NORTHAMPTON, United Kingdom, 24 July 2026 — Global Database today announced the general availability of the Regis API, giving developers programmatic access to Regis — a domain-specific AI model built for company verification, which plans and executes multi-step queries across government registers and returns structured, sourced data.
The Regis API covers more than 600 million companies sourced directly from over 400 government registries across more than 200 countries. Every value it returns carries the registry the data came from, the filing reference number and the moment it was retrieved — attached at field level, rather than as a footnote on the response as a whole.
Deterministic and agentic, on one key
The Regis API sits alongside Global Database's structured endpoints — company records, officers, shareholders, ownership, financials, credit and bank verification, already running in enterprise compliance and risk workflows — rather than replacing them. Together they give customers two complementary ways to reach registry data, depending on how much they already know about the question they are asking.
You already know the entity and the field.
registration_number09410808 country_codeGB datasetshareholdersParameterised endpoints for company records, officers, shareholders, ownership, financials, credit and bank verification. The same input returns the same output every time — predictable, cacheable and reproducible for audit.
The question spans filings, entities or borders.
Who ultimately owns this company, and is anything in the chain dormant?Regis plans which registers can answer each part, calls the tools it needs and assembles the result — for questions that do not map to a single endpoint.
Deterministic queries suit high-volume onboarding, batch enrichment and scheduled monitoring, where the entity and the required field are already known. Agentic queries suit investigations, group structure mapping and open-ended research — tracing an ownership chain across three jurisdictions, for example, where the path is not known before the query runs. Customers can use either mode, or both, on the same API key.
Why the citation matters more than the answer
As enterprises move compliance, onboarding and third-party risk work into AI-driven systems, the requirement placed on business data has shifted. It is no longer enough for a system to produce an answer. Regulated teams have to be able to show, months later and to someone who was not in the room, where each fact came from.
Most business data reaching AI systems today is derived. It is assembled from a range of inputs, standardised, scored and published as a provider's own record. That record can be perfectly accurate — but the citation resolves to the provider's dataset, which leaves one more step between the reviewer and the underlying filing.
Global Database sources directly from the government registries themselves. When the Regis API returns a company status, an officer or a shareholding, the reference it hands back is the registry's own.
"A regulator has never once asked whether the model sounded confident. They ask where the fact came from. So Regis is built so it cannot answer without saying — every value comes back with the register, the reference and the moment it was read."
Built for review, not just for demos
The Regis API was designed around the questions enterprise risk and procurement teams ask before a data source is approved:
- Retrieval, not recall. Answers are assembled from registry records retrieved at request time. The model orchestrates and structures the response; it does not supply the facts from memory.
- Missing data stays missing. Where a registry does not publish a field, the API returns null and names the registry that is silent, rather than filling the gap.
- Customer data is not used for model training. Questions and payloads sent to the API are excluded from training, with the controls set out in the data processing terms.
- Responses are reproducible. The registry, reference and timestamp returned on each field allow any answer to be re-checked against the primary source long after it was generated.
- Licensing matches the use. Internal use, customer-facing display and redistribution are licensed and priced separately, so contract scope is unambiguous.
The engineering problem underneath
Returning a sourced answer across four hundred registries is not a retrieval problem with a model bolted on top. Government registers were built independently, by different states, with no intention of interoperating.
A Companies House number, a German HRB, a French SIREN, a Dutch KvK number and a Delaware file number describe the same class of object using incompatible schemes. Legal forms do not map cleanly between jurisdictions. Names appear in different scripts and transliterations. Resolving an entity across that, and knowing which register is authoritative when two disagree, is the core of what Global Database has built over eleven years.
The second problem is lineage. Most retrieval systems cite at document level — the source they consulted. Regis attaches provenance at field level, threaded through an agent that may have queried several registries in sequence to assemble one answer, so each individual value carries its own registry, reference and timestamp rather than a single citation covering the response. Combined with a deliberate refusal to fill gaps — a field a register does not publish returns null, naming the register that is silent — this is what makes an answer defensible months after it was produced.
The cost of reaching data through a general assistant
Reaching company data through a general-purpose AI assistant requires two commercial relationships: one with the data provider and one with the AI vendor, with token usage billed separately and registry data passing through a third party on its way to the answer. For regulated teams that means a second data processing agreement, a second vendor review and a second answer to give about where the data travelled.
| Data provider + assistant | ||
|---|---|---|
| Commercial contracts | Two | One |
| Data processing agreements | Two | One |
| Assistant seat licences | Required | None |
| Pricing basis | Per token consumed | Per result |
| Bulk retrieval | Via model context | Direct, no inference |
| Data path | Crosses a third party | Single supplier |
Both columns return the same registry records. The difference is how many suppliers sit between the question and the register, and what the retrieval is billed as.
There is a second cost, less visible until the invoice arrives. When company data is pulled through a general assistant, every retrieval becomes an inference: the records travel into the model's context window and are billed as tokens. A single set of company financials runs to roughly 250 individual values per year, so a ten-year history means thousands of values paid for as tokens — to move data that an ordinary API call moves for nothing. Teams do this not because it is sensible but because a connector and a general assistant is the only tool they have, so every request is priced as if it required reasoning.
The Regis API includes the reasoning layer. Requests are priced per result rather than per token consumed, no assistant subscription or seat licences are required, and the question and the records it returns stay within Global Database — one agreement, one supplier on the register. Structured retrieval stays structured, and reasoning is used only where the question actually calls for it.
Three surfaces, one provenance contract
The Regis API is the third way customers can reach Global Database's registry data, alongside the company's MCP server — which exposes the platform as tools to Claude and other Model Context Protocol clients — and the hosted Regis assistant for compliance, risk and research teams working without code. All three run on the same sources and return the same provenance fields.
"Regis plans which registers can answer a question, resolves the same company across identifier schemes that were never designed to interoperate, and follows ownership from one register to the next until it reaches a person. What took the longest was taking capability away — it has no ability to answer from training, only from what it retrieved during the request. Anything can be built to sound confident. Building something that reliably stops is the work."
Availability
The Regis API is generally available today. The same key works across the REST endpoints and the MCP server. Access, committed volume, redistribution rights and on-premise deployment are arranged through the Global Database team.
Request a key and run the questions your team actually asks.
Request an API key Read the docs