Lawbot脱敏猫: Local Offline Legal Document Redaction for Secure AI Workflows

Quick Take

Lawbot脱敏猫 is a dedicated offline desktop legal document redaction software. It automatically detects, reviews, replaces, or masks confidential entities in contracts, pleadings, court judgments, formal notices, and photographic evidence locally on your machine. Sanitized texts can be analyzed using external LLMs, and real entities can subsequently be restored locally using an authorized key.

Product Positioning

According to verified facts, Lawbot脱敏猫 positions itself as the "security layer for legal AI," specifically catering to law firms, in-house legal departments, and corporate compliance teams. It is not a web-based SaaS platform and does not rely on external model APIs. Recognition, OCR, manual auditing, export, and reverse entity re-identification are executed locally on the client's device. The software enforces a default offline posture that actively blocks external network requests.

From an editorial perspective, Lawbot脱敏猫 does not aim to make large language models smarter; rather, it makes the ingestion of highly confidential legal materials into AI workflows manageable and compliant. Consequently, evaluations must center around privacy boundaries, pseudonym consistency, human audit overhead, supported file structures, and rule reuse, rather than generative output quality.

Core Capabilities

Automatic Detection of 15 Sensitive Information Categories

The software covers individual names, company/institution names and abbreviations, national ID numbers, phone numbers, bank account numbers, physical addresses, monetary sums, dates, email addresses, legal case/official document numbers, license plate numbers, official seals, WeChat IDs, Unified Social Credit Codes, and URLs. Official documentation notes an accuracy rate of approximately 95% for core sensitive entities. However, automated detection must always be paired with manual human review rather than assuming 100% precision.

Three Mixed Redaction Techniques

• Asterisk Masking: Replaces parts of characters while retaining partial readability.
• Color Blackout: Applies solid black, gray, white, or purple overlay strips on PDFs and images. Once exported, targeted text becomes completely unselectable and non-copyable.
• Pseudonymization: Substitutes real entities with uniform placeholders such as "Party A" or "Name 001". This format allows LLMs to extract arguments, draft summaries, perform cross-border translations, or review risks while preserving legal transaction structure.

Local Offline Processing and Hardware Requirements

Lawbot脱敏猫 supports Windows 10/11 x64 and macOS. For ordinary textual files, 8GB RAM is acceptable. For scanned documents, photographic evidence, or bulk tasks, hardware specifications of 16GB RAM, an SSD, and a recent CPU are recommended. The installer contains local inference models and OCR engines, allowing full operational capacity even when completely disconnected from the network.

Supported Formats and Batch Queuing

The tool accepts Microsoft Word (.docx), PDF, and image formats (.jpg, .jpeg, .png), with support for structured Markdown export. Users can drag and drop multiple DOCX, PDF, and image files into a parsing queue. Auditing can begin on completed files while subsequent documents are still processing in the background, which is ideal for voluminous litigation dockets and due diligence packages.

Document-Specific Routing and Word/PDF Deep Processing

Upon import, files can automatically route—or be manually assigned—to specialized recognition rules tailored to judgments, litigation briefs, commercial contracts, corporate disclosures, letters, or evidence screenshots. For Word files, document layout, formatting, tables, headers, and footers are preserved while scrubbing tracked changes and comments. For PDFs, native text is extracted directly, whereas scanned or hybrid PDFs undergo page-by-page OCR alongside pixel-level physical erasure.

Project Memory (.tmm), Custom Rules, and De-pseudonymization

The proprietary .tmm project memory file retains source files, OCR outputs, pseudonym mappings, blacklists/whitelists, masking coordinates, and review progress. Multiple files within the same case share consistent entity mapping. Blacklists enforce redaction, while whitelists prevent false positives on terms like stock exchanges; these rule sets can be imported and exported via .xlsx. When external AI tools generate responses, the software uses local authorized keys to replace pseudonyms back into true identities in one click.

