vaquill/open-us-law

Open US Law

New in v2026.09.1

Updated September 30, 2026.

• 5,081,709 sections, down from 5,276,632 in v2026.09 (-194,923).
• 308 files, down from 317.
• Three new columns: official_status, currency_note and adopting_citations. The first 24 columns are unchanged. See Columns added in v2026.09.1.
• Regulations for 40 jurisdictions, alongside the current Code of Federal Regulations, and 104,540 state attorney general opinions from 42 jurisdictions.
• Every section appears exactly once and carries the URL of the government publisher it came from.

This release supersedes v2026.09.

Getting a licence. This dataset is free for research, teaching, legal aid, journalism and other non-commercial use, with attribution. To get a licence to use this data commercially, email contact@vaquill.ai.

Why this exists

The law is public. Reading it should not cost money.

In practice, it does. A state's regulations sit behind a login. Court rules are scanned PDFs nobody can search. The annotated code that actually tells you what a statute means costs more per year than a legal aid clinic spends on rent. The people who most need to read the law are the least able to pay for the privilege, and everyone in this industry knows it and quietly accepts it.

We are not accepting it.

So here is every US statute, regulation, constitution, court rule, attorney general opinion, agency decision and guidance document we could get our hands on. Published by US federal, state and territorial governments, and collected by our crawlers. Cleaned, parsed, deduplicated, structured, and handed over. Five million sections. No key, no quota, no sign-up, no sales call. Download it and use it for research, teaching, legal aid, journalism, open source, anything non-commercial. If you want to build a commercial product on it, talk to us.

We built this because somebody had to, and because the people who could have done it years ago decided the paywall was more interesting.

If a tenant facing eviction, a solo attorney with no research budget, a clinic with three staff, or one stubborn developer in a garage ends up with the same raw material as a firm paying six figures a year for it, then this was worth every hour.

And selfishly: this is what we want to be remembered for. Not a product. This. That some people were crazy enough to take an entire country's law, put it in a file, and give it away.

If it helps one person get a fair hearing they would not otherwise have got, it has already paid for itself.

Use it. Break it. Build on it. Tell us what is wrong with it.

Built by Priyansh Khodiyar and the team at Vaquill AI. If you find an error, a gap, or a provision we got wrong, tell us and we will fix it in the next release.

---

5,081,709 sections of US primary law in one schema. Snapshot v2026.09.1, 308 files, 6.91 GB of Parquet.

Questions, ideas, or want to help? DM me on LinkedIn.

Watch the walkthrough

US law is public domain, but the structured version has never been openly available. Every state publishes its code differently, none offer bulk download, and the usable versions live inside commercial products. This is that layer, free and openly available.

The scrapers that produced it are public under Apache-2.0 at github.com/Vaquill-AI/open-us-law, so this corpus is independently reproducible rather than something you have to take on trust.

What is in it

Corpora that span many jurisdictions:

Corpus	Config	Sections	Jurisdictions
Statutes (state, territorial, and the US Code)	statutes	1,828,123	48
Regulations (state, and the current federal CFR with final Federal Register rules)	regulations	1,383,931	40
State attorney general opinions	ag_opinion	104,540	42
Agency guidance (including state insurance bulletins)	guidance	54,050	50
Court rules	court_rules	49,244	47
Constitutions	constitutions	11,864	51

Federal corpora, where a jurisdiction count would be meaningless:

Corpus	Config	Sections
Past CFR editions (historical, superseded text)	cfr_annual	886,066
Federal session laws (Statutes at Large)	session_law	375,967
Federal statute compilations	statute_compilation	80,288
Proposed federal rules	proposed_regulation	76,865
NLRB Board decisions	board_decision	66,594
Federal private laws	private_law	49,460
IRS written determinations	irs_written_determination	45,368
FTC administrative decisions	ftc_administrative_decision	17,973
Federal Register notices	notice	17,633
Agency rulings	ruling	7,278
SEC adjudications	sec_adjudication	5,790
MSPB decisions	mspb_decision	5,461
Immigration precedent decisions	bia_precedent_decision	3,292
Agency FAQs	faq	1,990
Presidential proclamations	proclamation	1,833
Office of Legal Counsel opinions	adjudication	1,371
Sentencing guidelines	guideline	1,281
Treaties	treaty	1,243
IRS notices	irs_notice	769
Executive orders	executive_order	737
Presidential documents	presidential_document	639
Senate treaty resolutions	resolution	406
Presidential memoranda	memorandum	401
CFPB enforcement actions	cfpb_enforcement_action	386
IRS revenue procedures	irs_rev_proc	378
IRS revenue rulings	irs_rev_rul	248
HHS civil rights resolution agreements	enforcement_action	146
IRS announcements	irs_announcement	83
Administrative guidance	administrative_guidance	11

Licensing, stated precisely

• The legal text is in the public domain. Under the government edicts doctrine (Georgia v. Public.Resource.Org, 590 U.S. 255 (2020)), edicts of government carry no copyright. We assert no rights over the text of the law.
• The compilation, normalization, and structuring in this release is offered under CC BY-NC 4.0: free for non-commercial use, with attribution. Commercial use of this compilation needs a licence from us; see below.
• The scrapers are Apache-2.0.
• Snapshots published before v2026.09.1 were released under CC BY 4.0, and that licence continues to apply to copies of those snapshots.

You cannot license what you do not own, so we are explicit about which of the three applies to what.

No third-party-licensed content is included in this release.

Using the data vs the API

This dataset is free for non-commercial use. Research, teaching, legal aid, journalism, public-interest work and open-source projects can download it and use it without asking us, as long as you attribute it. You never need to email us for that.

• Building something commercial? The compilation is not licensed for commercial use. Email contact@vaquill.ai for a commercial licence, or use the hosted API below, which is licensed for commercial use.

• Want the law kept current for you? We run a hosted API: the same law, refreshed continuously and searchable section by section, so you do not have to run the scrapers yourself. Sign up at app.vaquill.ai and generate an API key. Every account gets 500 free credits, no email needed.

• Non-commercial use of the API is free. If you are a legal aid organization, a researcher, a student, or building a non-commercial or open-source project and need more than the free credits, email contact@vaquill.ai and we will send you a free API key. Use this exact subject line so we can route it fast:

  Open US Law API - non-commercial - <your use case>

  Tell us briefly who you are and what you are building. The free key is best-effort and rate-limited, with no uptime guarantee. The dataset itself has none of those limits.

• Commercial use of the API is paid. Details at vaquill.ai/legal-api.

• Want it built for you? We also build products, workflows, and pipelines on top of this data. Email contact@vaquill.ai.

What makes this different from a raw scrape

Sections carry status, so you can exclude repealed law from an index rather than citing dead statutes as though they were operative:

• in_force: 3,326,749
• superseded: 950,551
• enacted: 426,263
• repealed: 149,796
• proposed: 76,865
• reserved: 45,789
• non_precedential: 45,368
• published: 17,633
• revoked: 12,651
• renumbered: 11,211
• transferred: 10,887
• plus omitted, rescinded, expired, deleted, recompiled, recodified, terminated, not_funded, withdrawn, rejected, not_yet_effective, relocated, pending, not_applicable, vacant, (none), suspended, vacated, unconstitutional, removed, vetoed, inactive

Sections also carry citation, full title/chapter hierarchy, breadcrumb path, subsection counts, and cross-references into the US Code and CFR where present.

One row is one section, and each section appears once. Long provisions are stored internally as multiple chunks and reassembled here, so a row is the complete text. Where a section has been amended, the row carries the current text; the prior wording is not merged in.

