Source files
Stay in the selected folder and remain authoritative. FileNyani writes product state only below .filenyani.
Concrete privacy boundaries
Your documents stay where you put them.
FileNyani reads the folder you explicitly choose, runs retrieval and generation locally, and keeps portable workspace state beside the source files. There is no FileNyani cloud account or remote inference service.
Your folder
Authoritative sources
Original files remain ordinary files. FileNyani never silently rewrites, moves, or converts them.
.filenyani
Portable conversations, extraction artifacts, reviews, saved field lists, and rebuildable derived state.
What lives where
Private by architecture, not by slogan.
Different categories of state have deliberately different owners.
Questions and answers
Stay on the device in folder-owned conversation files. No cloud conversation store exists.
Models
Download only after an explicit preparation action and live in browser or app-owned caches, never the workspace.
Desktop recents
Absolute recent-workspace paths stay in Desktop application data, outside portable .filenyani state.
Native API
Binds loopback only, requires a random session token, and does not expose engine credentials to the browser UI.
OCR evidence
Derived transcripts are labelled, revision-bound, and never replace the original PDF pixels as source authority.
Network activity
Only a few explicit reasons to use the network.
Hosted Web loads application assets. Models download only when you prepare them. Vercel-hosted Web records sanitized anonymous page views; query strings are removed and workspace content or product actions are not included.
| Edition | Analytics | Expected network use |
|---|---|---|
| Hosted Web | Anonymous sanitized page views only. | Application assets and explicit model acquisition. |
| Desktop | None. | Explicit model acquisition only. |
| CLI | None. | Explicit model acquisition only. |
| Docker | None. | Explicit model acquisition only. |
“Local” is not a claim that every edition starts offline.
The Hosted Web application and uncached model assets require initial network access. Once Native application and model assets are present, ordinary workspace processing does not require a FileNyani cloud service.
Inspect the evidence, not just the answer.
Local execution reduces data exposure. It does not make model output automatically correct. FileNyani keeps citations and trust states visible so you can judge the result.