Step-by-Step Usage Guide

1. Verify Environment: Ensure your computer runs Windows 10/11 x64 or macOS. Ensure at least 8GB RAM for text documents, or 16GB RAM plus an SSD for scanned materials.
2. Acquire License: Visit the official site. The factsheet notes a single-device personal license at ¥499 and a dual-device license at ¥599. Submit your name and email to obtain the download link and activation URL.
3. Install and Offline Activation: Open the client, copy your unique machine code, and activate offline via the provided invitation/activation link to bind the offline license.
4. Import Files: Drag contracts, petitions, judgments, notices, chat records, or payment receipts into the application.
5. Confirm Document Routing: Accept automatic classification or manually assign document types (e.g., judgments, contracts, evidence) to activate tailored detection profiles.
6. Run Automatic Identification: The local model annotates 15 sensitive fields at roughly 95% nominal precision.
7. Perform Manual Audit: Inspect the grouped list, delete false positives, click-and-drag to add missed items, and verify tables, headers, footers, and revisions.
8. Configure Schemes and Lists: Set case-specific masking depth. Add critical trade secrets to blacklists and standard regulatory bodies to whitelists.
9. Select Redaction Mode: Choose asterisks, irreversible color blackouts, or structural pseudonymization based on whether the document is going to public filing or AI prompts.
10. Save Project Memory (.tmm): Store work-in-progress mappings to keep pseudonym consistency across multi-document litigation bundles or future addendums.
11. Preview and Export: Verify pages visually. Export true-to-format Word, physically erased PDFs, or structured Markdown files designed for AI ingestion.
12. Submit to Cloud AI: Paste sanitized Markdown into Kimi, DeepSeek, or ChatGPT. The model interacts purely with pseudonyms, never seeing confidential identities, account details, or exact figures.
13. Restore Real Identities Locally: Import the AI's final analysis into Lawbot脱敏猫, select reverse pseudonymization, authenticate via your local key, and repopulate actual client names locally.

Common Use Cases

• AI-Assisted Drafting: Pre-clearing judgments, counterclaims, and evidence before prompting LLMs for precedent analysis or legal opinion outlines.
• External Disclosure: Blacking out accounts, bank numbers, case IDs, and personal names prior to exchanging records with opposing counsel, arbitrators, or regulators.
• Public Case Studies & Marketing: Redacting client details to publish compliant firm newsletters, training materials, or public legal analyses.
• M&A Due Diligence: Maintaining uniform pseudonyms across hundreds of transactional agreements and debt covenants.
• Cross-Border Translations: Providing sanitized contract structures to translation LLMs without exposing underlying enterprise identities.
• Financial & Visual Evidence Handling: Sanitizing payment slips, invoice receipts, and mobile chat screenshots using geometric shape masking.

Key Strengths

• Strict Privacy Isolation: True client-side offline execution, zero document uploads, and native interception of outbound web requests.
• Tailored Legal Taxonomies: Identifies high-frequency legal data points like court numbers, corporate credit codes, institutional abbreviations, and official stamps.
• Closed-Loop Workflow: Connects local sanitization, cloud AI intelligence, and local reverse re-identification seamlessly.
• Multi-Document Consistency: Unified .tmm project memory ensures "Party A" remains identical across all exhibits in a docket.
• Compliant Formats: Clean Word layout retention, true physical pixel-scrubbing on PDFs, and dedicated Markdown exports with automated processing summaries.

Limitations and Risk Factors

• No Absolute Accuracy Guarantee: The estimated 95% detection rate requires systematic human auditing. The publisher's disclaimer mandates that end users verify final compliance.
• High OCR Compute Cost: Multi-page scanned PDFs and images demand substantial CPU and RAM. Machines with 8GB RAM may experience slower processing.
• Desktop-Only Architecture: Lacks web-based collaboration or cloud sync; hardware migration requires machine code re-licensing per official rules.
• Platform Verification Needed: Intel-based Mac users must contact official customer support for compatible build packages.
• Local Endpoint Responsibility: Local data safety relies on workstation policies, disk encryption (e.g., BitLocker), and OS account permissions.
• Legal Compliance Remains with Counsel: Redaction mitigates data exposure but does not automatically fulfill jurisdictional, court-mandated, or client-specific privilege obligations without attorney oversight.

