LocalMask Product Documentation
Detection, review, redaction, export and local restoration run on your computer. Your data files stay local.Local offline document redaction software for professional workflows
LocalMask is designed for professionals and organizations with clear requirements for detection accuracy, data locality and a complete processing workflow. File access, sensitive-information detection, centralized review, content-level redaction, batch export and alias restoration run locally, while original files, detection results, processing records and alias mappings are not uploaded to the cloud.
LocalMask serves sensitive workflows across legal, finance, risk, audit, healthcare, international trade, tendering, procurement, government and enterprise operations, with continuing improvements for professional material in each field. Legal workflows are deeply supported across contracts, pleadings, case files, letters and evidence material, while documents, spreadsheets, images and scans are supported throughout the product.
LocalMask is currently used by legal departments at globally leading companies, law firms, banks, tax advisory firms, top-tier tertiary hospitals and professionals in public-safety work. Outside China, LocalMask is in use in countries including the United States, Germany, Poland, Romania and Liechtenstein.
Quick start
- 01
Select one or more files.
- 02
Choose the sensitive information categories and redaction mode.
- 03
Start redaction.
- 04
Review and adjust the detected results.
- 05
Export the redacted files.
- 06
If aliases were used, use Alias Restore to replace aliases in text or files returned by AI or another external system with the original information.
Core capabilities
File access, detection, review, redaction and export all run on the local computer, including in offline, intranet and controlled office environments.
LocalMask processes supported text layers and image pixels so the exported copy does not retain the processed source content.
Review detected items, categories, source locations and export decisions in one workspace. When you find a missed item, select the text or area to add an annotation, then batch-mark matching text or graphics throughout the current document.
Use fixed text, wildcards, regular expressions and graphic matching, with blocklists and allowlists to add or exclude specific content beyond the built-in models.
Results and review state are saved automatically as encrypted local records. Reopen them whenever needed without exporting, storing or importing project files.
Select multiple files or drag in an entire folder. Folder imports recursively process supported materials.
Use consistent aliases to preserve context, then restore content returned by AI or external systems locally.
Detect copied text locally, then write the redacted version back to the clipboard before pasting it into AI tools, web forms or external systems.
Batch and automatic exports preserve the original folder hierarchy for folder imports. Automatic Export is off by default and can be enabled at the bottom of the home-screen configuration panel with an output directory and filename suffix.
Detection accuracy and evaluation criteria
LocalMask continuously evaluates its core Chinese entity recognition model and improves it using performance observed across real-world materials. The results below are from the current production model's latest formal evaluation, completed on July 30, 2026. Evaluation data is strictly separated from training data.
Strict entity-level F1 Combines precision and recall. A prediction is correct only when both the entity type and its exact boundaries match.
Sensitive-information coverage Measures the proportion of sensitive characters covered by detection. It is not the same as entity-level recall.
We will continue expanding real-world evaluations, improving the model, and updating these public metrics after production model releases.
Supported materials
The following are common examples, not an exhaustive list.
Case files, reports, archives, supporting evidence, tender documents, bids, image-only PDFs and scanned PDFs.
Contracts, letters, tender documents, bids, procurement files, manuals, submissions and meeting minutes.
Payroll sheets, customer and supplier lists, quotation tables, procurement ledgers, financial spreadsheets, operational ledgers and statistical data.
Photos of documents and credentials, screenshots, chat screenshots and scanned images.
Plain-text material, edited drafts, prompts and notes.
Copied text from websites, chats, email, documents, AI prompts and web forms.
Sensitive information coverage
LocalMask covers many common types of sensitive information and supports custom rules for business-specific material.
Names, ID numbers, addresses, travel document numbers, phone numbers, emails, URLs, social accounts, ethnicity, gender, age, postal codes, vehicle plates, VINs and IP addresses.
Organization names, Unified Social Credit Codes, DUNS numbers, patent agency codes, judicial case numbers and official document numbers.
Patent numbers, patent application publication numbers, invention patent grant publication numbers, utility model grant publication numbers, design patent grant publication numbers, work registration numbers, software copyright registration numbers and software copyright certificate numbers.
Contract, medical record, visit, admission, and other reference numbers.
Bank accounts, card numbers, stock codes, invoice numbers, property certificate numbers, amounts and dates.
Handwritten signatures, seals, fingerprints, faces and QR codes.
Redact exported file names, directory names in preserved folder structures, and Excel worksheet names, and optionally remove Word headers and footers, common PDF watermarks, and PDF or Word comments and annotations.
