Privacy-First Architecture

Local-first by direction

Neuroartan should not require every thought, memory, voice sample, or document to begin inside company cloud storage. The long-term architecture supports a local-first vault where raw files can remain under the user’s control.

The vault may live on a local device, in a user-selected folder, or in an Apple-native storage layer such as iCloud Drive or CloudKit where supported by future native applications.

Storage layers

Neuroartan distinguishes between local storage, iCloud or CloudKit storage, Neuroartan encrypted cloud storage, and model-processing workspaces. These layers are not the same and must not be presented as the same.

A local or iCloud file can be indexed by reference, parsed locally where supported, or copied into Neuroartan Cloud only after explicit user approval.

Copy is not index

Neuroartan must clearly distinguish between indexing a source and copying a source. Indexing may create metadata or local references. Copying means Neuroartan receives a stored copy or approved derivative.

This distinction is essential for sensitive material such as private thoughts, memory records, voice files, grief notes, family history, model calibration data, and legacy settings.

Consent before deeper processing

Deeper processing requires clearer consent. Parsing, chunking, embedding generation, memory graph construction, voice processing, model calibration, and legacy continuity must be governed separately.

Neuroartan must avoid bundled consent that silently authorizes every future use of intimate data.

Recipe Workshop

Recipe Workshop is the planned intake layer for files, folders, vaults, transcripts, thoughts, and other user-selected sources. It should preview sources, filter file types, classify sensitivity, show what will be processed, and ask for consent before any upload or model use.

It must never silently ingest an entire mind archive.

Memory, voice, and legacy

Memory, voice, and legacy continuity require higher safeguards than ordinary application data. They may contain identity, emotion, family history, health-like references, voice patterns, loved-one access rules, or posthumous release controls.

These systems require separate consent, stronger access controls, export and deletion paths, and legal review before production activation.

Clinical and assisted onboarding boundary

Neuroartan may later support expert-assisted onboarding, elder-care workflows, or branch-based setup. These contexts require additional privacy, staff-access, consent, and legal review controls.

Neuroartan is not positioned as a therapy, diagnosis, treatment, or healthcare provider replacement unless a future legally reviewed product specifically establishes that boundary.

Current status

This page describes the architectural direction. Some capabilities may be in development, unavailable, or subject to legal and security review before launch.