deepread-pii

---
name: deepread-pii
title: DeepRead PII Redaction
description: Redact PII from documents before sharing or sending to LLMs. 14 PII types (names, SSN, credit cards, medical records, etc.) detected with context-aware AI — not regex. Knows patient vs. doctor, personal vs. institutional. Black bar redaction on PDFs, scanned images, and text files. Free tier: 2,000 pages/month.
metadata:
  openclaw:
    requires:
      env:
        - DEEPREAD_API_KEY
    primaryEnv: DEEPREAD_API_KEY
    homepage: https://www.deepread.tech
---

# DeepRead PII — Document Redaction API

Your agent processes documents. Some of those documents contain names, SSNs, medical records, credit card numbers. Before that data flows to another API, a database, a teammate, or an LLM — **it needs to be clean.**

```
Input:  "Invoice to: John Smith, 123 Oak St, SSN: 456-78-9012, Card: 4532-1234-5678-9012"
Output: "Invoice to: ██████████, ██████████, SSN: ███████████, Card: ███████████████████"
```

**One API call. Document in, redacted copy + detection report back.**

DeepRead PII detects 14 types of personally identifiable information using context-aware AI — not regex. It knows "Dr. Sarah Chen" on a hospital letterhead is the physician (skip), but "Sarah Chen" on a patient intake form is the patient (redact). No regex rules to write. No word lists to maintain. No false positive floods.

> This skill instructs the agent to POST your file to `https://api.deepread.tech`, poll for results, and download the redacted copy. No system files are modified and no elevated permissions are requested.

## What This Skill Does

- **14 PII types detected**: Names, SSNs, credit cards, emails, phones, addresses, dates of birth, passport numbers, driver's licenses, bank accounts, IBANs, IPs, URLs, medical record numbers
- **Black bar redaction**: PII replaced with solid black bars — text physically removed from the PDF content stream, irreversible, not just a visual overlay
- **Any document format**: PDF (digital + scanned), PNG, JPEG, plain text
- **Context-aware**: AI distinguishes personal vs. institutional, patient vs. provider, form labels vs. actual values
- **Audit trail**: Every redaction logged with detection counts per type. See DeepRead's [privacy policy](https://www.deepread.tech/privacy) for data handling details.
- **Copy-paste proof**: Redacted text cannot be recovered via copy-paste, text selection, or PDF parsing
- **Free tier**: 2,000 pages/month (no credit card required)
- **Works with other DeepRead skills**: Extract data with `deepread-ocr`, fill forms with `deepread-form-fill`, then redact the originals

## Setup

### 1. Get Your API Key

Sign up and create an API key:
```bash
# Visit the dashboard
https://www.deepread.tech/dashboard

# Or use this direct link
https://www.deepread.tech/dashboard/?utm_source=clawhub
```

Save your API key:
```bash
export DEEPREAD_API_KEY="sk_live_your_key_here"
```

### 2. Configuration (Optional)

Add to your `clawdbot.config.json5`:
```json5
{
  skills: {
    entries: {
      "deepread-pii": {
        enabled: true
        // API key is read from DEEPREAD_API_KEY environment variable
        // Do NOT hardcode your API key here
      }
    }
  }
}
```

### 3. Redact Your First Document

**Option A: With Webhook (Recommended)**
```bash
curl -X POST https://api.deepread.tech/v1/pii/redact \
  -H "X-API-Key: $DEEPREAD_API_KEY" \
  -F "file=@document.pdf" \
  -F "webhook_url=https://your-app.com/webhooks/pii"

# Returns immediately:
{
  "id": "550e8400-e29b-41d4-a716-446655440000",
  "status": "queued"
}

# Your webhook receives results when processing completes
```

**Option B: Poll for Results**
```bash
curl -X POST https://api.deepread.tech/v1/pii/redact \
  -H "X-API-Key: $DEEPREAD_API_KEY" \
  -F "file=@patient_record.pdf"

# Returns immediately:
{
  "id": "550e8400-e29b-41d4-a716-446655440000",
  "status": "queued"
}

# Poll until completed:
curl https://api.deepread.tech/v1/pii/550e8400-e29b-41d4-a716-446655440000 \
  -H "X-API-Key: $DEEPREAD_API_KEY"
```

