QingJiao Redact

Embrace frontier AI, protect your core data

Running large models in-house is expensive, yet pasting contracts, client records and financial data into external AI risks leaking them, and conventionally redacted text cannot be turned back. QingJiao Redact redacts on your own computer: names, ID numbers, bank details and company identifiers become traceable placeholders such as {{NAME_0001}}, the redacted file works with any AI model or agent, and the answer is restored from the document’s own mapping table.

Download the desktop app7-day free trial · no account needed · files stay on your computer

Mac · Apple siliconM-series chipsComing soonMac · IntelIntel processorComing soonWindowsWindows 10 / 11, 64-bitComing soon

Deploy a model in-house

GPUs and an operations team to run your own model, which often falls short of frontier models on complex work.

Use cloud AI as is

Nothing to deploy, but confidential data goes straight to the provider’s API, where it may be retained or used for training, and you cannot control where it goes.

QingJiao Redact + cloud AI

Redact on your machine with a small footprint, send only placeholders to the AI, and restore the answer exactly from the mapping table.

  • File panelRedacted and restored files kept apart, batch upload
  • Original previewSensitive data highlighted by type, one click to the redacted preview
  • FindingsMarked Model or Rule; remove, retype or allowlist each item
QingJiao RedactAlex MorganRedactChatSettingsWebsiteSign out
employment-judgment-demo.txtDone 1
✓Upload✓Detect3Review4Export
Send to chatRedactRedact filesDownload redactedRestoreUpload to restoreDownload restoredRestore this file
OriginalRedacted1492 characters · 41 redactions · 10 types
Edit originalRedaction style Placeholders (reversible) ⌄Data types ·16 ⌄Redact again

EMPLOYMENT TRIBUNALS (LONDON CENTRAL)

JUDGMENT

Claimant: Emma Hayes, born 15 March 1988, of 14 Kestrel Road, London N1 7GU. National Insurance number JG 10 37 48 D. Telephone 020 7946 0381. Email [email protected].

Representative: Daniel Brooks, solicitor.

Respondent: Bluesail Analytics Ltd, company number 09481276, of 40 Bank Street, London E14 5NR. Telephone 020 7946 0742.

Director: Oliver Grant, appointed 3 May 2019.

Second respondent: Marcus Reed, born 7 August 1979, of 18 Elm Road, Bristol BS8 1QU.

Emma Hayes brought a claim for unpaid wages against Bluesail Analytics Ltd and Marcus Reed, heard on 2 April 2025. Emma Hayes attended with Daniel Brooks as her representative; neither respondent attended.

Emma Hayes joined Bluesail Analytics Ltd on 1 June 2021 as a product manager on a monthly salary of £7,200, paid into her account at Northgate Bank, IBAN GB82 WEST 1234 5698 7654 32. Bluesail Analytics Ltd stopped paying her salary on 1 October 2024, and her wages from 1 October 2024 to 31 January 2025 remain unpaid.

Real screenshot ↗

Redact first, then hand the document to AI

A complete redact, AI, restore loop. With placeholders, Word, Excel, PowerPoint, text and PDF can all be restored; files redacted with black boxes or asterisks cannot. Try the interaction on the right.

  1. 1

    Import files

    Contracts, judgments, spreadsheets, slide decks, scans and images, one at a time or in batches.

  2. 2

    Detect locally, review by hand

    Rules with checksums catch structured numbers and an entity recognition model finds names, organisations and addresses. Every finding shows where it came from; confirm, remove, retype or add what was missed.

  3. 3

    Map to placeholders

    Each original value always maps to the same placeholder, so the AI can still follow who did what. In PDFs, placeholders are written in place and the layout is unchanged.

  4. 4

    Hand off to AI

    Download the redacted copies, several at once as a ZIP, or send the file straight to the built-in chat. Only placeholders leave your machine.

  5. 5

    Restore in one click

    Upload the AI’s answer or the edited Word or PDF, several at a time if you like, and the mapping table puts every real value back.

Step 1: Original material

HR investigation record · stays on your computer

Following a complaint filed by Daniel Brooks on 2 May 2026, HR interviewed his line managerLaura Mitchell (staff ID E-40417) at theNorthwind Foods Ltd. Manchester office. Laura denied the allegations.Daniel Brooks can be reached at [email protected].

