RIPPRE

07 — Document privacy

Pocket-ObSecura

In development

Share the file, not the secrets.

Pocket-ObSecura takes the spreadsheets, documents, PDFs, emails, and scanned pages your business runs on and produces safe, realistic copies. Names, account numbers, amounts, and identifiers are swapped for convincing stand-ins or blacked out, and spreadsheets, documents, and emails keep their shape — the same sheets, columns, paragraphs, and headers — so a vendor, a reviewer, or an AI tool can still work with them. Everything happens on your own Windows or Mac computer. Nothing is uploaded, ever.

You stay in control of every change. Pocket-ObSecura shows the original and the proposed copy side by side, lets you approve, deny, black out, or rewrite each finding, and asks before anything undecided goes out. Then it reads the new file back and flags any original value that is still there. Ready-made profiles cover the everyday jobs: preparing a file for an AI tool, legal redaction, accounting samples, investor diligence, and demo data.

  • Works offline, on your own computer
  • Checks every copy for leftover originals
  • Never changes your original files
Request early access Read the docs

Features

Opens the files you use

  • Works with Excel, CSV, Word, PDF, and email files, including Outlook messages, plus plain text and PNG or JPEG images.
  • Open one file or a whole folder, or drag files and folders anywhere onto the window.
  • Scanned PDF pages and pictures are read with built-in text recognition that runs on your computer, and text found on a picture is covered with black boxes.
  • Spreadsheets, documents, and emails keep their structure — the same sheets, columns, paragraphs, and headers — with only the sensitive values changed.
  • Tells you as soon as you open a file if it has parts that cannot be cleaned, such as a macro or an embedded object in a Word document.

Finds what is sensitive

  • Finds 18 kinds of sensitive information: names, company names, email addresses, phone numbers, dates, amounts, percentages, IP addresses, Social Security and employer ID numbers, card numbers, IBAN and SWIFT codes, passport, loan, purchase-order, and legal case numbers, and secret keys and access tokens.
  • On-device AI catches names and company names that pattern rules miss, such as lowercase or hyphenated names. It works offline, and the application works without it.
  • Six ready-made profiles — AI Upload, Legal Redaction, Accounting Sample Data, Investor Diligence, Demo Dataset, and Full Privacy Sweep — decide what to look for and what to black out. Or tick the kinds of data yourself.
  • Column headings and form labels are recognized as labels, not data, so a sanitized spreadsheet keeps its headings. They stay in the list, one click from being replaced.
  • Every finding shows how confident the detection is, so you can approve the sure ones in one step and check the rest.

Stand-ins that hold together

  • The same person or company gets the same stand-in in every file you have open, and a shortened mention — “Acme” for “Acme Holdings LLC” — resolves to the same one.
  • Stand-ins look real but are built to be safe: email addresses use reserved example domains, US phone numbers use the 555 exchange, card numbers use well-known test prefixes, and Social Security numbers come from a range the Social Security Administration never assigns.
  • Dates move by the same number of days throughout, so the gaps between them still make sense, and amounts shift by a few percent while keeping their currency and format.
  • Secret keys and access tokens are blacked out by default, and any finding can be blacked out instead of replaced.
  • Type your own stand-in for any finding when you need a particular value.

You approve every change

  • See the original and the proposed copy side by side, with each finding highlighted by its status.
  • Approve, deny, or black out each finding — or approve everything at once, or just the findings the app is most sure of.
  • A progress bar, per-type counts, and search show what is left to decide, even in a file with hundreds of findings.
  • Single-key shortcuts for approve, deny, redact, and ignore keep a long review moving.
  • By default, export asks before any undecided finding goes out, and an administrator can make that mandatory.

Checked before you share

  • Your original is never touched. Each export is a new file with “.sanitized” in its name.
  • After every export, Pocket-ObSecura reads the new file back twice — once as it reads any document, and once with an independent reader that looks inside every part of the file — and flags any original value still there.
  • Document properties such as author and company are cleared, and email routing headers are removed. Email attachments are removed rather than passed along unchecked.
  • Save an audit report of the session — files, fingerprints, counts by type and decision, settings, and verification results — with no sensitive values in it.