Columns added in v2026.09.1

Every file now has 27 columns. The first 24 are the columns every earlier release published, with the same names, order and types, so code written against v2026.09 keeps working. Three are appended:

• official_status (string). The publisher's own label for the copy, where it gives one. unofficial means the publisher itself says this copy is not the official text: typically an agency posting its own rules while the official compilation is published elsewhere. The text still comes from a government body, but cite the official compilation where it matters. Null means the publisher did not label the copy; it is not a promise that the copy is official. In this release: unofficial on 19,594 regulations sections for Oklahoma.
• currency_note (string). The publisher's own statement of how current the text is, such as a "current through" register issue, as the hosted API's currencyNote. Null where the publisher gives none. Present on 1,138,329 sections.
• adopting_citations (list of strings). The instruments that adopted or amended the rule: state register citations, register filings and adopting orders, and session-law citations, as the hosted API's adoptingCitations. Null where the source records none. Present on 324,978 sections.

Statute coverage by jurisdiction

Federal (the United States Code) is listed separately from the state and territorial jurisdictions, since it is a different corpus rather than another state.

Jurisdiction	Statute sections	Size
United States Code (federal)	60,667	80.8 MB
CA	161,428	60.7 MB
TX	122,630	31.2 MB
IN	83,148	23.2 MB
OR	72,664	20.4 MB
IL	72,251	31.4 MB
NJ	55,048	24.9 MB
WA	52,002	22.8 MB
NV	49,745	16.7 MB
AL	46,614	19.4 MB
LA	43,213	22.4 MB
MI	42,355	21.0 MB
NC	41,290	16.1 MB
NY	40,162	31.0 MB
SD	39,589	11.0 MB
MD	39,552	16.3 MB
OK	35,330	18.3 MB
VA	33,861	17.3 MB
OH	33,158	20.1 MB
NM	31,115	11.4 MB
MT	30,515	13.0 MB
SC	30,489	13.4 MB
MO	29,363	16.2 MB
ND	29,042	11.2 MB
NE	28,502	13.1 MB
IA	28,223	12.9 MB
MN	27,747	20.6 MB
UT	27,556	15.3 MB
WV	25,664	14.4 MB
NH	25,375	10.6 MB
ME	25,316	15.2 MB
FL	25,020	21.3 MB
AZ	24,956	14.2 MB
PR	24,763	10.8 MB
KS	24,361	12.4 MB
DC	24,036	10.4 MB
VT	23,521	10.7 MB
MA	23,144	13.8 MB
ID	22,754	11.8 MB
DE	22,669	11.1 MB
KY	21,194	13.1 MB
RI	21,107	8.6 MB
WY	20,995	7.8 MB
HI	19,197	9.2 MB
WI	18,158	12.5 MB
AK	17,935	7.2 MB
CT	16,082	11.8 MB
PA	14,617	7.9 MB

Sourcing

The publisher of every section in this release is a government body, or a body a government designates to publish the law: state supreme courts and bar associations that publish the rules they administer, state compilation commissions and attorneys general. Every row carries the URL it was published from. The build refuses to publish a file containing a non-government source URL, and sections whose source cannot be shown are left out rather than published without one, so this is enforced rather than asserted.

That is a statement about who publishes the law, not about how easy it was to reach. Our crawlers collected the text with automated requests, in some cases through US-based proxy servers because a site serves only US visitors. A copy the publisher itself labels unofficial ships with official_status set to unofficial.

Honest limitations

• This is a dated snapshot, not a live feed. New snapshots are published quarterly. Always cite the version.
• Coverage quality is not uniform. Some jurisdictions publish clean structured sources; others required considerable parsing, and section counts vary for genuine reasons, since states differ enormously in how they subdivide.
• Some state texts still carry site navigation. 44,454 sections, mostly state regulations, include navigation or header text from the publisher's web page alongside the law. We publish the text as stored rather than rewrite law text with an automated cleaner, and will remove it at the source.
• Past CFR editions are historical. Every row in the cfr_annual config is superseded text from an earlier annual edition, labelled as such. Use the regulations config for the regulation in force.
• This is not legal advice, and it is not a substitute for verifying current law. Statutes are amended continuously. Check the official source before relying on any section.

