Redactorr finds the sensitive values in your documents — in your browser, before anything leaves it. You approve every redaction, then hand AI a copy that is safe and still makes sense.
AUTO-APPROVE THRESHOLD (>95%) ACTIVE2 OF 2 CLEARED
02 May · Carlton Property Co. · Rent · −A$2,200
High-confidence detections (like validated statutory checksums) auto-approve by policy threshold. Ambiguous flags are surfaced for selective triage. You stay in full governance without line-by-line reading fatigue.
Approved values turn into typed tokens like [PERSON_1]. The document keeps its shape, so summaries, search, and answers still work.
STRUCTURE SURVIVES — AI AND PEOPLE CAN STILL READ THE DOCUMENT
04 · ONLY THE COPY MOVES04 / 04
FILE // harbour-mutual-may-demo.pdfCLEARED
Account holderSarah Mitchell[PERSON_1]
BSB063-241[BSB_1]
Account number1094 7823[ACCOUNT_1]
02 May · Carlton Property Co. · Rent · −A$2,200
RELEASED
The reviewed copy is what travels — to AI with your consent, to a secure share, to storage. The original and the Redaction Map stay put.
ORIGINAL + REDACTION MAP STAY GOVERNED · SENDING IS A SEPARATE CHOICE
Any format. Same discipline.
The engine is purpose-built for the complexity of these formats.
Documents
PDFDOCXRTFODTTXT
Spatial OCR and form field analysis
Track changes, headers, and footnotes preserved
Embedded objects and deep metadata parsed
Layout and structure survive redaction intact
Spreadsheets
XLSXCSV
Multi-sheet workbook traversal
Formula-resolved cell values processed
Named ranges and complex table structures supported
Strict schema adherence for downstream pipelines
Presentations
PPTX
Slide master and layout templates scanned
Speaker notes and hidden text extracted
Embedded diagrams and shapes traversed
OCR
Images & Scans
PNGJPEGTIFFWEBP
Local on-device OCR — no cloud dependency
Automatic skew correction and contrast boosting
Handwriting and low-resolution artefact handling
{
"key": [TOKEN]
}
Code & Data
JSONXMLHTMLMD
Nested keys, arrays, and deeply nested objects
Code blocks, HTML tags, and front matter handled
Output guarantees exact schema parity (it still parses)
Secrets & Keys
KEYSTOKENSCERTS
API keys, bearer tokens, and connection strings detected
Identified inside any of the supported formats above
Flagged as candidates — nothing auto-deleted
SECTION 03 — THE ANALYSISSTRUCTURE SURVIVES REDACTION
The analysis
Now let AI do the analysis.
The reviewed copy goes to the chat. The answers come back. The real values never left your browser.
You: Summarise this statement. Anything unusual about the rent? AI: [PERSON_1] paid A$2,200 rent on 02 May — the largest outgoing this month and up from the prior period. Subscriptions are steady; the closing balance is A$6,116. You: Do we hold earlier statements for [PERSON_1]? AI: Two — March and April, both on [ACCOUNT_1]. Rent was A$2,050 in each, so May is the first increase: up A$150, with no other recurring change.
Summarise this statement. Anything unusual about the rent?
SYS_ANALYSIS
[PERSON_1] paid A$2,200 rent on 02 May — the largest outgoing this month and up from the prior period. Subscriptions are steady; the closing balance is A$6,116.
USER_PROMPT
Do we hold earlier statements for [PERSON_1]?
SYS_ANALYSIS
Two — March and April, both on [ACCOUNT_1]. Rent was A$2,050 in each, so May is the first increase: up A$150, with no other recurring change.
●WHAT AI SEES · STRUCTURE INTACTTHE EXCERPT BEHIND THE ANSWERS
AI models don't need your raw data to reason. By replacing sensitive values with deterministic tokens (like [ACCOUNT_1]), the engine explicitly labels the meaningof the data for the LLM. The AI retains 100% of the document's narrative, structural, and relational context without ever being exposed to the underlying PII.
SECTION 04 — THE BOUNDARYWHAT STAYS ON YOUR MACHINE, WHAT CAN MOVE
The boundary
What stays on your machine. What can move.
Nothing leaves without you. Reviewing a document creates a safe copy — it sends nothing anywhere. Sending is always its own step, and it only ever carries the reviewed copy.
LOCAL HARDWARE
Your original documentRaw names, numbers and details — never uploaded, never sent.
The Redaction MapThe key that links each token back to its real value. Locked away from the copy.
NEVER SENT ANYWHERE
→
THE ONLY GATEYOUREVIEW + APPROVE
→
EXTERNAL NETWORK
The reviewed copyTokens instead of real values — safe to work with, still makes sense.
VAULT · saved for laterAI · with your consentSECURE SHARE · deliberate
Every document runs the full AEGIS pipeline — two preparation layers, twelve specialist engines, and a final gate that holds anything unresolved for your review. The full methodology is public.
