Law, understood.
We build AI systems that reason about law the way a lawyer does, starting from statutes and case law.
French law is unlike any other. It is written and codified, split between two separate court systems, and its decisions are extremely terse: a single citation from the Cour de cassation or one paragraph from the Conseil d’État is often enough to set a rule.
General-purpose models have learned mostly from English texts and common-law sources. They can read French law, but they struggle with the link between a code article and the judgment that applies it, or with what a single adverb changes in a court’s reasoning. We train our models directly on the sources of French law.
They learn from public legal texts and are tested on tasks designed by lawyers, which they never see during training. Our first models already outperform the best open models in French legal research.
Our method reaches beyond France. The 1804 Civil Code served as a model for many civil-law systems, from Belgium to Quebec and from Lebanon to French-speaking Africa. They have kept its structure, so the same approach carries over to them.
Our models will be released openly. A second generation, a public benchmark and a research paper will follow.
Alinea Labs was founded by a French public-law jurist who studies at the École normale supérieure (ENS-PSL) and at Université Paris 1 Panthéon‑Sorbonne, and who previously worked at the National Assembly and at the General Secretariat of the Government, the Prime Minister’s legal coordination office.