Anonymizer User Guide

Complete guide to protecting sensitive information with privacy-first PII detection and anonymization

Quick Navigation

→ Getting Started → Detection Methods → Basic Workflow → Sample Texts → File Upload → Entity Management → LLM Integration → Redact Mode → Export & Import → Privacy & Security → Limits & Review → Best Practices

1. Getting Started

The Anonymizer is a privacy-first proof of concept that detects and anonymizes Personally Identifiable Information (PII) inside your browser. It is useful for exploring local PII workflows, but it is not a certified anonymization or compliance system.

Privacy model: Your text and entity mappings stay local. Optional AI modes can download model files into your browser, but they do not send your text away for remote processing.

What is detected? The Anonymizer recognizes 75+ types of PII including:

  • Personal information: Names, emails, phone numbers, addresses
  • Financial data: Credit cards, IBANs, cryptocurrency addresses
  • Identification: SSNs, passports, driver licenses, tax IDs
  • Country-specific: Swiss AHV, German tax ID, UK NINO, and more
  • Technical: IP addresses, URLs, MAC addresses, UUIDs

2. Detection Methods

The Anonymizer offers several detection methods. Regex is the default and recommended starting point. OpenAI Privacy Filter is the advanced local AI option. AI4Privacy modes remain available as Legacy PoC options for comparison.

Regex - Pattern Matching

Default fast pattern-based detection

  • Instant results
  • Works completely offline
  • No model download required
  • Best for structured data such as emails, phones, IBANs, IDs, and URLs

Best for: quick checks and standard workflows

OpenAI Privacy Filter

Advanced local AI detection

  • Runs locally in WebGPU-capable browsers
  • Downloads model files from Hugging Face on first use
  • First load is about 950 MB
  • May catch contextual PII that regex misses

Best for: deeper review when download size is acceptable

AI4Privacy - Legacy PoC

Older experimental AI modes

  • English and multilingual variants remain selectable
  • Kept for comparison and experimentation
  • Not recommended as the primary AI path
  • Use Regex or OpenAI Privacy Filter first

Best for: comparison only

Model download note: When OpenAI Privacy Filter is selected, the browser downloads model weights, tokenizer, and configuration files. Your text, files, entities, mappings, and output are not sent to Hugging Face, OpenAI, or ASD123.ai for processing.

3. Basic Anonymization Workflow

1

Enter or Paste Text

Type or paste your text into the Input Text area. You can also drag-and-drop files.

Example: "My name is John Smith, email john@example.com, phone (555) 123-4567"
2

Select Detection Method

Choose from the dropdown: Regex - Pattern Matching for instant results, OpenAI Privacy Filter for advanced local AI detection, or AI4Privacy Legacy PoC for comparison.

3

Click Anonymize

Press the 🛡️ Anonymize button. The tool will detect all PII and replace it with placeholders.

Result: "My name is [PERSON_NAME_1], email [EMAIL_1], phone [PHONE_1]"
4

Review & Copy

Check the Detected Entities panel, review the anonymized output, and click 📋 Copy to use elsewhere.

4. Sample Texts

Use the small DE and EN buttons under the input area to load example texts quickly. They are intentionally small so the main workflow buttons stay in focus.

German Example

Contains names, address, email, phone number, customer number, insurance number, IBAN, doctor, and treatment date.

English Example

Contains names, address, date of birth, email, phone number, customer ID, account number, IBAN, doctor, clinic, and appointment date.

5. File Upload & Processing

The Anonymizer supports multiple file formats for easy text extraction:

📄

.txt Files

Plain text documents

📝

.docx Files

Microsoft Word documents

📕

.pdf Files

PDF documents (text-based)

Two Ways to Upload:

  • Click Method: Click 📁 Load File button and select your file
  • Drag & Drop: Drag file directly onto the Input Text area

⚠ Limits: Maximum file size is 10MB. Scanned PDFs (images) are not supported.