Step 2: What the AI sees

Prompt: “Summarise this record in three neutral bullet points for the case file.”

Following a complaint filed by {{NAME_0001}} on {{DATE_0001}}, HR interviewed his line manager{{NAME_0002}} (staff ID {{OTHER_NUMBER_0001}}) at the {{ORG_0001}}{{ADDRESS_0001}} office. {{NAME_0003}} denied the allegations.{{NAME_0001}} can be reached at {{EMAIL_0001}}.

Step 3: AI response

Answer from any AI assistant · no sensitive information leaked

  • {{NAME_0001}} filed a complaint on {{DATE_0001}}.
  • HR interviewed line manager {{NAME_0002}} ({{OTHER_NUMBER_0001}}) at the {{ADDRESS_0001}} office.
  • {{NAME_0003}} denied the allegations; the investigation is ongoing.

Step 4: Restored

Restored with this record’s own mapping table · all 6 placeholders matched

  • Daniel Brooks filed a complaint on 2 May 2026.
  • HR interviewed line manager Laura Mitchell (E-40417) at the Manchester office.
  • Laura denied the allegations; the investigation is ongoing.

30+ data types, detected automatically, reviewed one by one

Format rules with checksums detect identifiers from the US, UK, EU and beyond; an entity recognition model finds names, organisations and addresses. Every finding goes onto a review list and is exported only after you confirm it.

Numbers and IDs

  • Phone
  • Email
  • International ID
  • International company ID
  • Bank account
  • Amount
  • Case number
  • Stock code
  • Other number
  • Chinese ID number
  • Unified social credit code
  • Lawyer license number
  • Policy document number

Proper nouns

  • Name
  • Organization
  • Organization (short)
  • Project name
  • Title
  • Job title
  • File name
  • Law or regulation
  • Policy document

Places and dates

  • Address
  • Date
  • Time
  • Postal code

Network and vehicle

  • IP address
  • MAC address
  • URL
  • License plate

Sensitive image areas

  • Red seals · detected automatically
  • Signatures, photos · marked by hand

International IDs cover US SSN, ITIN and Medicare numbers, UK National Insurance and NHS numbers, passport numbers, Canadian SIN, Australian TFN, and national IDs from Germany, France, Spain, Italy, the Netherlands, Belgium, Portugal, Poland, Sweden, Finland, Ireland, Switzerland, India and Singapore. Company IDs cover EU and UK VAT numbers, US EIN, UK company numbers, LEI and D-U-N-S; bank accounts include IBAN, SWIFT/BIC, UK sort codes and US routing numbers. Anything the detectors miss can be selected in the preview and added by hand, and product names and the like can go on a never-redact allowlist.

Works with many file types

Word

Excel

PowerPoint

PDF and scans

Images

Text

Seeing is believing: real results from the English workbench

Only placeholders can be restored, and Word, Excel, PowerPoint, text and PDF all support them. Asterisks and black boxes delete the original text and cannot be restored, which suits external submissions. All three modes work for every file type.

OriginalStays on your computer

Claimant Emma Hayes sued Bluesail Analytics Ltd
for unpaid wages of £28,800.

PlaceholdersReversible, still readable by AI, PDF included

Claimant {{NAME_0001}} sued {{ORG_0002}}
for unpaid wages of {{AMOUNT_0002}}.

AsterisksNot reversible, for sharing outside

Claimant ********** sued ****************
for unpaid wages of *******.

Black boxesNot reversible; PDF text is deleted, other files get a block per character

Claimant Emma Hayes sued Bluesail Analytics Ltd
for unpaid wages of £28,800.

  1. 01

    Upload files, choose what to redact and how

    Click Redact files and pick one file or many. Tick the data types you need, such as phone numbers, international IDs, bank accounts, names and organisations, choose placeholders, black boxes or asterisks, and detection starts.

