REDACT<XXXXX>

Private redaction, fully local.

A standalone redaction suite for documents, PDFs, images, audio and video. The macOS DMG build is working now. Windows remains in pre-production. The platform intent extends to *nix, with iOS later, without changing the local-first security boundary.

$ redact run ./case-pack --profile uk-pii --local-only
Processing Local only Model boundary No external LLM Retention Zero source artefacts Unsafe output Fail closed
REDACT<XXXXX>
System status
Source files opened read-only
Processing local
Current REDACT<XXXXX> desktop interface

Current evidence

  • macOS DMG working
  • Windows pre-prod
  • Local Rust engine
  • No external LLM
  • Fail-closed export

Redaction fails when it is treated as a black box over text.

REDACT<XXXXX> treats redaction as a security function. The source file stays unchanged. Processing stays local. Output is created as a new artefact only after detection, review, irreversible rendering and verification.

The product boundary is the selling point.

The quiet promise is simple: sensitive material should not leave the machine just to be made safe enough to share, file, support, disclose or inspect with AI later.

01

Local processing

Files are read locally and processed without default upload, external LLM calls or network dependency.

02

Reviewable export

Findings can be inspected, kept, changed or approved before an irreversible redacted output is created.

03

Zero retained source

No retained logs, caches or app-managed original copies. Temporary processing state is purged after the run.

The intent is platform-agnostic, but the release posture stays honest.

REDACT<XXXXX> is being shaped as a local tool that can follow the user’s machine estate, not a cloud service with a desktop wrapper. The macOS DMG build is working now. Windows remains in pre-production. *nix comes after that. iOS is an eventual lane, not a current promise.

Working DMG

macOS

Current local build path with a working macOS DMG package.

Pre-prod

Windows

Current target for installer and desktop package preparation.

Intent

*nix

Planned desktop lane once local OCR, rendering and packaging gates are settled.

Later

iOS

Eventual mobile path after the desktop safety model proves itself.

The first version starts with documents, then expands into media.

The build path is deliberately staged. Document and PDF redaction prove the core safety model before image, audio, video, batch and policy automation are widened.

Phase 01

Document / PDF redaction MVP

Text, OCR, regex, PII, custom terms, review controls, flattened export and reports.

  • PDF and document intake
  • OCR for scanned pages
  • Manual region redaction
  • REDACTED-YYMMDD-HHMM-Filename
Expansion

Multimodal redaction suite

Faces, objects, transcripts, time ranges, policy profiles and watch-folder processing.

ImagesFaces, text, signatures, plates and selected regions.
AudioTranscript cue findings and local mute export.
VideoTimeline candidates and fail-closed export gate.
PoliciesUK, EU, US and Australia-first redaction packs.
ArchivesRecursive folder and archive processing.
ReportsExplicit output artefacts, not hidden processing logs.

This is already a local build, not a slide.

The current build includes a Rust redaction engine, CLI runner, Tauri desktop shell, React UI, local verification scripts, source escrow packaging and a working macOS DMG. Windows remains the active pre-production desktop lane; *nix and iOS remain future platform intent.

Current local REDACT<XXXXX> desktop demo, showing single-file and folder redaction outputs.

The next move is not more polish. It is assurance.

The sales story only works if the tool earns the security claim. The next work stays focused on evidence, release gates and safe operating limits.

Area Current focus Why it matters
MVP hardening Document workflows must become boringly reliable. PDF, document, image OCR, review decisions, report generation and temporary-state cleanup stay as the first release standard.
Assurance The tool needs proof, not claims. Verification checks, no-network evidence, source non-mutation tests, metadata removal and fail-closed paths remain part of the product surface.
Commercial fit The best buyers already understand the pain. Legal, FOI, healthcare, commercial and secure-support teams need redaction before disclosure, before vendor support and before AI-assisted analysis.