Thousands of sources, millions of domains, billions of attributes.One clean record per company.
Our AI engine curates what most teams never have time to assemble: registries, tax rolls, securities filings, sanctions and debarment lists, 118 million web domains, and more. Messy, unusable data in. Matched, scored, and sourced data out.
Thousands of sources. One register.
They reach the register through 92 upstream feeds. Every value keeps its trail.
Official registries
GLEIF · Companies House · SEC EDGAR · SIRENE · CNPJ · Handelsregister and many more. The system of record in each jurisdiction, pulled on its own publishing schedule.
The open web
We crawl the web at scale, continuously. Company websites are found, verified against the record, and read for what filings never carry.
Curated open data
Sanctions and debarment lists, Wikidata, IRS filings, and more. Independent open datasets that corroborate what the registers state.
The coverage atlas maps them jurisdiction by jurisdiction. How we count is on the methodology page.
One graph. 228M+ entities.
Records resolve into a single entity graph. Every edge carries its evidence.
Five stages. One engine.
One pass from raw rows to resolved and sourced entities. Close calls go to human review. Never forced.
Each station shows its operation on a synthetic record: junk stripped from a raw value; a raw string standardized to its canonical legal form; two near-identical rows collapsing into one; two match candidates weighed with the close call routed to human review; verified attributes appending with their sources.
Junk stripped
One canonical shape
Copies collapse to one
Candidates weighed
Cited attributes append
All Entities counts every registered legal form on the register: companies, sole traders, nonprofits, funds, and public bodies. Company Records is the incorporated-company subset of that count.
Right when it matches. Honest when it cannot.
When the engine links a record automatically it is right 99.8% of the time on our public benchmark. Anything uncertain goes to a person, never to a guess.
Watch a record resolve.
Weighted evidence in. One verified record out.
Every merge carries its evidence. The registers and filings and web presence that agree are weighed against each other. The confidence is scored and the record keeps the receipts. Below the threshold a person decides.
Missing values get worked out.
It works like Sudoku. Constraints from everything else we know pin the missing value.
Watch a company record settle fact by fact. Every derived value has to survive each constraint already on the record. Contradictions veto. How a record earns its attributes →
Every value carries its source.
- register
- Ohio Secretary of State
- EIN
- ··-···4821
- NAICS
- 327215 · flat glass
- jurisdiction
- US-OH
- status
- Active
- confidence
- 0.98
One record exactly as it ships. Ask where any field came from and it answers: which source · what date · how sure. Walk a full record →
Certified delivery
- rows
- 12,847
- format
- csv · json
- sha256
- 3f6b19c2…a67d41e
Every delivery ships with a verifiable checksum and every attribute in it keeps its source.
Open identifiers
Identifiers you can look up in the public record. Attached wherever the evidence supports them. How we handle identifiers is documented in the docs.