AI round trip and batch work

  • Paste any text — a draft email, a chat message, a log excerpt — and get it back masked with the same stand-ins as your open files, ready for an AI tool.
  • Paste the AI’s reply and the stand-ins turn back into the real names and numbers. It all happens in memory, nothing is written to disk, and redacted values stay redacted.
  • A command-line tool scans, exports, and verifies whole folders, and can watch a drop folder and keep writing safe copies of whatever lands in it.
  • Settings an administrator can enforce for the whole organization: allowed file types, a required profile, approval before export, metadata removal, and passing verification.

Why you can trust it

Pocket-ObSecura never connects to the internet, and that is enforced three ways: the build fails if networking code is added, a test inspects the finished program, and the app refuses to load networking components while it runs. There is no telemetry, no account, no online activation, and no update check. Text recognition and name detection run on your computer.

Your sensitive values are not kept. Documents and findings live in memory while you work and are gone when you quit; nothing from your documents reaches your disk except the copies you export and the reports you choose to save. When a format has parts Pocket-ObSecura cannot clean — a macro, an embedded object, text inside a picture on a PDF page — it says so before you share, instead of passing the file off as safe.

Version updates

Pocket-ObSecura is in development and has not been released. These are dated milestones from its recorded history, which begins in August 2026; every build so far carries version 0.1.0.

  1. Current build

    Pre-release

    Text, images, and the AI round trip

    • Opens plain-text files (.txt, .md, .log) and PNG or JPEG images. Text found on a picture is covered with black boxes, and embedded photo details are removed.
    • New AI round trip: mask pasted text with the same stand-ins as your open files, then turn an AI’s reply back into the real values, all in memory.
    • A new review screen: a drop zone to start, a progress bar and per-type counts, search, and single-key decisions.
    • Light and dark color schemes, remembered between sessions, and a progress strip during long scans and exports.
    • Drop files and folders anywhere on the window, or open them from the command line or the Windows right-click menu.
    • The command-line tool can watch a folder and keep exporting safe copies of new or changed files.

  2. Hardening build

    Safer exports and stronger verification

    • Organization policy settings are now enforced at import, export, and verification, and a damaged policy file applies every restriction instead of none.
    • Verification now checks every finding you did not deliberately keep, and reads the exported file a second time with an independent reader.
    • Word documents: tracked-change deletions, comments, footnotes, and document properties are now cleaned.
    • Email: attachments are removed rather than passed along unchecked, and every address header is rewritten.
    • Excel: formulas, sheet names, and cell comments are now cleaned.
    • Security hardening.

  3. Consistency build

    One name, one stand-in

    • A person or company now gets one stand-in throughout, including shortened mentions such as “Acme” for “Acme Holdings LLC”.
    • Column headings and form labels are no longer mistaken for names; they stay in the list for review.
    • Choose the text-recognition set for scanned pages, including a separately installed multilingual set for scripts such as Chinese, Cyrillic, and Arabic.
    • Fixed: scanning could fail every time in the desktop app, and the “Sanitized (Proposed)” preview now updates as you decide.

  4. Scanning build

    Scanned PDFs and on-device AI

    • Scanned PDF pages are now read with built-in text recognition that runs on your computer.
    • On-device AI finds names and company names that pattern rules miss, such as lowercase or hyphenated names.
    • Passport, SWIFT, loan, purchase-order, and legal case numbers are now detected.
    • A toolbar puts every step one click away, and export now asks before leaving findings undecided.
    • Fixed: email headers were not cleaned, and verification could pass while an original value remained in a Word, PDF, or email file.

  5. First recorded build

    Where the recorded history starts

    • A desktop app for Windows and macOS that opens Excel, CSV, Word, PDF, and email files, proposes stand-ins for the sensitive values it finds, and writes a separate copy.
    • Six ready-made profiles, side-by-side review, and a command-line tool for batch work.

Documentation

How to set up Pocket-ObSecura, use it day to day, and fix what goes wrong.

Open the Pocket-ObSecura docs

Build with Rippre

Ideas become businesses through execution.

hello (at) rippre (dot) com