Sovereign document intelligence
Cortessa answers from your approved document set with every claim traced to a cited source span. It runs offline on a sovereign workstation on your desk. Blind-tested on real, public documents: 0 fabrications passed the gate.
Internally executed blind protocol on real, public corpora: the engine never saw the sealed key; a separate source-only oracle authored it and a different model family judged the answers. Independent external replication is a pilot objective. Figures are from interim development hardware.
Every output is provenance-gated. The gate applies a deterministic, versioned containment policy before a claim can be shown as Verified; the cited source remains open for inspection. Cortessa never certifies — the engineer signs off.
Shown only when a claim is provably a near-verbatim span of a cited source.
Unprovable claims are shown as Draft — never passed off as fact.
When the fact is not present, the system refuses rather than guess.
A live question against the 156-document public NASA demonstration corpus, answered by the on-box model with nothing leaving the machine. The provenance gate holds all three paraphrased claims at Draft — 0 verified — while exposing the cited source and every consulted passage for inspection.
Supersession is a claim about time — Rev E replaced Rev D. Effectivity is a claim about configuration — Rev D still applies to hulls 1–4, Rev E from hull 5, both current at once. Serving the wrong one to the wrong variant is a safety event, and a search engine cannot tell the difference.
Lead capability
Cross-references your whole corpus and surfaces where documents disagree — numeric conflicts, supersession, broken references, stale content — as a cited findings list for an engineer to adjudicate.
Recall 1.0 and precision 1.0 on the seeded regression set, across four seeded error classes. Separately, a blind run over a real NASA programme document family returned 15 / 15 genuine findings.
Cited answers
Natural-language questions answered from your corpus, every sentence cited. It refuses when the fact is not present.
~97% retrieval and 76.9% strict factual accuracy across the 480-question blind evaluation; all 16 model fabrication attempts in that run were held back. In a separate exam, 12 / 12 out-of-corpus questions were refused with 0 fabrications.
Sign-off-ready briefings
Turns the corpus into a cited, sign-off-ready briefing, where the engineer accepts, revises, or rejects each drafted claim. In development — additional testing underway.
Sovereignty, provenance, air-gap operation: a product that clears that bar clears it for everyone else. Cortessa runs disconnected, on a dedicated sovereign workstation on the analyst's desk — a single executable; no server, no cloud, no internet dependency. Nothing leaves the machine to answer a question.
The same guarantee travels to civil aerospace, nuclear, energy, rail, pharma, legal and government estates. One product, one guarantee, many document estates.
Concept — not currently shipping
The sovereign workstation, taken to its conclusion: Cortessa is being brought to a sealed, self-contained appliance — the whole system in a single unit, running offline inside your boundary. No cloud, no server, nothing leaving the machine.
No third-party CDNs, no external web fonts, no analytics or trackers that phone home, no external calls at runtime. Every asset is self-hosted. It loads inside an accreditation boundary, disconnected — the same constraint Cortessa is built to.
There is a scoped, paid pilot worth a conversation — on your hardware, behind your firewall, against success criteria you set.