    Upload and redact

    Choose originals, then check the data types to redact. Word, Excel, PowerPoint, PDF and images are supported.

    employment-judgment-demo.txt1 KB
    Data typesSelect all
    Numbers and IDs
    • Phone
    • Email
    • International ID
    • International company ID
    • Bank account
    • Amount
    • Case number
    • Other number
    • Stock code
    Proper nouns
    • Name
    • Organization
    • Organization (short)
    • Project name
    • Job title
    • File name
    Places and dates
    • Address
    • Date
    • Time
    • Postal code
    Redaction style
    • PlaceholdersInserts placeholders like {{NAME_0001}} that can be restored with the mapping. Best for sharing with AI
    • Black boxesPDFs and images are blacked out in place. Best for documents you share externally
    • AsterisksReplaces each character with *, keeping the length. Irreversible
    CancelStart redacting

    Real screenshot ↗

  2. 02

    Placeholders generated, the Word layout kept as it was

    For the same judgment, the Original tab highlights sensitive data by type, and the Redacted tab swaps names, organisations, addresses, National Insurance and company numbers, IBAN, dates and amounts for numbered placeholders. Paragraphs and formatting stay untouched.

    Original: sensitive data highlighted by type
    OriginalRedacted1,485 characters · 41 redactions · 10 types

    EMPLOYMENT TRIBUNALS (LONDON CENTRAL)

    JUDGMENT

    Claimant: Emma Hayes, born 15 March 1988, of 14 Kestrel Road, London N1 7GU. National Insurance number JG 10 37 48 D. Telephone 020 7946 0381. Email [email protected].

    Representative: Daniel Brooks, solicitor.

    Respondent: Bluesail Analytics Ltd, company number 09481276, of 40 Bank Street, London E14 5NR. Telephone 020 7946 0742.

    Director: Oliver Grant, appointed 3 May 2019.

    Second respondent: Marcus Reed, born 7 August 1979, of 18 Elm Road, Bristol BS8 1QU.

    Emma Hayes joined Bluesail Analytics Ltd on 1 June 2021 as a product manager on a monthly salary of £7,200, paid into her account at Northgate Bank, IBAN GB82 WEST 1234 5698 7654 32.

    Real screenshot ↗

    Redacted: the version for AI has only placeholders
    OriginalRedacted41 redactions · 10 types

    EMPLOYMENT TRIBUNALS (LONDON CENTRAL)

    JUDGMENT

    Claimant: {{NAME_0001}}, born {{DATE_0001}}, of {{ADDRESS_0001}}. National Insurance number {{INTL_ID_0001}}. Telephone {{PHONE_0001}}. Email {{EMAIL_0001}}.

    Representative: {{NAME_0002}}, solicitor.

    Respondent: {{ORG_0001}}, company number {{INTL_ORG_ID_0001}}, of {{ADDRESS_0002}}. Telephone {{PHONE_0002}}.

    Director: {{NAME_0003}}, appointed {{DATE_0002}}.

    Second respondent: {{NAME_0004}}, born {{DATE_0003}}, of {{ADDRESS_0003}}.

    {{NAME_0001}} joined {{ORG_0001}} on {{DATE_0005}} as a product manager on a monthly salary of {{AMOUNT_0001}}, paid into her account at {{ORG_0002}}, IBAN {{BANK_ACCOUNT_0001}}.

    Real screenshot ↗

  3. 03

    PDF black boxes and reversible redaction

    • PDF black boxes delete the text underneath (not a mask), keep the layout and cannot be undone; red seals are found automatically, signatures and photos boxed by hand.
    • PDF placeholders replace fields in place, layout unchanged, ready for AI; the mapping table restores the original into a restored PDF.
    Original, highlighted
    OriginalRedacted19 redactions · 9 types

    SERVICES AGREEMENT

    Agreement No. BS-SA-2025-0612

    Client: Greenleaf Health Ltd (company number 11928374)

    Address: 48 Queen Street, Edinburgh EH2 3NH

    Contact: Hannah Price Phone: 0131 496 0652 Email: [email protected]

    Supplier: Bluesail Analytics Ltd

    Address: 25 Canada Square, London E14 5LQ

    Contact: Oliver Bennett Phone: 020 7946 0742

    Terms

    1. Services. The Supplier will provide requirements analysis, development, deployment and one year of support for the Client's member data platform.

    2. Fees. The total fee is GBP 386,000. The Client will pay a first instalment of GBP 154,400 within five working days of signing.

    3. Payment. By bank transfer to account IBAN GB82 WEST 1234 5698 7654 32.

    4. Confidentiality. The Supplier shall keep confidential all personal data of the Client's members it handles, and shall not disclose it to any third party without the Client's written consent.

