Query the whole US primary-law corpus as a database. 229 Parquet files in an object store, readable in place over HTTPS with DuckDB or loaded into Postgres, plus direct access to the vector index
The API answers one question at a time. When you need the corpus itself, take it.Every dated snapshot ships as Parquet in a public object store. You can query it
where it sits, load it into your own warehouse, or pull the whole thing down and
never call us again.
-- Runs as-is. No key, no download, no signup.INSTALL httpfs; LOAD httpfs;SELECT citation, section_title, word_countFROM read_parquet('https://oss-data-us.vaquill.ai/v2026.08/us_ak_constitutions.parquet')WHERE section_title ILIKE '%search%';
That query reads a remote Parquet file over HTTPS and returns in well under a
second. The object store serves range requests, so DuckDB fetches only the
column chunks it needs rather than the file.
-- Whole corpus into a local database file. Globs are not supported over plain-- HTTPS, so read the file list out of the manifest and pass it as an array.INSTALL httpfs; LOAD httpfs; INSTALL json; LOAD json;CREATE TABLE law ASSELECT * FROM read_parquet( (SELECT list(f.url) FROM (SELECT unnest(files) AS f FROM read_json_auto('https://oss-data-us.vaquill.ai/index.json')) WHERE f.file LIKE '%.parquet'));
-- After downloading the tarball and unpackingCREATE TABLE law ( act_id text PRIMARY KEY, citation text, citation_short text, state text, jurisdiction text, document_type text, title_number text, title_name text, chapter text, chapter_name text, section_number text, section_title text, breadcrumb text, display_path text, act_status text, text text, word_count bigint, source_url text, last_amended_year bigint, subsection_count bigint, cross_references_usc text, cross_references_cfr text, public_laws_referenced text, year bigint);-- then COPY from a Parquet-aware loader, or convert with DuckDB first
The corpus is also held as an embedded, chunked retrieval index. Direct access
to that index is available under a data licence, which matters if you would
rather not spend the compute re-embedding several million sections to get back
to where we already are.Talk to us at contact@vaquill.ai about scope.
This is not a hosted replica we operate for you. There is no connection string
to rent. You take a copy and run it on your own infrastructure, which is the
point: the copy is yours, and it keeps working if we do not.
The legal text is public domain under the government edicts doctrine
(Georgia v. Public.Resource.Org, 590 U.S. 255 (2020)). Our compilation, the
schema, citations, hierarchy and cross-references, is CC BY 4.0. The ingestion
pipelines are Apache-2.0.Attribute as “Open US Law by Vaquill AI, CC BY 4.0”. Attribution can be waived
under a commercial licence.
The free snapshot and a paid bulk licence are not the same thing. Anything
you take from the public object store is CC BY 4.0 and irrevocably yours,
on every version we have published.A paid bulk delivery, where we ship you the corpus under a separate signed
agreement rather than answering calls, is different: on termination or expiry
you delete stored copies of our compilation within 30 days. There the corpus
is the deliverable, so open-ended retention would let one period buy it
outright. See Rights and licensing.
