01 — What MiMi is

A memory, and a cognition that acts on it.

Memory without cognition is a database. Cognition without memory is a language model. MiMi is the union of both.

The graph

The resolved, longitudinal record of the market — entities, relationships and evidence, cleaned, connected, dated and corrected across time. Not storage. A picture of how the market behaves, that improves the longer it is watched.

The cognition

The reasoning that traverses that memory to turn a question into a course of action — with its evidence attached and its confidence stated.

MiMi powers GovIntel, the government-contracting platform, and the same graph answers due-diligence, supplier-risk and credit questions that have nothing to do with federal contracting.

Every month the market produces a public, non-excludable stream of evidence — awards, registrations, dockets, exclusions, terminations, financing statements. A model built on it improves on its own and depends on no single participant. The defensibility is not the raw data, which is public to everyone; it is the resolved, corrected model, which cannot be rebuilt overnight.

02 — The graph

Self-sourced. No purchased data licences.

Every layer is joined onto one resolved company identity, across 71,959,093 companies in 19 countries. Counted against production, not estimated — and rising, because the record does.

197 tables · 77 GB · refreshed on scheduled ingestion. Counts grow daily.

Every company in the graph has a free public page. United States, United Kingdom, Denmark, Japan, France, Estonia, Latvia, Australia, Belgium, Mexico, Israel, Italy, Canada, Poland, Norway, Finland, Czechia, Romania, Ireland. U.S. contractors ranked by federal agency.

03 — What it answers

One question, asked eighteen ways.

Every question below is the same underlying one — who is this company, and what does the record say about them? What changes is which layers carry the answer. None of these are inferred from a score; each resolves to a dated, sourced record.

And the harder version of each: ask it about a company with no federal contract history at all. Most systems have nothing to say about a first-time bidder or a second-tier supplier. Across 265,947 federal subawards and 35,475,794 registered businesses, that is the population MiMi was built for.

Asked most often by:

A credit report and a MiMi dossier get confused for each other. They answer different questions — and if yours is the first one, buy the credit report.

A credit report tells you whether to extend terms. A dossier tells you whether to enter the relationship at all — and what to verify first. We publish no score, because a rating is our judgement to defend and a filing is simply true.

04 — How it answers

Under three laws.

Law 1

Traceable to evidence. Every finding resolves to a sourced, dated, attributable record a human can inspect. No finding exists that cannot be pointed at.

Law 2

Confidence is stated. Every call carries a calibrated confidence, and MiMi says plainly when it does not know. Where a record cannot be attributed to a company with certainty, it is reported as unattributed rather than assumed.

Law 3

It informs judgment. The human decides. MiMi makes the decision better informed and more explainable — it never makes it.

MiMi reports facts with provenance, not proprietary scores. A sourced fact can be checked at its origin. A score is a judgment that hides its reasoning, and it is the wrong instrument for a decision that has to be defended.

05 — Access

Want access?

MiMi is available for due diligence, supplier and counterparty risk, credit and lending, M&A screening, and government-contracting business development. Access is arranged directly — tell us the question you are trying to answer and we will tell you whether MiMi answers it.

Or explore the contracting platform MiMi powers at govintel.ai.