to your affiliates and customers, and models you train never have to be deleted. AI data companies can take a data-company license to build environments and task packs for their customers.
Can we evaluate before we buy?
Yes. After a short buyer check, we send a de-identified sample under evaluation terms, with the datasheet and the provenance file.
Can you collect data you do not list?
Yes. Tell us what you need. We collect it from US companies for your order, with scope, volume and timing confirmed in writing before you sign. Teams that ask for an industry get first look at new data in it.
Do you tell other buyers what we bought or paid?
No. We never show one buyer what another paid or who else licensed a dataset. We tell you whether a dataset is already licensed and on what exclusivity, so your license is clean.
What if we find personal data you missed?
Report it within 48 hours. We fix the master copy, send a corrected version to every buyer of that dataset, require deletion of the old one, and re-measure.
Do you license to buyers outside the US?
Yes, when they pass our screening. Buyers outside the US must be outside the countries named in the Department of Justice bulk data rule, with no covered person owning or controlling them, and every such sale carries an onward-transfer clause.