Supports fixed text, wildcards, regular expressions and graphic matching.
Redaction scope settings
Organization names and official document numbers have independent redaction scope settings.
Choose to redact all public authority names, leave public authority names in cited public-authority documents unredacted, or leave all public authority names unredacted. This setting does not affect companies, commercial banks, schools, hospitals, or other organizations.
Choose to redact all official document numbers or leave the official document numbers of cited public-authority documents unredacted.
Redaction modes
LocalMask provides four common redaction output methods, plus formula-based transformations for dates and amounts. Choose a preset redaction-area color, or customize it with the color picker or a hexadecimal color value.
Replace sensitive content with letter or number aliases such as Person A or Person 1 while preserving relationships among people, organizations and addresses. Pseudonymization can automatically associate recognized names of the same organization, such as abbreviations and former names, so they are not mistaken for separate entities after redaction. Matching content uses the same alias within a batch. To keep aliases consistent across batches, you can reuse the alias mapping from a previous processing record.
Mask the full detected value when exported material should not expose readable source text.
Remove the detected content while preserving its original position and the rhythm of the layout, producing an effect closer to manual blanking.
Keep a small amount of context at the edges while masking the middle of IDs, accounts, card numbers or phone numbers.
Shift dates forward or backward by days, months or years so exported content does not retain the actual dates while preserving chronological order.
Apply addition, subtraction, multiplication or division consistently to preserve scale and calculation semantics while reducing exposure of actual amounts.
Custom rule engine
Custom rules stay on the local computer. Blocklists add content that should be redacted, while allowlists exclude content that should remain visible; when they overlap, the allowlist takes precedence. After selecting a logo, icon or other distinctive graphic, future documents can automatically find and mark that graphic.
Match fixed content such as customer names, project codes, internal abbreviations and public names.
Use * to match variable content.
Suitable for complex rules.
Select a logo, icon or other distinctive graphic so future documents can automatically find and mark that graphic.
Example: redact only your company while keeping other organization names
If you only want to redact your own company name while leaving other organizations visible, configure it as follows:
- 01
Before processing the files, turn off automatic Organization Name detection so other organization names are not marked.
- 02
Open Blocklist under Custom Rules, select Fixed Text and choose Organization Name as the label.
- 03
Enter one actual form per line, such as the full legal name, a commonly used abbreviation, the English name or a brand name.
- 04
After detection, verify in the review workspace that every form of your company name was found and other organizations were not marked.
Copy the prompt below into an AI tool and answer its questions with fictional or already-redacted examples. Do not send real sensitive information.
I need a regular expression for text matching. First ask me for several examples, then ask step by step which parts remain fixed, which parts vary, the character types and lengths of the variable parts, and which part of each example should ultimately be matched. If the information is incomplete, keep asking questions instead of making assumptions. Once the requirements are clear, generate the expression. It must be ready to paste directly, without /.../ delimiters, lookarounds or backreferences. It must not match empty content and should not span lines by default. Finally, explain the rule, test it against my examples and identify possible false matches.
Review workspace
After detection, review every item in the workspace. The document appears on the left and detected content is grouped by category on the right. The final workspace state determines the export.
Switch between two views at the top. Original verifies detected locations and source content; Redacted shows the output after pseudonymization, masking or removal.
When you find a missed item during review, select the text or area to add an annotation. For repeated text or the same graphic, batch-mark every match in the current document. Existing annotations can also be relabeled, assigned a different alias or removed.
Use the Search tab on the right to find text or annotations and jump to the relevant page. For long documents, select the page number at the top, enter a page and press Enter to jump directly to it.
Select a column header to redact the entire column, ignore the column or restore default detection, then adjust individual exceptions.
Mark logos and other graphics
- 01
Select the logo, icon or other graphic to process.
- 02
Choose a label, then select Annotate for the current graphic. If the same graphic should be processed in future documents, select Add to Blocklist.
Mark all occurrences of the same logo in the current document at once instead of selecting each one individually.
Processing records are saved automatically
After each run, the results and review state are automatically saved as an encrypted local processing record. When you need to continue, open the corresponding record from the home screen to review, adjust and export without exporting or importing a project file.
A local processing record is created automatically when processing finishes, with no separate save step.
Open the corresponding record from the home screen to return to the review workspace and continue.
There is no need to export, store and later re-import a project file.
Export
After review, choose an export format based on the recipient and purpose.
Preserves the main structure and layout of the source file for final delivery or continued editing.
