Overview

Install and activate

Download the installer and run it. The beta build is not code signed, so SmartScreen warns once. The offline installer needs no network connection at any point, and carries all seven detection languages and the Accuracy tier; the light one starts on English and can fetch the other six detection languages, but never the Accuracy tier. The interface language is a separate setting and a separate list: five languages, both installers, nothing to download. Activation is a signed license key you paste in; Omit verifies it locally with Ed25519 and never contacts a server.

Verify offline

Disconnect the network and run a redaction end to end: Omit reads the file, runs detection, applies the policy, and writes the output, all without a live connection. If any step required the network, it would fail here instead.

Redact a file

Drag a file, text, email, CSV, Excel, PDF, or a scanned image, onto Omit Redact. It scans on-device, groups every detection by type, and waits for your review. Approve, reject a false positive, or let it run automatically. A new _redacted copy is written next to the source; the original is never modified.

Bulk mode

Point Redact at a folder instead of a single file and it walks every supported document inside, applying the same policy and writing each output next to its source, with one combined audit report for the whole batch.

Redaction operators

Each of roughly 25 entity types maps to one operator, so you control exactly how every kind of PII is handled.

The reversible vault

Pseudonymized values are sealed in an encrypted vault (AES-256-GCM, keys derived with Argon2id and protected by Windows DPAPI). Paste an anonymized output back into Omit and, with the correct key, pull the original values out. A wrong key returns opaque data, never plaintext.

Detector coverage

Google Cloud DLP publishes the widest public catalogue of PII types, so we treat it as the yardstick. Every text infoType in that catalogue is mapped to an Omit detector, and the mapping is a file in the engine rather than a claim on a website: a test fails the build if a single type is left unmapped. Note the last row: 22 of these types are detected only with the broader coverage mode switched on, so a default install covers 191 of the 213.

A granularity difference is where the two catalogues slice the same information differently. Google separates FIRST_NAME, LAST_NAME, MALE_NAME and FEMALE_NAME; Omit detects all four as PERSON and redacts them identically. Google's US_STATE is our LOCATION. The information is caught either way, which is why these count as mapped rather than missing, but the label you see in the audit report is the coarser one. The eight catch-all types are a weaker case again: things like GENERIC_ID and GOVERNMENT_ID are open-ended by definition, so they are caught by a broad identifier detector rather than a purpose-built one, and that detector is part of the opt-in coverage mode.

Catalogue coverage and detection quality are different claims, and only the second one matters in production. Coverage says a detector exists; it says nothing about how often it fires correctly on your documents. For that, we publish measured results against Google Cloud DLP on the home page, including the entity types where we do worse. Read both.

Languages

Three separate lists, and they are not the same list. The app you read, the part that recognises names, and the part that recognises identifiers each cover a different set, so the useful answer to whether Omit works in your language depends on which of the three you mean.

The interface is switchable in Settings and opens in English. Both installers carry all five and neither downloads anything to get them, though it is not finished: about 45 of the 394 interface strings, roughly one in nine, are still English in the four other locales, mostly in the vault screens, the re-identify flow and the hints explaining detection modes. That is every multi-word string byte-identical to the English one in all five files; a further handful of single words are untranslated and not counted. Detection is the wider list on purpose, because the documents people redact are wider than the interfaces they are comfortable reading. Only the offline installer carries all seven detection languages in the box; the light one starts on English and fetches the others on request.

Outside those seven you keep every structured detector, because an IBAN or a passport number is matched by pattern and checksum and does not care what language surrounds it. What you lose is the model layer that finds names, organisations and places from context. For a sheet of account numbers that costs nothing. For a file full of personal names it costs a great deal, and you should run your own documents through the beta rather than take our word for it.

Audit reports

Every run writes an HTML audit report listing what was detected and how each item was handled, redact, anonymize, or de-anonymize, each timestamped. Hand it to your compliance team or attach it to a procurement review.

Admin policy

Fleet admins set a ceiling policy that intersects with each user's local settings, so a rollout can restrict which operators and entity types are available without touching every machine by hand.

Offline licensing

License keys are signed offline and verified locally with Ed25519. There is no license server to reach, no seat check-in, and no telemetry tied to activation.