Your analysts spend hours classifying documents. Cadmus traces every sentence back to the governing guide in real time.
Air-gapped. On-premises. Human-in-the-loop. No data leaves the network.
Request a Briefing
The classification problem your analysts already know they have
Cleared analysts make thousands of classification decisions a week. Every sentence, every paragraph, every portion marking requires a judgment call against the governing Security Classification Guide. Get it wrong in one direction and you create a spillage incident. Get it wrong in the other and you overclassify, blocking information sharing, slowing mission-critical decisions, and adding downstream costs in storage, SCIF infrastructure, and workforce time.
The estimated overclassification rate across the federal government is 50 to 90 percent. A single document can take one to three hours to classify. The current process does not scale, and the tools most analysts rely on are a PDF of the SCG and their own memory.
What Cadmus does
Sentence-level classification
Cadmus classifies text in real time as the analyst writes. Every sentence receives a classification level: not a document-level guess, but a granular, portion-level determination.
Traceable SCG citations
Every classification suggestion links to the specific SCG provision, EO 13526 reason, dissemination control, and declassification instruction.
Ask Cadmus
Analysts can ask natural-language questions and receive cite-backed, actionable guidance drawn from the indexed classification guides.
Audit logging
Every suggestion, every analyst decision, and every override is recorded. The audit trail creates a defensible record for security review.
Cross-domain pre-validation
Cadmus pre-validates redactions and caveats before air-gap transfers: JWICS to SIPR, classified to unclassified.
SCG ingestion
Administrators configure required classification guides at the division level. Cadmus indexes, vectorizes, and applies the correct sources automatically.
How Cadmus deploys
Cadmus runs air-gapped, on-premises only. It deploys inside JWICS, SIPR, and SCIF-local environments. There is no cloud component, no internet dependency, and no data leaving the network.
All inference runs locally. The LLM operates entirely within your accreditation boundary, under your security controls, on your hardware.
Kubernetes and OpenShift · IL4 and IL5
The analyst decides. Cadmus suggests.
Cadmus does not classify documents. Cadmus supports the analyst who does.
Every suggestion includes the governing provision so the analyst can verify the reasoning before accepting it. If the analyst disagrees, they override with a single click. The override is logged with the analyst's rationale, creating a complete audit trail of human decisions, not machine decisions.
No classification is applied without human confirmation.
Questions we get asked
Pricing
Cadmus is licensed as an annual enterprise subscription, tiered by classification network level, concurrent users, and SCG library size.
Contact us for pricing and deployment details.
Request a BriefingSee Cadmus in action. Schedule a briefing and we will walk your team through the platform.
Request a BriefingAir-gapped. On-premises. The analyst always decides.