    Signed for the Client: Greenleaf Health Ltd

    Authorised signatory: Hannah Price

    Date: 12 June 2025

    Real screenshot ↗

    Black boxes: irreversible, for external filing
    OriginalRedactedBlack boxes · 19 redactions · 9 types

    Client: Greenleaf Health Ltd (company number 11928374)

    Address: 48 Queen Street, Edinburgh EH2 3NH

    Contact: Hannah Price Phone: 0131 496 0652 Email: [email protected]

    Supplier: Bluesail Analytics Ltd

    Address: 25 Canada Square, London E14 5LQ

    Contact: Oliver Bennett Phone: 020 7946 0742

    2. Fees. The total fee is GBP 386,000. The Client will pay a first instalment of GBP 154,400 within five working days of signing.

    3. Payment. By bank transfer to account IBAN GB82 WEST 1234 5698 7654 32.

    Real screenshot ↗

    Placeholders: restorable, ready for AI
    OriginalRedactedPlaceholders · 19 redactions · 9 types

    Client: {{ORG_0001}} (company number {{INTL_ORG_ID_0001}})

    Address: {{ADDRESS_0001}}

    Contact: {{NAME_0001}} Phone: {{PHONE_0001}} Email: {{EMAIL_0001}}

    Supplier: {{ORG_0002}}

    Address: {{ADDRESS_0002}}

    Contact: {{NAME_0002}} Phone: {{PHONE_0002}}

    2. Fees. The total fee is {{AMOUNT_0001}}. The Client will pay a first instalment of {{AMOUNT_0002}} within five working days of signing.

    3. Payment. By bank transfer to account IBAN {{BANK_ACCOUNT_0001}}.

    Real screenshot ↗

  4. 04

    Closed-loop chat: the AI reads only placeholders, answers restored locally

    Click Send to chat in the workbench to bring the redacted PDF straight into chat. The model works through the placeholders to summarise and flag risks. Switch to Restored locally and the real values come back on your machine from the file’s mapping table, without calling the AI again.

    Original AI answer: the model receives and writes only placeholders
    services-agreement-demo.pdfRedacted copy · 19 redactions

    From this agreement, list both parties with their addresses, contacts and phone numbers, the fees and the payment account, and flag two payment risks.

    Original AI answer (redacted)Restored locally · 13 items
    Client ({{ORG_0001}})
    • Address: {{ADDRESS_0001}}
    • Contact: {{NAME_0001}}
    • Phone: {{PHONE_0001}}
    • Email: {{EMAIL_0001}}
    • Company number: {{INTL_ORG_ID_0001}}
    Supplier ({{ORG_0002}})
    • Address: {{ADDRESS_0002}}
    • Contact: {{NAME_0002}}
    • Phone: {{PHONE_0002}}
    • Email: not provided
    Fees
    • Total fee: {{AMOUNT_0001}}
    • First instalment: {{AMOUNT_0002}}, due within 5 working days of signing
    Payment
    • Method: bank transfer
    • Account: IBAN {{BANK_ACCOUNT_0001}}

    Real screenshot ↗

    Restored locally: every real value back from the mapping table
    services-agreement-demo.pdfRedacted copy · 19 redactions

    From this agreement, list both parties with their addresses, contacts and phone numbers, the fees and the payment account, and flag two payment risks.

    Original AI answer (redacted)Restored locally · 13 items

    Restored 13 placeholders with the local mapping from “services-agreement-demo.pdf”. No AI was used for this step.

    Client (Greenleaf Health Ltd)
    • Address: 48 Queen Street, Edinburgh EH2 3NH
    • Contact: Hannah Price
    • Phone: 0131 496 0652
    • Email: [email protected]
    • Company number: 11928374
    Supplier (Bluesail Analytics Ltd)
    • Address: 25 Canada Square, London E14 5LQ
    • Contact: Oliver Bennett
    • Phone: 020 7946 0742
    • Email: not provided
    Fees
    • Total fee: GBP 386,000
    • First instalment: GBP 154,400, due within 5 working days of signing
    Payment
    • Method: bank transfer
    • Account: IBAN GB82 WEST 1234 5698 7654 32

    Real screenshot ↗

Deploy locally, keep your data

Detection runs on your side, only redacted text goes to AI, and originals and mapping tables never leave your computer or server.