Exports redacted content as lightweight Markdown for AI analysis or further processing.
Exports a fixed-layout copy for viewing across devices, archiving or external sharing.
Organizes redacted content from PDFs, scanned PDFs or images into a more compact, editable Word document with optimized paragraphs and table structure, suitable for AI uploads, search or continued editing.
Shows detected locations with highlights and comments for internal review. It still contains sensitive source text and is not suitable as an externally shared redacted result.
Exports only the processed text without the original layout, suitable for AI and text-processing workflows.
In Pseudonymization mode, summarizes categories, sensitive source text, aliases and occurrence counts. Reused mappings not found in the current run are retained with a count of 0 for future imports. Store it securely.
In other redaction modes, summarizes categories, sensitive source text, redacted results and occurrence counts for internal verification and records. Store it securely.
Pseudonymization and restoration
LocalMask can replace sensitive content with consistent aliases such as Person A or Person 1, so exported content does not retain source identities while preserving relationships among people, organizations and addresses. Pseudonymization can automatically associate recognized names of the same organization, such as abbreviations and former names, so they are not mistaken for separate entities after redaction. Matching content uses the same alias within a batch. To keep aliases consistent across batches, you can reuse the alias mapping from a previous processing record.
When AI or another external service returns text or a file containing those aliases, restore it locally. Alias mappings and restoration remain on the local computer.
- 01
Open Alias Restore from the corresponding processing record.
- 02
Paste the text returned by AI, or select the file to restore.
- 03
LocalMask restores the content using the alias mapping in that record.
- 04
Review the result and save the restored text or file.
Automatic Export
Enable it at the bottom of the home-screen configuration panel, then select an output directory and filename suffix. Folder imports retain their original directory hierarchy. Automatic Export is off by default.
Performance settings
Select the CPU indicator in the bottom status bar to adjust the thread count based on your device's performance. LocalMask optimizes local processing performance across platforms and device configurations, and some devices can also enable hardware acceleration for faster processing.
Large-document capability
LocalMask supports documents with thousands to around ten thousand pages. Processing time depends on the document and the device; use the tested scale and factors below as a reference.
LocalMask supports documents with thousands to around ten thousand pages. In internal testing, a 1,000-page document took about 3–4 minutes on the test device; a user processed a Word document of about 8,000 pages in over 40 minutes.
Actual time depends on hardware, file type, page complexity, scan quality, image count and the number of detected items.
We continue to improve large-document processing performance and the review experience.
Clipboard redaction
Clipboard redaction handles sensitive text copied from websites, chats, email, documents and other applications without sending the content to a server. LocalMask detects the copied text locally and can write the redacted result back to the system clipboard before you paste it into an AI tool, web form or external system.
- 01
Open Clipboard redaction settings and enable the feature.
- 02
Copy the text you want to process. LocalMask detects sensitive information on the local computer.
- 03
Review the result when needed, then copy the redacted content with the action button or configured shortcut.
- 04
Paste the redacted text into the intended AI tool, web form, email or external system.
Legal document redaction
Legal documents are a core training and continuous-improvement focus for LocalMask. The core Chinese entity recognition model is specifically trained on contracts, pleadings, case files, letters and evidence material, with an emphasis on names, organizations and addresses in complex contexts. Other detection capabilities cover identification numbers, contact details and case numbers.
LocalMask serves lawyers, law firms and in-house legal teams with clear requirements for detection accuracy, data locality and a complete processing workflow, including local detection, centralized review, content-level redaction, batch export and alias restoration. For cited statutes, regulations, official documents and other public material, the redaction scope of public authority names and official document numbers can be adjusted as needed to preserve the integrity and readability of the citation. Beyond legal material, it also supports sensitive workflows in finance, healthcare, tax, HR, international trade, procurement and government, with continuing improvements for different professional materials.
Overall public evaluation results for the core Chinese entity recognition model: 98.25% strict entity-level F1 with 99.26% sensitive-information coverage on the full held-out test set, and 97.48% strict entity-level F1 with 99.04% coverage on the formal human-annotated benchmark. See Detection accuracy for the evaluation methodology.
Use cases
Redact bank statements, accounts, card numbers, transaction details, credit material, audit workpapers, financial statements and compliance material before sharing, archiving or AI analysis.
Detect and redact relevant identifiers in patent applications, grant publications, work registrations and software copyrights before AI analysis, external collaboration or public circulation.
