Caselaw Access Project

The Caselaw Access Project (CAP) is the product of Harvard Law School Library scanning its entire physical collection of American case law and turning it into consistent, machine-readable data. It covers all official, book-published state and federal United States case law: every volume designated as an official report of a court's decisions, which comes to 6.9 million decisions reaching from the 1600s through to 2020.

Access is genuinely open now. Earlier agreements that metered how much you could download have expired, so there is no cap, and a free bulk API plus downloadable bulk files let you take as much as you need. Each case arrives as structured data with the full opinion text alongside metadata such as court, jurisdiction, decision date, and citations. A complete mirror is published on Hugging Face under the free-law organisation, which is often the quickest way to load the whole corpus into a pipeline.

For RAG, this is the foundation for any system that answers questions over United States case law: legal research assistants, citation lookup, precedent search, and tools that summarise or compare decisions. The consistent format and rich metadata make it straightforward to filter by court or era before you chunk (split into passages) and embed the text.

A few honest caveats. The data stops at 2020, so it knows nothing more recent, and it comes from official reporters, which means unreported and very recent rulings are simply absent. Older opinions were captured by optical character recognition from scans, so expect occasional transcription errors in the oldest volumes. Case law is also unusually citation-heavy and context-dependent, so a plain retrieval setup benefits from parsing and linking citations rather than treating each opinion as flat text. On licensing you are on firm ground: the opinions are public domain and Harvard's digitisation is released under CC0, a full waiver of rights, so you can build on it commercially with no attribution required. Pair it with CourtListener for more current coverage.