CORTEX
CORTEX is nexmin's AI engine: two officers, four offices. Scriba describes and compresses; Pensa formulates working hypotheses and reads the process between sessions. It is not a generalist model with a long prompt: it is a chain of specialised processes that split the risk — facts go to the synthesis, judgement goes to the hypotheses — and everything it produces arrives as a draft the therapist signs.
The architecture is born of a clinical boundary: describing and judging are different trades. A model that writes the note while interpreting the case contaminates the record with premature hypotheses; if the hypothesis later proves wrong, the data is no longer trustworthy. So Scriba sticks to what was said — every statement reviewable against the transcription — and hypotheses travel separately, signed as such, to be validated or discarded. The two officers are fed by specialised pieces: the clinical transcription with its non-verbal events, the process variables scored with to-the-minute evidence, and the voice analysis. Each piece does a bounded, reviewable job, instead of a single model opining on everything at once. Governance is the Trust Loop: the AI proposes, the therapist decides. Risk signals are flagged for a person to look at; nothing enters the clinical record without a signature. And sovereignty is standard: CORTEX processes in the European Union and sessions never train models.
Inside nexmin
Every session analysis in nexmin is run by CORTEX; its results reach the session tabs as drafts approved or edited one by one.
Related terms
Last updated: 2026-07-28