Datapublica

Use this when the user names a person associated with Danish companies and you need a resolved person identifier before retrieving details, history, or relationships. Do not use for companies or unresolved registry participants.

search_people

Use this when the target is a registry participant that is not resolved as a company or person, or when the company and person searches cannot represent the registry identity. Do not use as the first choice when the target is clearly a company or resolved person.

search_participants

Use this when the user wants companies similar to one to five known Danish companies and you have their CVR numbers. Rank by business activity, products and services, and customer types, with optional deterministic filters. Results contain rank but no similarity score and are a candidate pool rather than an exhaustive market list.

find_similar_companies

Use this when you have a DAR address identity, such as a company's registered_address_identity from search_companies, and need the property that holds that address. A floor and door resolve to the condominium or building on leased land holding them, otherwise the main property. Returns the property's BFE number, kind, and address; follow with get_property for details.

get_property_at_address

Use this when the user asks which companies are or were registered at one property's addresses (CVR beliggenhedsadresse), including condominiums inside it. Pages of rows with a next cursor; tab current lists open registrations and previous lists ended ones. Registration at an address does not imply ownership.

get_property_companies

Use this when the user asks which Danish properties one company owns. Includes properties held directly and through subsidiaries, with the ownership chain, effective share, address, use, areas, latest valuation and latest sale for each, portfolio totals, and former holdings with their disposal sale and buyers. Lists are capped by limit while totals and formerPropertyCount cover every holding; listsTruncated marks a capped list, and SQL gives complete lists. Effective shares can be ranges because CVR reports ownership in bands. Do not use for a company's ownership graph; use get_company_portfolio instead. The result adds Datapublica overview and properties links for the company; link individual properties as https://datapublica.dk/properties/{bfe}.

get_company_properties

Use this when the user asks which Danish properties one person owns. Accepts a resolved person identifier from search_people or a property owner's personId (ejf_ identifiers are owners known only from the property register). Includes properties held directly and through companies, with effective share, address, use, areas, latest valuation and latest sale, totals, and former holdings, with lists capped like get_company_properties. Owners with name and address protection cannot be looked up.

get_person_properties

Use this before advanced analysis across the full active Datapublica dataset. It provides the tables, fields, relationships, semantic rules, domains, examples, and schemaVersion required to author custom read-only SQL for screening, rankings, comparisons, aggregations, trends, and multi-domain analysis. Do not use it to retrieve company records or execute SQL. The root contract also provides machine-readable company-link templates.

get_sql_contract

Use this when the root SQL contract uses modular delivery and an analysis requires this declared data domain or one of its dependencies. Modules cover company registry data, people and ownership, financials and filings, risk and signals, classifications and geography, capital and investments, property (buildings, units, parcels, valuations), and property ownership (owners, trades and sale prices). Pass the exact schemaVersion from the root contract.

get_sql_contract_module

Use this when the user asks for curated narrative sections from one identified company’s XBRL annual report, covering business and development, risks and material matters, or commitments and security. This is not the complete annual report: boilerplate and cross-scope duplicates are suppressed, contentPlain is capped at 20,000 characters, isTruncated marks excerpts, and there is no pagination. A section’s financialScope can be report_wide; sourceReportUrl can be null; PDF-only filings return no sections; and availableReports[].locked is always false for token-authenticated calls. Supply financialScope when no scoped numeric annual filing can provide a default. Do not use it for numeric financial comparisons or full-report retrieval. A resolved CVR adds Datapublica overview and financials links; preserve sourceReportUrl as claim evidence.

get_company_annual_report_sections

Use this when the user asks about capital increases, reductions, funding events, subscription prices, implied valuations, provenance, or attributed investors for one identified company. Do not use for recent investment activity across many companies; use get_investment_feed instead. A resolved CVR adds Datapublica overview and investments links; preserve event source URLs as evidence.

get_company_capital_events

Use this when the user asks how one identified company changed over time, including registry, role, ownership, auditor, and filing events. Do not use for the company’s current profile or cross-company event analysis. Events with a resolved CVR add Datapublica overview and timeline links; preserve filing/report URLs as event evidence.

get_company_timeline

Use this when the user asks which properties one company bought or sold. Returns each registered trade the company took part in directly, with its side (buyer or seller), share, the property, price, dates, method, and counterparties. Prices cover the whole sale; propertyCount above 1 means several properties.

get_company_property_trades

Use this when the user asks for chronological registry activity involving one unresolved participant identity. Do not use for the participant’s relationship graph.

get_participant_timeline

Use this when the user asks how a resolved person’s company roles, ownership, or other registry activity changed over time. Do not use for the current relationship graph; use get_person_portfolio instead.

get_person_timeline

Use this when the user asks which properties one person bought or sold. Accepts the same identifiers as get_person_properties and returns each direct trade with side, share, the property, price, dates, method, and counterparties. Prices cover the whole sale; propertyCount above 1 means several properties.

get_person_property_trades

Use this to run validated custom, read-only DuckDB SQL that may scan the full active Datapublica dataset and return bounded results. Appropriate for company and property screening, rankings, comparisons, aggregations, trends, and analyses combining registry, financial, ownership, people, industry, geography, risk, Gazelle, capital, investment, property, valuation, and property sale data. Do not use unvalidated SQL, write statements, or repeated entity lookups when one analytical query is more appropriate. Select cvr_number when presenting individual companies and construct links from the contract templates without extra entity lookups; never invent links for aggregate-only results.

execute_sql

Use this only as a fallback when a complex Danish or English analytical question cannot be converted confidently into SQL from the active contract. It can translate screening, ranking, comparison, aggregation, trend, and multi-domain questions, but consumes natural-language credits and does not execute the result. The question text is sent to OpenAI under API terms with storage disabled and is not used for model training; see the Datapublica privacy policy.

translate_to_sql

Use this after authoring or modifying custom DuckDB SQL and before execution. It validates full-dataset analytical queries against the active schema, read-only safety rules, semantic domains, financial scope, currency rules, and plan access without running them.

validate_sql

Use this when you already have a stable identifier for one Danish company and need its complete company profile, including registry details, status, addresses, industries, roles, ownership, filings, financials, and available risk fields. companyRelations.fullyLiableParticipants lists the holder of an enkeltmandsvirksomhed or PMV (100%) and I/S partners or K/S general partners (share not registered); they are separate from legalOwners and beneficialOwners. employment exposes legal-entity CVR headcount and full-time equivalents (FTE/årsværk) with latestHeadcount, latestFte and observations, each carrying exact values or intervals, period, frequency and source. These are separate from annual-report average_employees and group figures. Latest metric periods can differ; compare matched periods and preserve missing values. Do not use to find an unknown company or compare multiple companies; use search_companies or SQL instead. The result includes Datapublica navigation links; destinations do not imply records or paid access in every section.

get_company

Use this when you have a stable participant identifier and need the registry bundle for an unresolved participant identity. Do not use for resolved companies or people.

get_participant

Use this when you have a stable resolved-person identifier and need that person’s roles, ownership, auditor relationships, or related companies. Person profiles do not contain contact details. Do not use a display name as the identifier; call search_people first when necessary.

get_person

Use this when you have one property's BFE number and need its full record: address, municipality, main use, land and floor areas, plot ratio, BBR buildings, units and technical installations, cadastral parcels, condominiums of a divided main property, current and previous owners with shares, public valuations by year, sales with prices, buyers and sellers, and a combined history. Sale prices cover the whole sale; propertyCount above 1 means the price covers several properties. Owners with name and address protection appear as type protected without identity. Geometry is omitted unless includeGeometry is true.

get_property