Legal data for research and legal AI
Building a research tool, a litigation-analytics model or a legal agent? Here is what the LAW.co corpus is good for today, and where you will need other sources.
3 relevant datasets
U.S. Court Opinions (LAW.co corpus)
Federal and state court opinions in the LAW.co corpus: opinion text, case name, court, jurisdiction, decision date, docket number, reporter citations, judges, parties and attorneys. Legacy corpus is Caselaw Access Project-structured; newer opinions are ingested from CourtListener plain text behind LAW.co's copyright gate.
- Coverage
- U.S. federal, U.S. state (all 50)
- Records
- Not yet verified
- WEB
- REST
- MCP
- DOWNLOAD
LAW.co AI Case Summaries
LAW.co-generated plain-language summary, holding, key issues, topics and procedural posture for a prominence-selected subset of cases. Only enrichment that passed LAW.co's QA gate (published / approved / auto) is exposed. Machine-generated; not citable as authority.
- Coverage
- U.S. federal, U.S. state
- Records
- Not yet verified
- WEB
- REST
- MCP
- DOWNLOAD
LAW.co Case Citation Graph
Directed citation edges between cases (citing case → cited authority, with pin cite), seeded from CAP cites_to data and extended with eyecite extraction on ingest. Exposed per case as 'authorities cited' and 'cited by'.
- Coverage
- U.S. federal, U.S. state
- Records
- Not yet verified
- WEB
- REST
- MCP
- DOWNLOAD
Good fits
Retrieval-augmented answers grounded in opinion text with a citation and source URL per passage. Exploring how cases cite each other. Plain-language summaries for triage, clearly labeled as AI-generated.
Poor fits (today)
Anything that needs a citator or good-law status, statutes or regulations, complete coverage of a specific court, or bulk export for model training. LAW.co's terms prohibit bulk extraction and model training on the corpus without a written agreement, and that agreement is pending.
Nothing on the marketplace is legal advice. Opinion text is not official or certified; verify against the court's own record before relying on it.
A practical retrieval pattern
Search with a focused keyword query, fetch the full records you want to use, and pass the opinion text to your model with the citation, court, date and source URL attached. Ask the model to quote and cite only from those passages. Show the provenance alongside the answer so a reviewer can trace each claim.
Treat AI summaries as navigation aids: good for deciding which opinions to read, not for quoting. Keep them in a visually distinct block, the way the API keeps them in a separate enrichment object with isDerived set to true.
Citing what you retrieve
Cite the case by name, reporter citation, court and year, as you would from any source, and link the canonical URL when one is present. Credit LAW.co and the upstream source (CAP or CourtListener) where you display records. Never cite a summary or holding as the court's words.
Evaluating results
Test retrieval on questions where you already know the right cases. Check that answers quote opinion text rather than summaries, that every citation matches a record you retrieved, and that the system reports the sampled-search warning instead of implying the result set is complete.
Provenance on every record
Whatever the topic, each legal record carries the same provenance fields, so you can tell source text from derived work and cite it correctly.
- provider / sourceId
- Who served the record and its id at that provider.
- upstreamSource
- Where the provider got it (for example CAP or CourtListener), or null if not stated.
- canonicalUrl
- Public URL for the record, when one exists.
- licenseId
- The license that applies, such as public domain for court-authored text.
- isDerived
- false for original text and facts; true for AI or editorial layers.
- retrievedAt
- When Data.co fetched it from the provider.
Access today
Individual cases can be read now on LAW.co's own public case pages. REST API and MCP access through Data.co are live for court opinions, case summaries and the citation graph. Previews on this site are synthetic samples titled [DEMO] that are never billed; the API returns live records. Bulk downloads stay closed until the provider confirms redistribution rights.
Each dataset page shows its status channel by channel, with the reason for anything that is not open, so you can plan an integration without guessing.
Legal Research: common questions
Can I build a legal chatbot on this data?
Yes, around single-record retrieval over the live API or MCP server. Using the corpus to train or fine-tune models needs a written agreement with LAW.co, which is not in place.
Is search semantic?
The upstream search Data.co can use is keyword-based and sampled. LAW.co's own vector search is not exposed to Data.co, and its embeddings are never redistributed.
Do I need a lawyer to use it?
Not to use the data, but nothing here is legal advice. Anyone acting on a legal question should consult a qualified attorney and the official record.
More in the legal hub
Case Law
U.S.
Court Opinions
Federal and state court opinions from the LAW.co corpus: opinion text, court, decision date, docket, citations, judges and parties, with source provenance..
Jurisdictions
Which U.S.
Federal
U.S.
State
U.S.
Statutes
Full statute corpora are not currently offered on the Data.co Marketplace.
Regulations
Federal and state regulations are not currently offered on the Data.co Marketplace.
Start building with live data.
API and MCP access is live. Sign up at app.data.co, create a scoped key, or talk to us about your use case.