6. Entity Management

The Detected Entities panel shows all found PII with their placeholders and original values:

👁️ View Options

  • • Tiles View: Compact grid layout
  • • List View: Detailed list format

🔀 Sorting Options

  • • By Appearance: Order as they appear
  • • Alphabetical: Sort by entity type

🗑️ Managing Entities

  • Remove Entity: Click the trash icon (🗑️) on any entity to remove it and restore the original value in the output
  • Entity Stats: View total and active entity counts at the bottom of the panel
  • Manual Selection: Use "Anonymize Selected Text" to manually anonymize specific portions

7. LLM Integration & Deanonymization

Use the Anonymizer to safely process text with AI services like ChatGPT, Claude, or others:

1

Anonymize Your Text

Use the main anonymization feature to protect your PII. Copy the anonymized output.

2

Send to LLM

Paste the anonymized text into ChatGPT, Claude, or any AI service. The placeholders like [PERSON_NAME_1] will be preserved.

3

Get AI Response

The AI will process your text and maintain the placeholders in its response.

4

Deanonymize

Paste the AI response into the LLM Input field and click 🔓 Deanonymize. All placeholders will be replaced with original values!

💡 Pro Tip: Export your entities as CSV before closing the page. This lets you deanonymize responses even days later!

8. Redact Mode

For maximum privacy when sharing with untrusted parties, use Redact Mode:

🛡️ Anonymize Mode

Default mode with unique placeholders:

[PERSON_NAME_1], [EMAIL_1], [PHONE_1]

🔒 Redact Mode

All entities replaced with [redacted]:

[redacted], [redacted], [redacted]

When to use Redact Mode:

  • Sharing documents with untrusted parties
  • Public posting or publishing
  • When you don't need to reverse the anonymization
  • Maximum privacy protection

✓ Good to know: Even in Redact Mode, you can still deanonymize using the entity list, as the mapping is preserved internally!

9. Export & Import Entities

Preserve your entity mappings for later use or share them with trusted recipients:

💾 Export Entities

  1. 1. After anonymization, click 💾 Export CSV
  2. 2. Save the file with a descriptive name
  3. 3. Store securely (file contains PII!)
CSV Format:
"Placeholder","Original","Type","Active"
"[EMAIL_1]","john@example.com","EMAIL","true"

📥 Import Entities

  1. 1. Click 📥 Import button
  2. 2. Select your previously exported CSV file
  3. 3. Entity list will populate automatically
  4. 4. Now you can deanonymize LLM outputs!

⚠ Security Warning: CSV files contain the original ↔ placeholder mapping. Protect these files like you would the original sensitive data. Use encryption when storing or sharing.

10. Privacy & Security

Privacy Guarantees

  • ✓ Local Processing: Detection and anonymization run in your browser.
  • ✓ No Text Transmission: Your text and files are not sent away for remote processing.
  • ✓ Memory-Only Storage: Entity mappings stored in RAM only, cleared when you close the page.
  • ✓ Optional Model Downloads: OpenAI Privacy Filter downloads model files from Hugging Face, but not your text or mappings.

Best Security Practices:

  • Use on trusted devices only (not public computers)
  • Clear browser cache after processing highly sensitive data
  • Encrypt CSV exports before storing
  • Always review anonymized output before sharing
  • Use Redact Mode for untrusted recipients

11. Limits & Review

The Anonymizer is an experimental proof of concept. It can miss sensitive information, classify harmless text as PII, or produce incomplete mappings. Always review the detected entities and the final output manually.

Not a compliance guarantee: The tool does not certify GDPR, nDSG, HIPAA, legal, medical, financial, or other regulated workflows. Use it as a local aid, not as your only control.

12. Best Practices & Tips

💡 For Maximum Privacy

  • • Use Redact Mode for public sharing
  • • Clear all data after each session
  • • Don't share entity CSV files publicly
  • • Process offline when possible

🎯 For Best Detection

  • • Start with Regex mode for structured PII
  • • Use OpenAI Privacy Filter for deeper local AI review
  • • Review detected entities manually
  • • Format text properly (standard dates/phones)
  • • Remove false positives from entity list

⚡ For Performance

  • • Use Regex mode for instant results
  • • Process large docs in batches
  • • Close unused browser tabs
  • • Desktop browsers work best

🔄 For LLM Workflows

  • • Export entities before using LLM
  • • Keep CSV files organized
  • • Test deanonymization first
  • • Document your workflow

📚 Additional Resources

Need Help?

Supported Entity Types

75+ types including names, emails, phones, credit cards, SSNs, passports, IBANs, IP addresses, URLs, Swiss AHV, German tax IDs, UK NINO, and more.

Try the Anonymizer Now →