$ sha256sum chunks.jsonl ✓ matches manifest
Cryptographically verifiable, licensed-for-AI datasets from primary sources. Every record hashed, every release Merkle-rooted, every ID stable across releases — so your retrieval layer stands on data you can defend.
sample access opens soon · full provenance chain on every record · primary sources only
9f2c41ab…c135e41a✓ verifiable release"why_merkleset":
Every record carries its source URL, fetch timestamp and SHA-256. Every release ships a manifest with a Merkle root over every chunk hash — recompute it from the files you downloaded and check any record yourself.
source_url · fetched_at · sha256
An explicit RAG/embedding-use license instead of a gray area, with indemnification on Pro and Enterprise. Your legal review is one page, not forty.
embedding_use: permitted
record_id is a UUID v5 of the dataset plus the natural key; chunk_id is a content address. The same chunk in two releases carries the same ID and the same hash, so you can compare releases yourself and re-embed only what moved.
record_id · chunk_id · stable across releases
US federal procurement and regulatory ship today; courts, SEC filings and e-commerce are on the roadmap. All collected from primary sources — never resold vendor feeds.
procurement ✓ regulatory ✓ courts · sec · ecommerce
"how_it_works":
01 collect
Official bulk files and public APIs — SAM.gov and the Federal Register today. Every raw response is stored as its own evidence record and hashed at fetch time.
02 normalize
Cleaned, structured, chunked for retrieval. Chunk IDs are content addresses, so an unchanged chunk keeps its ID across releases.
03 release
JSONL plus a manifest per release, carrying every file's SHA-256 and the Merkle root. Recompute the root before a single byte enters your index.
"catalog":
RFPs from SAM.gov's daily bulk extract, with normalized NAICS/PSC metadata. Award history from FPDS and RFP attachments are next.
sam.govdaily
The Federal Register today, EUR-Lex on the roadmap. Versioned releases with stable IDs, so you can see what changed between them.
federalregister.govdaily
Planned: state trial court records normalized across fragmented portals, with personal data screened out before hashing.
state court portalsroadmap
Planned: EDGAR filings and earnings materials, chunked for retrieval with stable section anchors.
sec.gov / edgarroadmap
Planned: structured product facts for agentic commerce — attributes and availability, never copied prose.
public product feedsroadmap
Missing a corpus? We build custom niches and connectors on the Enterprise plan.
contact us
"pricing":
demo
$0
sample access opens soon
starter
$99/mo
per dataset / month
pro
$299/mo
per dataset / month
enterprise
Custom
custom agreement
"faq":
Yes — that is the point. Every dataset ships under an explicit license that permits RAG, embedding, and retrieval use. Pro and Enterprise licenses include indemnification. No scraping gray zones: we collect from primary sources, honor robots.txt and opt-outs, and screen personal data out of every record before it is hashed.
Every record carries its source URL, fetch timestamp, and SHA-256 of the raw response. Every release ships a manifest with a SHA-256 for every file and a Merkle root over the ordered chunk hashes. Recompute the file hashes, rebuild the root from chunks.jsonl, compare — a few lines of your own code, none of ours. We ran exactly that against a full-history release with a separate implementation, and the roots matched.
Cadence is per dataset and per plan: Starter ships weekly releases, Pro ships daily. Each release is versioned, and because record and chunk IDs are stable across releases you can compare any two of them yourself and see exactly what moved.
JSONL for records and retrieval-ready chunks, plus a manifest.json per release carrying every file's SHA-256 and byte count, the record and chunk counts, quality metrics, and the Merkle root over the ordered chunk hashes.
Not from a sample dataset today — sample access is closed while the corpus grows, and it opens soon. In the meantime, tell us which corpus you are evaluating: we will walk you through a real release — the manifest, the per-file SHA-256 and the Merkle root — and you can recompute the root with your own code before any money changes hands.
On the Enterprise plan we build custom niches and connectors against your source list, with freshness terms agreed in the contract. Tell us what your retrieval layer is missing via the contact form.