Skip to content

OfflineMask 2.0.0

Choose a tag to compare

@fongfan999 fongfan999 released this 19 Oct 16:24
· 84 commits to main since this release

The first public release. The number starts at 2.0.0 by the owner's choice; there was no 1.0.0.

Masking and restoring

  • Paste text, review what the built-in rules found, hide or keep each value, copy the result and paste the AI's reply back to restore the original values (0004, 0061, 0070).
  • Built-in rules recognise emails, phone numbers and labelled banking identifiers without any model files (0005, 0012, 0017).
  • An optional recognition pack, downloaded from Settings and verified by size and SHA-256, adds names, addresses, organizations, birth dates, card and login details (0042, 0054, 0056). On macOS arm64 it runs through a native helper (0049).
  • Spreadsheets (XLSX, XLS, ODS, CSV): per-column choices, a masked copy with formatting kept, and restoration of the AI's returned file (0076, 0085, 0086).
  • Documents (DOCX, ODT): text, tables, headers, footers, notes and comments masked with formatting kept (0078, 0084). TXT and MD files open on the text page (0083).
  • Every file ends with a review of each hidden value and type before saving (0079).

Ẩn nhanh and quick actions (macOS, Windows)

  • A small button appears beside text selected in any application when a click would hide something; one click masks or restores in place (0096, 0100, 0101).
  • ⌘⇧M / Ctrl+Shift+M does the same from the keyboard, with the clipboard as fallback (0090, 0091).
  • Background mode with a menu bar / tray icon, an optional login item, and macOS Services for selected text and Finder files (0092, 0094, 0097).

History and privacy

  • Private sessions stay in memory unless a six-digit passcode is set; then they are kept in an encrypted vault on the device, with no recovery for a forgotten passcode (0062, 0065, 0071).
  • No telemetry, no automatic network access, no update check; the pack download is the only network action and the user starts it (0050, SECURITY.md).

Known limits

  • PDF and image input are not implemented.
  • Only macOS on Apple Silicon has been tested by hand; Windows x64 builds in CI and awaits manual qualification (0077); Linux is untested.
  • Builds are not yet signed or notarized (release notes).
  • Detection has documented misses and over-masking; review the result before sharing it (evidence).

Binaries are not attached yet; build from source as described in the README.