03COMMANDERorchestrates 12 specialist engines in parallel
04–14SPECIALISTSidentity · financial · medical · legal · location · AU identifiers…
×12
15WARDENfinal gate — unresolved candidates stay flagged for youHOLD
→YOUapprove, edit, reject, retag. The only release authority on the pipeline.GATE
Global tools treat Australian identifiers as edge cases. We treat them as the default.
Every AU government ID, healthcare number and financial identifier ships built in — not as an add-on.
38AU government identifiers
TFN, ABN, ACN, Medicare, BSB, DVA, WWCC, Director ID, ImmiCard, Centrelink CRN, IHI, NDIS, PBS, USI and more — the identifiers Australian documents actually carry, built into one engine.
8Statutory verification
Real identifiers adhere to strict statutory formatting rules — Australian TFNs, Medicare cards, and ABNs carry specific structural patterns. The engine verifies these standards locally, so numbers that merely look like identifiers are dropped, not flagged.
MAPPEDAustralian laws + regulations
Privacy Act 1988, Healthcare Identifiers Act 2010, My Health Records Act 2012, NDIS Act 2013, Migration Act 1958 and more — entity types are linked to their governing legislation where applicable.
BSBBank directory checks
Every BSB is validated against an institution registry that ships inside the engine — Big 4, Macquarie, Bendigo, ING, Suncorp, BOQ, St.George, Bankwest and more. No server call needed.
LOCALPrivacy Act 1988 · data boundary
Detection and redaction run in your browser and original values stay local — the boundary is part of the coverage, not a separate promise. Copies move on only when you send them.
DOMAINIndustry configurations
Construction, healthcare, legal, insurance, immigration, aged care, government, real estate and more — each configuration is tuned to that industry’s documents and the identifiers that matter in them.
Trust isn’t a paid feature.
The free checker and the workspace run on the same infrastructure — the register opposite ships with every plan. Security is the product, not an upgrade.
It started with a simple question: why does every AI tool demand your raw data?
3,000+ detection patterns · every edge case hunted downRead our story →
SECTION 06 — THE PACKSYOUR INDUSTRY, DETECTED
The packs
27 Industry packs. One dedicated engine each.
From healthcare and finance to FOI and defence. Enable an industry pack for industry-specific detection, add custom patterns for your organisation’s own identifiers, or manually redact anything else. Your data. Your rules.
SECTION 07 — THE VAULTSTORED, HELD, GOVERNED APART
The vault + workflows
Control the copy, map, share, and audit trail.
Redaction is the first trust boundary. Redactorr then helps teams move the reviewed working copy into vault, sharing, AI, agents, and knowledge workflows while the original-value map stays separate and governed.
01 · INTAKE
Review the sensitive original
Detection and redaction run in your browser during the review step, so candidate values can be approved, edited, or rejected before export.
Local review · 0 uploads
02 · TRANSFORM
Create the safe working copy
The reviewed redacted copy becomes the useful document object for vault storage, sharing, AI, and knowledge workflows.
Safe object · Governed copy
03 · GOVERN
Keep the map separate
The original-value map is treated as a separate governed object, not bundled into the default copy or share package.
Isolated map · Sealed apart
04 · AUDIT
Log who sees what
Sharing, AI context, map access, expiry, and review outcomes become visible product states rather than loose feature claims.
AI is downstream from document review. Redactorr provides clean, structurally intact redacted context to LLMs while original raw values and the Redaction Map remain locally sealed and governed.
Reviewed safe copy enters the AI workspaceOriginal-value map is withheld unless policy allows recoveryOutputs keep source lineage and approval state
SEALED VAULTRaw originals and the Redaction Map stay strictly governed in local memory.
Custom Patterns
Mould the engine to your exact use case.
Industry packs handle the standard entities, but every organisation has unique identifiers. Define custom Regex patterns, exact-match dictionaries, or proprietary IDs. Redactorr compiles your custom rules alongside our engines into a unified, high-performance pipeline.
Proprietary project codes
Internal employee IDs
Custom database formats
// CUSTOM_PATTERN_COMPILER[ONLINE]
PATTERN_ID:CUST_ORG_099
PATTERN_NAME:Project Alpha Codes
ENGINE_TYPE:REGEX_COMPILED
EXPRESSION:
/PRJ-[A-Z]{3}-\d{4}/g
ACTION:REDACT_MATCHES
Monitoring datastream for custom patterns...
SECTION 08 — THE PRICESAME ENGINE, EVERY PLAN
THE PRICE
Clear pricing. Zero enterprise gates.
Every tier runs the exact same client-side isolated security architecture. Upgrade only when team size, document formats, or automation volume warrants it.
Free Forever
A$0/month
Full client-side redaction engine for individuals processing plain text.
20 document scans per day 20 core foundation patterns Plain text & clipboard snippets 100% in-browser WebAssembly Zero server transmission Manual override & whitelisting
Everything in Professional 150 AI sessions / seat / month Encrypted 30-day Team Vault Multi-seat permission controls Centralised billing & invoicing Priority SLA support