Skip to content

Home

Jump to bottom
Nick edited this page Dec 26, 2025 · 7 revisions

Masker API - PII Redaction for LLMs

Privacy-first API that removes personal information from text and JSON before sending to ChatGPT, Claude, or any LLM.

📖 API Documentation: https://masker.kikuai.dev/docs
🌐 Landing Page: https://kikuai-lab.github.io/masker/

🔐 Privacy Guarantee

We store NOTHING. We log NOTHING. Your data is processed in-memory and immediately discarded.

  • ✅ No database - All processing is in-memory
  • ✅ No file storage - Nothing is written to disk
  • ✅ No content logging - Only metadata (method, path, status) is logged
  • ✅ Stateless - Each request is independent
  • ✅ GDPR compliant - No personal data retention

Quick Start

curl -X POST "https://masker.kikuai.dev/v1/redact" \
  -H "Content-Type: application/json" \
  -d '{
    "text": "Contact John Doe at john@example.com",
    "mode": "placeholder"
  }'

Response:

{
  "redacted_text": "Contact <PERSON> at <EMAIL>",
  "items": [
    {"entity_type": "PERSON", "start": 8, "end": 16, "score": 0.85},
    {"entity_type": "EMAIL", "start": 20, "end": 36, "score": 1.0}
  ],
  "processing_time_ms": 15.29
}

Features

  • ✅ Text & JSON - Process plain text or entire JSON structures recursively
  • ✅ Multiple PII Types - EMAIL, PHONE, CARD (regex), PERSON (NER)
  • ✅ Flexible Modes - mask (***), placeholder (<TYPE>)
  • ✅ Privacy First - Stateless, no logging, no data storage
  • ✅ Multi-Language - English & Russian NER support
  • ✅ Fast - Average processing time: 15-25ms

Documentation

Use Cases

  • Clean user messages before sending to ChatGPT/Claude
  • Anonymize support tickets for AI analysis
  • Process form data before LLM classification
  • Sanitize logs for AI-powered monitoring

Supported PII Types

Type Description Detection Method
EMAIL Email addresses Regex
PHONE Phone numbers International formats
CARD Credit card numbers Regex + Luhn validation
PERSON Person names spaCy NER (EN/RU)

License

AGPL-3.0

Built by KikuAI Lab

Clone this wiki locally