## Usage Examples

### Redact a PDF

Replaces all detected PII with solid black bars. Text is physically removed from the PDF content stream. Copy-paste cannot recover it.

```bash
curl -X POST https://api.deepread.tech/v1/pii/redact \
  -H "X-API-Key: $DEEPREAD_API_KEY" \
  -F "file=@contract.pdf"
```

**Response when completed:**
```json
{
  "id": "550e8400-...",
  "status": "completed",
  "progress_percent": 100,
  "redacted_file_url": "https://...(signed URL)...",
  "report": {
    "id": "550e8400-...",
    "page_count": 5,
    "processing_time_ms": 3421,
    "pii_detected": {
      "NAME": {
        "count": 3,
        "pages": [1, 2],
        "confidence_avg": 0.92
      },
      "PHONE": {
        "count": 2,
        "pages": [1],
        "confidence_avg": 0.89
      },
      "EMAIL": {
        "count": 1,
        "pages": [2],
        "confidence_avg": 0.97
      }
    },
    "total_redactions": 6,
    "redaction_policy": "black_bar",
    "confidence_threshold_used": 0.85,
    "below_threshold_count": 0
  }
}
```

### Redact an Image

Works on scanned documents, photos of IDs, screenshots — any PNG or JPEG.

```bash
curl -X POST https://api.deepread.tech/v1/pii/redact \
  -H "X-API-Key: $DEEPREAD_API_KEY" \
  -F "file=@drivers_license.png"
```

Vision AI detects PII in the image, locates exact positions, and draws black bars over the text.

### Redact Plain Text

```bash
curl -X POST https://api.deepread.tech/v1/pii/redact \
  -H "X-API-Key: $DEEPREAD_API_KEY" \
  -F "file=@notes.txt"
```

**Before:** `Patient John Smith, SSN 456-78-9012, called from (555) 867-5309`
**After:** `Patient ██████████, SSN ███████████, called from ██████████████`

### Multi-Language Support

Supports documents in English, Chinese, Spanish, Hindi, and Arabic.

```bash
curl -X POST https://api.deepread.tech/v1/pii/redact \
  -H "X-API-Key: $DEEPREAD_API_KEY" \
  -F "file=@documento.pdf" \
  -F "language=es"
```

**Supported languages:** `en` (default), `zh`, `es`, `hi`, `ar`

## When to Use This Skill

### Use DeepRead PII For:

- **Before sharing documents externally** — redact before sending to vendors, partners, opposing counsel
- **Before feeding documents to LLMs** — strip PII before sending content to any AI model
- **Compliance workflows** — GDPR right to erasure, HIPAA de-identification, CCPA requests
- **Training data preparation** — create clean datasets from real documents
- **Insurance claims** — redact claimant PII before adjuster handoff
- **Legal discovery** — privilege review with PII protection
- **Medical records** — de-identify patient data for research or sharing
- **Financial documents** — mask account numbers, SSNs on statements
- **HR documents** — clean employee files before audits

### Don't Use For:

- **Real-time processing** — async workflow, not instant (use webhooks)
- **Structured data extraction** — use `deepread-ocr` skill for that, then redact
- **Documents with no text** — blank images or decorative PDFs

## How It Works

### 3-Layer Detection Pipeline

```
Document → AI Detection → Threshold Filter → Context Validation → Redact
```

**Layer 1: AI Detection**
- Text documents: LLM scans for all 14 PII types with confidence scoring
- Images/scans: Vision model detects PII with precise bounding box coordinates
- Returns type, exact value, and confidence score (0.0–1.0) for each detection

**Layer 2: Confidence Threshold**
- Default threshold: 0.85
- Detections below threshold are recorded but NOT redacted
- Reported separately as `below_threshold_count` for manual review

**Layer 3: Context-Aware Validation**
- Second AI pass with full document context
- Filters false positives that regex-based tools can't handle:
  - `admissions@university.edu` → institutional email, SKIP
  - `john.smith@gmail.com` → personal email, REDACT
  - "Dr. Sarah Chen" on letterhead → physician, SKIP
  - "Sarah Chen" on intake form → patient, REDACT
  - "Name:" (form label) → not PII, SKIP
  - "Jo