Your legal AI queries documents. A document cannot tell you whether a holding still applies, which circuits it binds, or whether the language came from the majority or a dissent. We extract the assertion from the document and make it queryable.
Federal case law 1951 to present. California 2000 to present.
Most legal data products give you documents.
We give you structured assertions.
Without it, "grounded in authoritative sources"
means "grounded in PDFs."
The JSON record for a case isn't a document summary. It's the assertion itself: typed, traced, and jurisdiction-stamped. Every record carries what was decided, who decided it, which courts are bound by it, and whether it's still standing.
A citation tells you a court spoke.
The assertion tells you what it decided.
The pipeline extracts assertions at the sentence level. Each one is typed, cross-referenced, speaker-attributed, and jurisdiction-stamped at extraction time, before it ever reaches storage.
The output is a federated network of named knowledge graphs, one per assertion category. Holdings are stored separately from dicta. Citations are stored separately from treatment records. Each graph is queryable independently and traversable as a unified network.
This is pipeline-extracted, not generated. The same input produces the same output every time. There are no stochastic components between document and assertion record.
Every extracted assertion moves through a two-state quality machine before export. Assertions that pass become part of the network. Assertions that fail are withheld. The pipeline does not ship uncertain output.
Extraction accuracy is measured against a human-labeled gold set, not self-reported on a single metric. The evaluation scores across seven independent classification categories, testing the full output chain from sentence-level extraction to final assertion record.
A 97% overall score means the system gets the right answer across intent, attribution, polarity, routing, bucket, modality, and conditional detection, at the same time, on the same sentence.
Human-labeled gold set. Multi-category scoring. Evaluated on held-out federal case law corpus.
Opinions added as published. Updated continuously.
Legal authority is not static. Omniarch tracks the full treatment arc of every holding, from establishment through every subsequent court that cited, narrowed, or overruled it.
KeyCite and Shepard's are the best citation verification tools available. They were built for human lawyers in a research session.
Omniarch structures what was actually decided and makes it queryable at the moment it matters. We're complementary to Westlaw, not a replacement for it.
The question isn't "does this citation exist?"
The question is "does this holding apply here, now?"
Usage-based licensing. Enterprise contracts scoped by coverage and volume. Contact us to scope →