Built so it can’t hallucinate.
Lex.NY is a research engine for New York law. It is supervised by a New York–licensed attorney (SDNY/EDNY). Its architecture makes it physically incapable of citing a case that doesn’t exist.
The problem this solves
Modern general-purpose AI tools confidently produce legal text. They will name cases, quote holdings, and cite section numbers. The text sounds correct because the model is trained to produce plausible text. That isn’t the same as truth.
The first time I asked a major commercial AI a question I would have billed a client three hundred dollars to research, it cited a case that does not exist. Not a misquote — a fictional opinion with a confident citation, supposedly from the New York Court of Appeals.
Under New York Rule of Professional Conduct 7.1, I cannot ship a tool to clients that confidently invents law. So I built one that can’t.
The architecture
Every claim Lex.NY produces is tied to a source that exists in the corpus: 1.3M+ NY opinions and statute sections stored in AWS Aurora PostgreSQL, alongside the citation graph (4.9M citation edges in Aurora, traversed with recursive CTEs over CITES / APPLIES edges). Retrieval is corpus-only — pgvector semantic search and Postgres full-text search run against the same Aurora engine.
The retrieval pipeline runs before the language model ever sees the question. The model’s job is to summarize what was retrieved, in context, with a numbered citation after every factual claim. If the retrieval layer didn’t find a source, the model has nothing to cite — and the system prompt requires it to say so rather than invent one.
Full architecture and code: github.com/banksythequantLab/lex-ny (Apache-2.0).
What this is — and isn’t
- A research tool for finding NY case law and statutes
- A citation-anchored summary engine — every claim is sourced
- A way to navigate the citation graph (cited-by, citing)
- Open-source, attorney-supervised, free to use
- Legal advice on your specific situation
- An attorney-client relationship
- A substitute for a qualified NY attorney
- Comprehensive — the corpus has gaps and the model can be wrong
Using Lex.NY responsibly
Every citation Lex.NY shows is drawn from a real New York opinion or statute — the links go to CourtListener, the NY Senate’s OpenLegislation site, and other primary sources, so you can open the actual case. The whole point of citation-anchored AI is that the citations are real and traceable. What is not guaranteed is the legal interpretation: the analysis may be wrong, so do not rely on it to make legal decisions.
If you’re facing an actual legal problem, engage a qualified NY attorney rather than relying on this or any other AI tool. Lex.NY is what an attorney uses to start a research question, not what a non-attorney uses to finish a case.
About the author
Derek Soltis is a New York attorney (SDNY/EDNY), recognized as a Super Lawyers Rising Star. JD Rutgers, MBA/MS Fordham. He operates 28usc1782.com (international discovery practice) and Nota.Lawyer (LLC formation platform). Lex.NY is one front of a broader research and infrastructure project.
Rate limits + fair use
The Lex.NY public API is rate-limited at 10 requests per minute per IP address, with the streaming endpoint at 5/min. Each call runs pgvector retrieval against AWS Aurora and a fast hosted model under a strict-citation prompt, and capacity is finite. If you need higher throughput, fork the repo — the architecture is reproducible.
Every case. Every statute. Every cite verifiable.