Resolve the entity
Translate names and local identifiers into the correct legal-entity record.
Canonical company object
Give AI agents company answers they can cite, calculate and decline.
Return resolved entities, ownership paths, source states and explicit gaps through API or MCP—so an agent can explain what is known instead of improvising what is missing.
For enterprise AI, data-platform, risk-product and software teams grounding agents in company and ownership intelligence.
Grounded company responseEvidence assembled
SubjectWho ultimately owns Northstar Trading Ltd?
ResolvedEntity resolved · ownership query executed
Decision questionCan the agent identify the right company, support every relationship and refuse an unsupported conclusion?
Government registry dataLegal entity and two shareholder relationships returned with source metadata
Source factVerified public-source evidenceOne reviewed disclosure supports a control relationship
Separate evidenceZavia calculationTwo indirect paths combine to a 67% effective equity interest
Formula shownUnresolved or conflictingVoting rights are unresolved and must not be stated as control
ReviewWorkflow outcomeAgent answers the equity question and declines the unsupported control claim
One evidence model throughoutExact legal entityNamed source and dateVisible ownership pathExplicit evidence gaps
The operating problem
Retrieval alone does not make a company answer trustworthy.
Names must resolve to legal entities, relationships must retain provenance and indirect ownership must be calculated across a graph. A text snippet cannot carry those controls by itself.
Decision to support
Can the agent identify the right company, support every relationship and refuse an unsupported conclusion?
01
Which exact legal entity does the user's question refer to?
02
Which facts are registry-reported, publicly verified or calculated?
03
Where should the agent stop and return an unresolved state?
How the workflow fits
Move from submitted company to reviewable evidence.
Zavia provides the company and ownership layer. Your team keeps control of policy, judgment and the final decision.
Retrieve the evidence
Return company, shareholder and group data with source metadata.
Typed evidence payload
Traverse and calculate
Follow ownership paths and expose deterministic formulae for indirect interests.
Explainable ownership graph
Respond with boundaries
Return supported conclusions, conflicts and unresolved gaps in machine-readable states.
Grounded agent answer
What your team receives
A decision file—not a data dump.
Every output is built to be reviewed by a person, passed into a system and revisited when the underlying company changes.
Entity-resolution output
Candidate and selected legal entities with jurisdiction, identifiers and match context.
Provenance model
Source type, name, record date and retrieval date attached at field and relationship level.
Ownership graph
Companies, people, edges, percentages, formulas and unresolved branches in structured form.
Change delivery
Monitoring events and webhooks that tell the agent when a prior ownership claim may be stale.
Evidence architecture
Do not flatten four different evidence states into one answer.
The distinction is visible to analysts and available to systems through structured delivery.
Government registry data
The company fact or relationship as reported by the named official registry or government source.
Verified public-source evidence
A named public disclosure matched to the legal entity and kept separate from official registry facts.
Zavia calculation
A result derived from supported source facts, with the entity path and formula available to reproduce.
Unresolved or conflicting
A missing, unclear or inconsistent part of the record that the workflow must not convert into certainty.
The procurement test
What a skeptical buyer will challenge.
These are the controls that prevent a useful company-data workflow from becoming an overclaimed black box.
“The model will still hallucinate beyond the payload.”
Zavia returns explicit supported and unresolved states. The calling application must constrain generation and preserve citations in the final answer.
“A citation at document level is too weak.”
Provenance is attached to the field or relationship it supports, with calculations kept separate from source-reported facts.
“Data rights may not permit agent use or redistribution.”
Delivery and downstream rights must be defined in the commercial licence for the intended workflow. Technical access does not expand usage rights.
Use the same evidence everywhere
Web platform, API, MCP, bulk data and monitoring.
Agreed web, API, MCP, bulk-data and monitoring implementations are designed to preserve the same resolved entities, relationship paths, evidence types, available dates, calculations and unresolved states.
Compare access methodsStructured decision contextCan the agent identify the right company, support every relationship and refuse an unsupported conclusion?
- entity
- resolved
- registry_evidence
- attached
- ownership_paths
- visible
- calculations
- reproducible
- unresolved
- explicit
Frequently asked questions
Scope, evidence and workflow boundaries.
Need to test a specific jurisdiction or policy? Talk to the Zavia team.
Use your own company cases
Test the evidence before you change the workflow.
Start with a legal entity your team already knows. Compare the sources, ownership path and unresolved gaps against your current process.