Detect and redact contacts, contact details, addresses, organization names and accounts in bilingual contracts, commercial invoices, packing lists, bills of lading, purchase orders and customer or supplier material before sharing, translation or AI processing.
Process sensitive information in internal reports, approval material, collaborative documents and archives before circulation.
Process employee records, resumes, compensation and performance data, invoices, reimbursements, payroll and financial spreadsheets.
Redact tender documents, bids, procurement files, supplier material, quotation tables and evaluation material before sharing, archiving, AI analysis or internal circulation.
Process sensitive information locally before submitting contracts, reports, screenshots, chat records or spreadsheets to AI.
Detect and redact names, ID numbers, contact details, medical record numbers, visit numbers, admission numbers and faces in health checks, medical records and diagnostic or laboratory reports before sharing or AI processing.
Complete local redaction and human review of mediation material, government documents and document batches before submission, collaboration or archiving.
Platform, installation and licensing
Enterprise licensing and device management
LocalMask supports silent deployment on Windows, macOS and Linux, along with enterprise batch deployment and centralized licensing for teams and organizations. Windows includes an MSI installer for centralized deployment, while macOS and Linux can be deployed through scripts or enterprise device management tools. Enterprise administrators can view license capacity and device status, approve device requests in batches, and disable or re-enable devices. When personnel leave or devices change, license capacity can be reclaimed and reassigned.
Continuous maintenance and issue response
LocalMask uses a rapid-iteration, continuous-delivery update model. Routine product issues are typically fixed, regression-tested and released within two business days after they are reported, without waiting for a scheduled major version.
Detection accuracy, professional-material support, document handling and usability continue to improve.
Check for and install updates in the app by selecting “Check for updates” at the bottom, without downloading and reinstalling from the website each time.
System requirements
Windows 10 64-bit, x64, dual-core CPU with four threads and 4GB memory.
Windows 10/11 64-bit, quad-core CPU with eight threads and at least 8GB memory.
macOS 12 Monterey or later with 4GB memory.
At least 8GB memory; 16GB or more for large documents, high-resolution scans or many images.
Desktop Linux on x86_64 or ARM64 / aarch64, with at least 4GB memory recommended.
At least 8GB memory; validate the target distribution and desktop environment before procurement.
No dedicated GPU or CUDA is required.
Integrated graphics are sufficient.
Comparison with PDF editors
Ordinary PDF editors focus on manually annotating, covering or editing individual files.
LocalMask provides a complete workflow for detection, review, redaction, export and local restoration.
After visual covering, underlying text may remain copyable, searchable or extractable.
LocalMask emphasizes content-level redacted exports by processing text layers and image pixels.
Multiple files and pages require manual inspection one by one.
LocalMask batch-processes PDFs, Word documents, spreadsheets, images, scans and text.
Blacking out everything can reduce the readability of the context.
Pseudonymization preserves relationships among people, organizations and addresses for AI analysis.
Processed items are scattered and difficult to verify consistently.
The workspace centralizes detected items, categories, source locations and export status.
Privacy and security boundaries
- Source-file copies, detection results and review content within processing records are stored locally in encrypted form and are not uploaded to cloud services.
- When exporting supported Word or PDF files, LocalMask processes identifiable document properties and metadata such as authors, comment authors and revision authors to reduce identity information left outside the main content.
- LocalMask supports local redaction of sensitive material before it is sent to AI, cloud services, web forms or external systems.
Frequently asked questions
Does redaction require an internet connection?
No. Detection, review, redaction, export and local restoration all run on the local computer and can be used offline or on an intranet.
Will my original files be uploaded?
No. LocalMask does not upload original files, detection results, processing records, redaction settings or alias mappings to LocalMask or any third party.
Is LocalMask suitable before sending material to AI?
Yes. Detect and process sensitive information locally, then export redacted Markdown, Word or another format for AI analysis. Pseudonymization replaces direct identifiers with stable aliases while preserving relationships among people, organizations and addresses.
Is LocalMask suitable for legal professionals?
Yes. LocalMask supports contracts, pleadings, case files, letters and evidence material. It detects sensitive information such as names, organizations, addresses, identification numbers, contact details and case numbers, then handles detection, review, redaction and export locally. LocalMask also supports other sensitive workflows in finance, healthcare, tax, international trade, procurement and government.
Is LocalMask specifically trained for legal documents?
Yes. The core Chinese entity recognition model is specifically trained and continuously improved for contracts, pleadings, case files, letters and evidence material, with an emphasis on names, organizations and addresses in complex legal contexts. LocalMask also covers professional identifiers such as case numbers.
