Company data your assistant can cite.

A file in, the same file back, no row guessed.

Claude sends the rows to resolve, which names the company behind each one or says it cannot, then to enrich, which adds the columns you asked for. A whole file goes to submit_job and comes back as your CSV: every line and column kept, Firmus's columns appended.

# crm_export.csv, as attached
Company;Zip
Danone SA;75009
Boulangerie Martin;
Zzyzx Quantum Holdings;69002

Recorded on 2026-11-01 on a copy of the register's publication of 2026-08-01; six of the appended columns shown. A row the match cannot settle is ambiguous and lists up to three candidates instead of picking one, and a name nothing matches is not_found: neither is a guess. An empty cell is written empty, as in the file.

Four tools, and a source on every answer.

Claude picks the tool from what you ask. Every answer names the register it read and the date of what it says, so Claude can cite rather than assert.

resolve

Looks for the French company behind each row — a name with a postal code or a city, an address, a VAT number, a SIREN or a SIRET. A row it can settle comes back with the SIREN, a confidence and the fields the match rests on; one it cannot comes back ambiguous, with its candidates, or not found. Never a guess.

enrich

Adds the columns you choose to identified companies: name, legal form, activity, workforce band, head office, active establishments, and the latest event of each kind. Each row carries the date the register last changed the company.

timeline

Lists what the register recorded about one company and its establishments, newest first, each event with the date it happened and the date Firmus recorded it.

lookup

Fetches one company by SIREN, or one establishment by SIRET: status, legal form, activity, workforce band, dates and address, with the date the register last changed the record.

A file too large for one call goes to submit_job, which answers with a job id rather than the rows, and job_status, which hands Claude the link to your result once the job is done.

No contact data: no emails, no phone numbers, no people's profiles, and no search by person. Firmus answers about companies, from the public register.

What changed, and since when.

timeline answers with the changes the register recorded: a cessation or a reactivation, a rename, a new activity or legal form, a head office that moved, an establishment opened or closed, a workforce band that changed. Each event says when it happened, when Firmus recorded it, and where it came from.

# one event, as timeline answers it
{
  "kind": "legal_entity.head_office_moved",
  "severity": "notable",
  "occurred_on": "2026-05-11",
  "recorded_at": "2026-11-01T08:51:16Z",
  "source": "sirene_daily",
  "bulk": false,
  "subject": { "kind": "legal_entity", "number": "332976976" },
  "payload": {
    "to": {
      "registered_city": "PARIS", "registered_address": "25 RUE D'ASTORG",
      "registered_postal_code": "75008"
    },
    "from": {
      "registered_city": "MONDEVILLE", "registered_address": "ROUTE DE PARIS",
      "registered_postal_code": "14120"
    }
  }
}

The log is new: it starts no earlier than 2026-10-29, when its daily recording shipped, and nothing older was filled in. A change's own date can be older, when the register reports it late. An empty timeline means nothing has been recorded, not that nothing happened.

The data

The register, counted.

France is the register covered today: INSEE's Sirene.

Between two monthly publications we apply the register's daily change feed every morning — about 33 000 record changes a weekday — so a company that changed yesterday answers correctly today. On every record and every enriched row, data_updated_at is the date the register last changed it, never the date anything here was rebuilt.

For developers

The same answers, over HTTP.

Each tool is a thin wrapper over one endpoint of the API, and a key from the console calls the endpoint directly. The field below searches the register live, on a shared demo key.

# the same search, from your terminal
curl -H "Authorization: Bearer fk_pub_acilFaSqbqW0a1DjYZZwq1IIDPpWAnj4" \
  'https://api.firmus.sh/v1/fr/suggest?q=bnp+pariba&limit=5'

# only fr is indexed today
GET /v1/{country_code}/legal-entities/{registry_number}
GET /v1/{country_code}/legal-entities/{registry_number}/establishments
GET /v1/{country_code}/legal-entities/{registry_number}/events
GET /v1/{country_code}/establishments/{establishment_number}
GET /v1/{country_code}/suggest?q=
POST /v1/{country_code}/resolve
POST /v1/{country_code}/enrich
GET,POST /v1/{country_code}/watchlists
GET,PATCH,DELETE /v1/{country_code}/watchlists/{id}
GET,POST,DELETE /v1/{country_code}/watchlists/{watchlist_id}/watches
GET /v1/{country_code}/watchlists/{watchlist_id}/segments
PUT,DELETE /v1/{country_code}/watchlists/{watchlist_id}/segments/{segment_id}
GET /v1/{country_code}/events?cursor=
GET,POST /v1/{country_code}/segments
GET,DELETE /v1/{country_code}/segments/{id}
POST /v1/{country_code}/segments/{id}/export
POST /v1/{country_code}/jobs
GET /v1/{country_code}/jobs/{id}
GET /v1/{country_code}/jobs/{id}/result

A contract that holds still.

# the legal record behind a pick, trimmed
curl -H "Authorization: Bearer fk_pub_acilFaSqbqW0a1DjYZZwq1IIDPpWAnj4" \
  'https://api.firmus.sh/v1/fr/legal-entities/552032534'

{
  "country_code": "fr",
  "registry_number": "552032534",
  "kind": "company",
  "name": "DANONE",
  "legal_form": { "code": "5599", "scheme": "insee_cj_2018" },
  "activity": { "code": "70.10Z", "scheme": "naf2008" },
  "workforce": { "range_code": "42", "scheme": "insee_tranche_effectifs", "year": 2023 },
  "size_category": { "code": "GE", "scheme": "insee_categorie_entreprise", "year": 2023 },
  "status": "active",
  "registered_on": "1955-01-01",
  "ceased_on": null,
  "registered_address": {
    "line": "59 RUE LA FAYETTE", "postal_code": "75009", "city": "PARIS"
  },
  "partially_disclosed": false,
  "data_updated_at": "2026-04-10T11:28:37Z",
  "country_data": { "sigle": null, "economie_sociale_solidaire": false, "unite_purgee": false, … }
}

Pricing

A flat fee, with the rows included.

Every plan includes a number of rows a month that resolve and enrich look up, counted together: a row that could not be matched is still a row looked up, and a file resolved and then enriched spends its matched rows twice. Once they are spent both refuse until the month ends, and nothing is billed per row.