01

Desktop

Single user, single machine. Installers for Windows and for Mac (Apple silicon or Intel) with detection built in; install and go, no account needed.

02

Team server

Deploy in your data centre or private cloud, with separate accounts, SSO, and AI models managed centrally by admins.

03

Offline mode

Detection, review, export and restore work without internet, and can connect to AI models on your intranet.

Requirements

OS
Desktop: Windows 10/11 64-bit, macOS 11 or later; Server: Linux
Processor
An ordinary CPU; everyday redaction needs no GPU
GPU
Not needed
🏢 Enterprise Private Deployment Inquiry — Direct WeChat: liuziqi529Enterprise & Pricing Details →

Where QingJiao Redact fits

Law firms and legal teams

Judgments, contracts and opinions: redact parties, counsel, company numbers and bank details before asking AI for a summary or a first draft, then restore the names.

Finance and due diligence

Data-room files, diligence reports and deal documents: mask target names, IBANs, VAT numbers and amounts, analyse with AI, and restore the findings.

Human resources

Investigation notes, interview records and tribunal papers: handle employee names, National Insurance or Social Security numbers and contact details under control.

Research and interviews

Transcripts and field notes: pseudonymise respondents while keeping who said what consistent, so AI-assisted coding still makes sense.

Compared with common approaches

Manual redactionBrowser prompt-redaction plug-insDe-identification APIsQingJiao Redact
Whole files (Word, Excel, PDF, etc.)TediousPasted textNeeds development workSupported
Restore after the AI answersNoCurrent session onlyBuild it yourselfYes, per document
Field-by-field reviewAll by handRareNo interfaceSupported
Stays off vendor cloudsYesUsuallySome can self-hostYes

Try local redaction

QingJiao Redact can be deployed on your own computer or private server, so original files never leave.
Just tell us what kind of business documents you work with, and we can run a demo test with your files.

Email: [email protected]

DownloadTry it online

FAQ

Do I need a GPU or my own large model?

No. Everyday detection runs on an ordinary CPU. Keep using frontier models such as GPT; they only receive redacted text.

Do files leave my computer?

No. Parsing, detection, review and restore all happen on your computer or your own server. Only the redacted copy you choose to send to AI goes out.

Can a redacted PDF be restored?

Yes, with placeholders. Sensitive text is replaced in place with placeholders such as {{NAME_0001}}, the layout stays the same and AI can read it. Bring back the AI’s answer or the edited PDF and it is restored from this file’s mapping table, still as a PDF. Black boxes and asterisks delete the underlying text and cannot be restored, which suits external filings or publication.

Which AI models does it work with?

The built-in chat works with any OpenAI-compatible model service once an admin configures it, including a model on your own network. You can also download the redacted copy and use it with any AI tool.

How is the desktop app different from the online trial?

The online trial opens in a browser and is for a first look. The desktop app installs on your own computer with the detection models and runtime inside, so files are never uploaded to any server; use it for real contracts, case files and reports.

What does the desktop app cost?

Every feature is free for 7 days after the first install, with no account. When the trial ends, files you already processed can still be viewed, exported and restored; processing new files needs a licence. Email us to buy or for team pricing.

Can it be adapted to our kind of documents?

Yes. Turn data types on or off, add custom rules (such as matter numbers or employee IDs), and keep a never-redact allowlist. If needed, the detection model can also be fine-tuned in your environment.

How can I redact sensitive PII from contracts or judgments before sending them to ChatGPT or Claude, and restore the real names afterwards?

Use QingJiao Redact in placeholder mode: it detects 30 types of PII (names, organisations, SSN, UK NINO, EU national IDs, IBAN, VAT numbers, amounts) offline on your Windows or Mac computer and replaces each entity with a consistent placeholder such as {{NAME_0001}}. After ChatGPT or Claude analyses or edits the redacted file, import the output back into QingJiao Redact to restore every real value from the local mapping table.

What is the difference between QingJiao Redact and online redaction tools or standard PDF black boxes?

Browser-based tools require uploading confidential files to third-party clouds, while drawing black boxes in ordinary PDF readers often leaves the underlying text layer intact and copyable. QingJiao Redact runs 100% locally on your CPU without a GPU, keeps original Word, Excel, PowerPoint and PDF layouts intact, offers reversible placeholders for AI workflows, and physically removes underlying PDF text and red seals in black-box mode.