When is LocalMask the better fit?
LocalMask is a strong fit when you need to process contracts, pleadings, case files, evidence material or other highly sensitive documents in professional work and value public accuracy metrics, fully local processing, human review, content-level redaction, batch handling, local processing records and alias restoration. It serves lawyers, law firms, in-house legal teams and professional workflows in finance, healthcare, tax, international trade, procurement and government.
Are processing records saved automatically?
Yes. Results and review state are automatically encrypted and stored on the local computer. Open the corresponding record from the home screen whenever needed to continue reviewing, adjusting and exporting, without manually exporting or importing a project file.
Which files and materials are supported?
LocalMask supports PDF, Word (DOC / DOCX), Excel (XLSX / XLS / XLSM), CSV, TXT, Markdown, scans, images and clipboard text. Select multiple files or drag in a folder to process supported file types recursively.
Do images of credentials, contracts or similar documents embedded in Word or PDF files need to be extracted separately?
No. Import the Word or PDF file directly. LocalMask detects and processes both the images and text in the document, so there is no need to extract the images for separate processing.
Can LocalMask process file names, directory names, worksheet names, headers, footers and watermarks?
Yes. LocalMask can redact exported file names, directory names in preserved folder structures, and Excel worksheet names, and optionally remove Word headers and footers, common PDF watermarks, and PDF or Word comments and annotations.
Does LocalMask support clipboard redaction?
Yes. Enable Clipboard redaction, then copy text from a website, chat, email, document or other application. LocalMask detects sensitive information locally and can write the redacted result back to the clipboard before you paste it into an external tool.
How accurate is detection?
The core Chinese entity recognition model achieves a strict entity-level F1 of 98.25% with 99.26% sensitive-information coverage on the full held-out test set, and a strict entity-level F1 of 97.48% with 99.04% coverage on the formal human-annotated benchmark. Actual results depend on file quality, layout and content.
Can a redacted export preserve the original layout?
A redacted copy in the original format preserves the source file's structure and layout. You can also export a PDF for viewing and archiving, an editable Word document for AI uploads, Markdown for AI analysis, or plain text as TXT.
What if a scanned PDF is too large to upload to an AI service?
Export redacted Markdown or Word (DOCX). Markdown is lighter and well suited to AI analysis; choose DOCX when you need to continue editing in Word. Both are usually easier to upload and process than the original scanned PDF.
How large a document can LocalMask process?
LocalMask supports documents with thousands to around ten thousand pages. In internal testing, a 1,000-page document took about 3–4 minutes on the test device; a user processed a Word document of about 8,000 pages in over 40 minutes. Actual time depends on hardware, file type, page complexity, scan quality, image count and the number of detected items.
Is LocalMask only adding a visual overlay?
No. LocalMask processes supported text layers and image pixels so exported redacted copies do not retain the processed source content instead of merely adding a removable visual overlay.
Is manual review still needed?
Yes. LocalMask assists with detection and redaction. Review detected items before export to confirm that the scope and output meet your requirements.
Can LocalMask process scanned PDFs, rotated images and handwritten signatures?
Yes. LocalMask can process scans, image-only PDFs, images with common rotations and handwritten signatures. Low resolution, complex backgrounds, skew, distortion and handwriting can affect detection.
Can LocalMask handle password-protected PDF and Word files?
Yes. Passwords are verified and used only on the local computer. The correct password is required; LocalMask does not bypass or crack unknown passwords.
What is the difference between an annotated Word copy and a redacted copy?
The redacted copy is for final delivery. The annotated copy is for internal review only: it retains sensitive source text and uses highlights and comments to identify it, so it must not be shared as a redacted result.
How can I review Excel files?
Review and adjust detected results, or select a column header to redact the entire column, ignore it or restore default detection. In Pseudonymization mode, aliases in Excel and CSV files can also be restored locally.
Which custom rules are supported?
LocalMask supports fixed text, wildcards, regular expressions and graphic matching. Blocklists add content to redact, while allowlists exclude content that should remain visible. After selecting a logo, icon or other graphic, LocalMask can automatically find and mark it in future documents.
Does LocalMask support pseudonymization and restoration?
Yes. Matching sensitive content can be replaced with consistent aliases so exported content does not retain source identities while preserving context. Pseudonymization can also automatically associate recognized names of the same organization, such as abbreviations and former names, so they are not mistaken for separate entities after redaction. When AI or another external service returns text or a file containing those aliases, restore it locally.