your memory carries fifteen processes. nexmin’s carries them all.

how it is built.

Each session you validate contributes to a structured record: recurring themes, open threads, active hypotheses and key events in sequence. It has two layers — session fragments searchable by meaning, and a compiled portrait of the process — and Pensa rereads the second before each new analysis, so the reading starts with what is already known.

Only sessions you validate write to it. Sensitive changes ask for explicit acknowledgement. If the AI proposes a reading you do not share, you correct it: memory keeps your version, and later readings start from there.

what it is for, in practice.

Returning after months. Someone comes back after six months. Instead of rereading an entire file, you open the memory: the current reason for seeking help, the threads left open, the hypotheses in the air. You enter the session with your head in the right place.

Reading the process. Looking at one session is manageable; rereading thirty every time is not. Memory compresses the trajectory so Pensa can read the process between sessions — what changes, what remains, what emerges — without reprocessing the complete history.

The process as the unit, not the label. Memory does not accumulate diagnoses: it accumulates process. It is the material for nexmin’s phenomenological model — observing how a consciousness evolves, rather than filing a category.

your rules apply here too.

Memory is encrypted in the European Union, never trains models, and is deleted when you delete the patient — the right to erasure has no exceptions for technical convenience. nexmin does not decide what is remembered: it proposes, and you sign off. Protection is layered: TLS 1.3 in transit and column-level encryption (Fernet, AES-128-CBC + HMAC) for fields that may hold clinical material.

frequently asked questions.