Légifrance (unofficial)

Describe what this server covers, how it is sourced and what it does NOT cover. Call it before telling a user that the law "does not contain" something, and whenever a search comes back empty: the gap may be in this server, not in the law. Each gap comes with a fallback. It also reports the PISTE environment (sandbox or production), whether the configuration is complete, the data source and licence with today's consultation date, and the note on ELI identifiers. No arguments, no network call.

Compare the versions of a code or law article (LEGIARTI) in force on `date_a` and `date_b` (YYYY-MM-DD; `date_b` defaults to today in Paris). Returns both versions with their validity dates and `etat`, and a compact word-level diff: removed and added passages with a little context. `date_fin` is the day the next version (or the repeal) takes effect, absent when there is no end date; `etat` is the Légifrance code (ABROGE_DIFF: still in force, repeal scheduled). `identical` is true when both dates fall in the same version. To read a single version use fr_get_text.

Read a consolidated law, decree or code by its LEGITEXT id (from fr_search on LODA or CODE): metadata, citation, source_url and a table of contents of sections and articles with their `article_id`. Returns the version in force on `date` (YYYY-MM-DD; default: the version named by a suffixed text_id, else today in Paris); without `date_fin_version` that version has no end date. The table of contents is paged: up to `toc_limit` entries (default 100) from `toc_offset`; continue with `next_toc_offset`. It does not return article text: read an article with fr_get_text. For a collective agreement (KALITEXT) use fr_get_convention.

Read a company-level agreement (accord d'entreprise) by its ACCOTEXT id from fr_search on fond ACCO. Returns METADATA ONLY: company, SIRET, IDCC, APE code, sector, themes, signatory unions, dates, citation and source_url. Légifrance distributes the full text only as a .docx attachment, which this tool does not return: point the user to source_url.

Read a collective labour agreement text (convention collective, avenant, annexe) by its KALITEXT id from fr_search on fond KALI: title, IDCC, parent convention, status, citation, source_url and a table of contents. Article entries carry KALIARTI ids: read them with fr_get_text. Page the table of contents with `toc_offset`/`toc_limit` (default 100) and `next_toc_offset`. For a company-level agreement use fr_get_company_agreement.

Read a court decision by id: JURITEXT (Cour de cassation, courts of appeal), CONSTEXT (Conseil constitutionnel) or CETATEXT (Conseil d'État, administrative courts), as found by fr_search. Returns court, formation, solution, case numbers, date, the native `ecli` when Légifrance has one (never invented), citation and source_url. By default it adds a short `summary`: the official summary when published, else the operative part or the beginning of the text; a short decision comes back whole in `text`. Use `detail`="full" with `offset`/`max_chars` (default 4,000 characters) to read the text page by page (follow `next_offset`), or `focus` to get the best matching passages (default 3).

Read a CNIL deliberation or decision by its CNILTEXT id (from fr_search on fond CNIL): number, nature (e.g. Sanction), status, dates, citation and source_url. By default a text of about 1,200 characters or less comes back whole in `text` and a longer one as a `summary`: the operative part of a sanction or formal notice, else the beginning of the text. `detail`="full" pages through the text with `offset`/`max_chars` (default 4,000 characters) and `next_offset`; `focus` returns the most relevant passages (default 3) and takes precedence over `detail` and `offset`. Personal data is redacted by the CNIL in published texts.

List the legal links of an article (LEGIARTI) or a decision (JURITEXT, CONSTEXT, CETATEXT): texts it cites, texts citing it, and case law it refers to. Article links come from Légifrance with their ids; decision links are parsed from the decision's references, have no id and may carry an inexact code: pass one to fr_resolve only if its citation names a code. Each link has a `relation`, a `citation` and, when known, an `id` and `source_url`; `relation` as a parameter keeps one kind of link. At most 100 links are returned: `truncated` is true when there are more, and there is no next page.

Read one article by id: LEGIARTI (from fr_get_act or a CODE search hit) or KALIARTI (from fr_get_convention). Returns its citation (for example "Code civil, art. 9"), source_url, status, validity dates and parent code or text. By default (`detail`="summary") an article of up to 2,000 characters comes back whole and a longer one as a summary of about 1,200 characters (`summary_truncated` is true when the summary stops before the end of the text). A summary is never the whole text: use `detail`="full" before quoting a threshold, amount or rate; it returns up to `max_chars` (default 4,000) from `offset`, and `next_offset` tells where to continue; `focus` returns the most relevant passages instead (default 3). Each LEGIARTI id is one version of the article: `date_debut` and `date_fin` bound its validity; without `date_fin` it has no end date, and without `date_debut` its entry into force is not set yet (see `etat`).

Turn one free-text legal citation into Légifrance ids, for example "article 1240 du code civil", "loi n° 78-17 du 6 janvier 1978", "Cass. civ. 1re, 4 juin 2025, n° 24-13.470" or an ECLI. Returns `candidates` (id, kind, citation, source_url, confidence high/medium/low), best first; nothing is guessed or fabricated, and an empty list comes with a `note` saying why. Pass `date` (YYYY-MM-DD, default: today in Paris) for the article version in force on that day. Then open a candidate with the fr_get_* tool for its kind; to check all citations of a text at once use fr_verify_citations.

Keyword search in one Légifrance fond (default LODA): laws and decrees (LODA), codes (CODE), case law (JURI, CONSTIT, CETAT), CNIL deliberations (CNIL), collective agreements (KALI) or company agreements (ACCO). Returns one page of hits (default 10) with id, title, snippet, citation and source_url; ask for the next `page` while `has_more`. `match` defaults to any word: use `all` for several words, `exact` for a phrase, fr_resolve for one known article. `date_from`/`date_to` filter by date; for KALI, `in_force_only` keeps the texts in force (not their articles' status). LODA and CODE hits are the versions in force on `version_date` (default: today in Paris); a hit `date` is the date Légifrance gives for that fond, not always the act date. Open a hit with the fr_get_* tool for its id prefix (LEGITEXT: fr_get_act, JURITEXT/CONSTEXT/CETATEXT: fr_get_decision, CNILTEXT: fr_get_deliberation, KALITEXT: fr_get_convention, ACCOTEXT: fr_get_company_agreement); CODE hits list `articles`.

Search several Légifrance fonds at once (default LODA, CODE, JURI, CETAT, CONSTIT) and return the top `per_fond` hits of each (default 3, max 10), grouped by fond with their totals. Use it for a first overview when you do not know where the answer is; then refine with fr_search on the relevant fond, which supports paging and filters. A fond that fails yields an error entry without failing the others.

Check every legal citation found in a text of up to 20,000 characters, for example a draft answer, against the official Légifrance record. Recognises code articles (a bare number needs article/art.; one prefixed by L, R or D does not), laws and decrees, court decisions, ECLI, CNIL deliberations and collective agreements; found=0 does not prove that a text cites nothing. For each citation returns its `span`, a `status` (verified, not_found, ambiguous, unresolvable or invalid), the official record when found and a `note` explaining any mismatch; `verified` means the article exists, not that it says what the draft claims. A quoted excerpt is compared with the official text (`quote`) only when it is glued to the citation: no `quote` means NOT checked. It checks the first `max_citations` (default 50); if `truncated` is true, call again on the rest of the text. Use it before giving the user an answer that cites law; for a single citation fr_resolve is lighter.