Price Verification Notice

Factsheet records indicate an official catalog price of ¥499 for a single-device personal license and ¥599 for a dual-device license, involving email submission, QR code payment, and a web activation URL. Because licensing tiers, discounts, compatibility patches, and device limits fluctuate, always check the official website at https://www.lawbotai.cn/ to confirm current commercial terms before purchasing.

FAQ

Does Lawbot脱敏猫 upload my legal documents to cloud servers?

No. As documented, Lawbot脱敏猫 is a standalone desktop application. Entity detection, OCR, review, file rendering, and identity restoration execute strictly on your local hardware. The application does not upload files, does not call third-party model APIs, and actively intercepts outgoing web requests.

What is the detection accuracy rate, and can I bypass manual review?

You cannot bypass manual review. The manufacturer claims an approximate 95% accuracy rate on core sensitive legal entities. To catch the remaining edge cases, the software provides search verification tools, grouping panels, and drag-to-add features. Legal compliance remains the user's responsibility.

Is it safe to feed pseudonymized legal text into Kimi, DeepSeek, or ChatGPT?

Safety is achieved by sanitizing data before sharing. Because external LLMs only see pseudonyms (e.g., "Company A", "Individual 001"), they never encounter real identities, case IDs, or financial values. Nonetheless, users should inspect exported files and follow their organization's data governance standards.

Can AI-generated draft responses be converted back to real names?

Yes. By loading the AI-generated text back into Lawbot脱敏猫 and selecting reverse pseudonymization, the software cross-references local memory using your authorized key to repopulate real entity names automatically on your computer.

Can it handle scanned PDFs and photographic evidence?

Yes. The client features an integrated local OCR engine that extracts text from scanned or mixed PDFs page by page. It also offers pixel-level physical erasing on images and scanned pages to prevent copy-paste leaks or visual transparency recovery.

What operating systems and document formats are supported?

It runs on Windows 10/11 x64 and macOS. Supported input formats include Microsoft Word (.docx), PDF, and image files (.jpg, .jpeg, .png), with export options for native formats and clean Markdown.

How can a legal team maintain consistent redaction rules across matters?

Teams can establish uniform policies by saving custom detection profiles and exporting shared blacklists and whitelists in .xlsx format, ensuring consistent standards across different fee earners.

What is the current purchase price for the tool?

Factsheet data notes ¥499 for a single-device personal license and ¥599 for a dual-device license. However, software pricing and promotion terms change periodically; verify final figures directly on the official Lawbot portal.

Alternative Approaches

Because verified sources do not provide a competitive comparison matrix of direct commercial rivals, unconfirmed vendors are excluded. Potential operational alternatives include:

• Manual Redaction & Two-Person Review: Maximum control without software purchases, but labor-intensive and error-prone across massive file dockets.
• Pre-OCR via External Tools (e.g., WPS): Recommended strictly as a hardware optimization pre-step to convert scans into text before local processing.
• Direct Ingestion into Kimi, DeepSeek, or ChatGPT: General LLMs lack local offline redaction, cross-file pseudonym tracking, and offline reverse mapping; unredacted uploads of sensitive files violate client confidentiality.
• Custom Internal RegEx & Blacklist Protocols: Developing in-house rules and review procedures. When considering third-party software, prioritize verification of true offline execution, auditability, and pseudonym consistency.

Sources and Verifications

• Primary tool factsheet 《Lawbot脱敏猫》, , verified on 2026-08-22T17:58:16.0087318+00:00.

• Official site portal: Lawbot脱敏猫, https://www.lawbotai.cn/, verified on 2026-08-22T17:58:16.3314576+00:00.

• All functional claims, operating requirements, licensing workflows, and pricing tiers derive strictly from the verified factsheet above. Variable parameters should be confirmed via real-time official channels.

• 《Lawbot脱敏猫》