Changelog

v2026.09.1 (September 30, 2026)

• Three new columns, official_status, currency_note and adopting_citations, appended after the existing 24.

v2026.09 (September 29, 2026)

• 5,276,632 sections in 317 files, up from 2,978,617 in v2026.08.
• Regulations for 49 jurisdictions, 104,540 state attorney general opinions from 42 jurisdictions, and past editions of the Code of Federal Regulations in their own cfr_annual config.
• More federal law: proposed rules and Federal Register notices, session laws and statute compilations, and decisions from the NLRB, FTC, SEC, MSPB, the Board of Immigration Appeals and the Office of Legal Counsel.
• The compilation moved to CC BY-NC 4.0. Earlier snapshots keep CC BY 4.0.

Reproducing this dataset

Every scraper that produced this corpus is open source: github.com/Vaquill-AI/open-us-law (Apache-2.0).

You can rebuild any jurisdiction from its official source and check the result against what is published here. That is deliberate: an open corpus you cannot audit is just a file you are asked to trust.

Verifying your download

SHA256SUMS.json lists every file with its SHA-256, row count and byte size.

import hashlib, json
sums = json.load(open("SHA256SUMS.json"))
for entry in sums:
    h = hashlib.sha256(open(entry["file"], "rb").read()).hexdigest()
    assert h == entry["sha256"], entry["file"]

totals.json gives the section count per corpus.

Bulk download (mirror)

Every file is also mirrored on a zero-egress Cloudflare R2 bucket, on a domain we control, so a copy exists independent of the Hub:

• Directory listing: https://oss-data-us.vaquill.ai/index.html
• Everything in one file: https://oss-data-us.vaquill.ai/v2026.09.1/open-us-law-v2026.09.1-parquet.tar (6.91 GB)
• Machine manifest (per-file SHA256): https://oss-data-us.vaquill.ai/index.json

The mirror is byte-identical to the Hub, carries the same SHA-256 for every file, and supports HTTP range requests, so you can query the Parquet directly without downloading it. index.json and latest.json always describe the current snapshot, so those two URLs never change between releases.

Loading

from datasets import load_dataset

# a whole corpus
ds = load_dataset("vaquill/open-us-law", "statutes", split="train")
rules = load_dataset("vaquill/open-us-law", "court_rules", split="train")
opinions = load_dataset("vaquill/open-us-law", "ag_opinion", split="train")

# one jurisdiction
ca = load_dataset("vaquill/open-us-law", data_files="us_ca_statutes.parquet")

Files are named us_{jurisdiction}_{corpus}.parquet. Each config lists its files exactly, so a config never picks up a neighbouring corpus.

More open source from Vaquill AI

Everything below is public and Apache-2.0 or CC BY unless noted.

Data and benchmarks

• open-legal-answer-benchmark - reproducible benchmark of US legal-answer quality. Verified questions, a standard-library scorer, results anyone can rerun.
• legal-mt-benchmark - English to Hindi legal machine translation on the WMT25 Legal Domain Test Suite, 7 metrics, all outputs published.

MCP servers

• courtlistener-mcp - CourtListener (US federal and state courts, PACER, eCFR). Hosted, bring your own key.
• canlii-mcp - CanLII, the Canadian legal database.
• vaquill-mcp - the Vaquill research API over USC, CFR, state law and case law.
• integrations - Slack, Discord, Teams, WhatsApp, Telegram and WordPress connectors.

Tools for the way lawyers actually work

• ms-word-addin - contract review, playbooks, drafting and research inside Word.
• google-docs-addon - the same for Google Docs, with native tracked-change redlines.

Reference

• awesome-legaltech - a curated list of open source legal tech, models, datasets and companies.
• playbooks - attorney-grade negotiation playbooks for US commercial contracts, clause by clause.

Citation

@misc{openuslaw2026091,
  title  = {Open US Law: Structured Primary Law of the United States},
  author = {Vaquill AI},
  year   = {2026},
  note   = {Snapshot v2026.09.1},
  url    = {https://huggingface.co/datasets/vaquill/open-us-law}
}