← artjoms.ai
PRODUCT · PRIVATE · ON REQUEST

Legal RAG

A private research assistant for law and regulation. Ask in plain language; the answer comes with the article it stands on, and it says so when the law it holds does not cover the question.

8
jurisdictions
invite
accounts
on-prem
deployment
WHAT IT IS

Answers you can check

Legal RAG reads the law before it answers. Each question is matched against the statutes and regulations it holds, and the answer is written only from what it finds, with the article number next to each point.

RAG stands for retrieval-augmented generation: the model does not answer from memory. It first retrieves the relevant articles, then writes from them. That is what makes an answer checkable. You open the article and read it yourself.

When the texts it holds do not cover a question, it says so. Anything taken from general knowledge is marked as such and never presented as a citation, and a rule from one regime is never cited in place of another's.

HOW AN ANSWER READS

Structured like a first opinion

  1. Direct answerYes, no, or it depends, and on what.
  2. Legal basisThe provisions the answer rests on, cited by act and article.
  3. DetailsConditions, thresholds, exceptions and deadlines, as the text states them.
  4. In practiceWhat it means for the case in front of you and what to check next.
  5. ScopeWhat was searched, what the texts did not cover, and where general knowledge was used.

First information, not legal advice. It shortens the way to the right article; it does not replace the lawyer who reads it.

WHAT IT KNOWS

Coverage

8 jurisdictions, indexed article by article:

  • Switzerland: 21 acts, among them the Code of Obligations and the Civil Code, AMLA (GwG) with the FINMA AML Ordinance, FinSA, FinIA, FMIA, the Banking Act, data protection, direct federal and cantonal tax, VAT, withholding tax, social insurance (AHV, IV, BVG, EO), the CDB 20 due-diligence agreement with commentary, and FINMA guidance on stablecoins.
  • European Union: MiCA, the AI Act, AMLR, AMLD6, the AMLA Regulation, DORA and the crypto provisions of CRR3, with GDPR and the EBA ML/TF risk-factor guidelines alongside.
  • United Kingdom: Proceeds of Crime Act 2002, Money Laundering Regulations 2017, SAMLA 2018.
  • United States: 31 CFR Chapter X (Bank Secrecy Act and AML rules), the GENIUS Act, the CLARITY Act.
  • Hong Kong: AMLO (Cap. 615), the HKMA AML/CFT Guideline, the SFC guidelines for virtual asset trading platforms.
  • Singapore: Payment Services Act 2019, MAS Notices 626 and 1014 on AML/CFT.
  • UAE and Dubai: Federal Decree-Law No. 10 of 2025 and Cabinet Resolution No. 134 of 2025, CBUAE AML/CFT guidelines, the VARA rulebooks, the DFSA AML Module (DIFC).
  • International standards: FATF Recommendations, risk-based approach and virtual asset guidance, red-flag indicators; Basel Committee SCO60 on crypto exposures and sound ML/TF risk management; IOSCO crypto recommendations; Wolfsberg correspondent banking principles.

New texts are added as they are parsed. Each answer names the texts it drew on.

HOW IT WORKS

Built to run inside a perimeter

HYBRID SEARCH
Meaning and exact wording together, so a question in everyday words and one quoting the legal term both find the article.
RERANKED
A second model re-reads the candidates and keeps those that actually answer the question.
ARTICLE-LEVEL
Every article is indexed on its own and labelled with act, article, jurisdiction, regulator and language.
OPEN-WEIGHT
Open-weight language models. No external model API; no question or document goes to a third party.
ON-PREM
Runs on a machine you own, in your office or data centre, and can run fully offline.
MAINTAINED
New and amended texts pass through a parsing pipeline (OCR, cleaning, indexing) before they reach the index.
PRIVATE · ON REQUEST

Private workspace. There is no public sign-up: accounts are issued on request, for teams evaluating an on-prem deployment. What it produces in public is the legal digest — free, daily.

Request access

Read by a person. Accounts are issued by hand; you will hear back by email.