A local, privacy-first tool for anonymizing PII in text and sensitive regions in images before sharing or sending to cloud AI.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install has-privacy
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install has-privacy using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is HaS Privacy?
HaS (Hide and Seek) is a specialized privacy protection skill that runs entirely on-device to ensure data sovereignty. It utilizes advanced AI models to perform high-accuracy text anonymization across eight languages and precise image masking for 21 different privacy categories including faces, IDs, and financial documents. By integrating this into your workflow through Openclaw Skills, you can effectively mitigate privacy risks when interacting with cloud-based LLMs or sharing data with third parties.
The tool is split into two primary components: has-text, which uses a 0.6B parameter model to identify and replace sensitive entities with semantic tags, and has-image, which utilizes a YOLO11 segmentation model for pixel-level masking. This dual approach provides a comprehensive safety net for developers and organizations handling sensitive information in diverse formats.
HaS Privacy Use Cases
- Redacting PII from documents, resumes, or contracts before sharing them externally.
- Anonymizing sensitive text inputs before sending prompts to cloud LLMs and restoring the response later.
- Masking faces, license plates, and ID cards in photos before public distribution or storage.
- Scanning local workspaces and directories for unintended privacy leaks.
- Cleaning technical logs of sensitive user data before handing them to support or operations teams.
How HaS Privacy Works
- The skill identifies the target files (text or image) and determines the appropriate model to load locally.
- For text tasks, a local llama-server is initiated to process inference without sending data to an external server.
- The scan command identifies sensitive entities like names, addresses, or biometric data based on open-set entity types.
- The hide command replaces sensitive text with structured semantic tags and generates a secure mapping file.
- For images, the skill detects visual privacy carriers and applies a user-defined mask (mosaic, blur, or solid color).
- The seek command allows for the seamless restoration of anonymized text tags back to their original values using the saved mapping.
HaS Privacy Setup
To get started with this skill, ensure you have the necessary dependencies installed via your package manager:
# Install required binaries
brew install uv llama.cpp
# The system will automatically download the necessary models to your local directory:
# Text Model: has_text_model.gguf (~639 MB)
# Image Model: sensitive_seg_best.pt (~119 MB)
Before running text anonymization tasks, start the local inference server:
llama-server -m ~/.openclaw/tools/has-privacy/models/has_text_model.gguf --port 8080
HaS Privacy Data Schema & Taxonomy
HaS Privacy organizes data through structured semantic tagging and mapping files to ensure accuracy and reversibility in text processing:
| Component | Format | Description |
|---|---|---|
| Semantic Tags | <EntityType[ID].Category.Attribute> |
Replaces sensitive text with context-aware placeholders. |
| Mapping Table | JSON Dictionary | Stores the relationship between tags and original values for restoration. |
| Image Masks | Pixel Modification | Applied directly to the output image using mosaic, blur, or fill methods. |
| Scan Report | JSON / Text | Detailed breakdown of detected risks, locations, and sensitivity levels. |
All Openclaw Skills within this suite prioritize creating new files rather than overwriting original data.
HaS Privacy Advanced Features
- Support for 8 languages including Chinese, English, French, German, Spanish, Portuguese, Japanese, and Korean.
- Open-set entity detection allowing for custom, natural language privacy types beyond predefined categories.
- Cross-lingual restoration capabilities using model inference for complex translation workflows.
- Configurable image masking strength and methods to balance privacy with visual utility.
- Batch processing support for directory-wide scans and anonymization tasks.
- Seamless integration into automated AI agent pipelines for privacy-first development via Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Virtual Imam
A respectful, voice-enabled virtual Imam that guides users through the five daily prayers and Friday Jumu'ah khutbahs in multiple languages.
Setup Automatik VPS Manager
A powerful automation engine for deploying and managing self-hosted open-source applications and infrastructure on Linux VPS servers.
Setup Automatik
A powerful VPS management engine for automating the installation of Docker-based applications and infrastructure tools.
Solax Summary Fetch
This skill retrieves real-time energy production and inverter status data from the Solax Cloud API and returns it as structured JSON.
CareerForge CV Generator
An automated tool that orchestrates job searches and generates tailored, ATS-optimized CVs using AI.
Short Video Hook Generator
An automated generator for high-engagement, 3-second video hooks designed to maximize retention on short-form video platforms.
Featured*
