scrub-ai

Shield your prompts. Sanitize sensitive content before sharing with AI assistants.

---

The Problem

Every day, developers copy sensitive content into AI assistants without thinking:

❌ Stack trace with internal hostnames    → pasted into ChatGPT
❌ Application logs with session tokens   → pasted into Copilot
❌ Config files with database passwords   → pasted into Claude
❌ kubectl output with cluster names      → pasted into AI
❌ AWS CLI output with account IDs        → pasted into ChatGPT

Once that data leaves your machine, you have no control over it.

scrub-ai fixes this — it detects and masks sensitive content before you share it with any AI tool.

---

Features

• 🛡️ Secrets detection — API keys, tokens, passwords, private keys
• ☁️ Cloud detection — AWS account IDs, ARNs, GCP project IDs, Azure subscriptions
• 📡 Network detection — IP addresses, internal hostnames, internal URLs
• 🕵️ PII detection — emails, phone numbers, names via Presidio (optional)
• 🎯 Confidence scoring — filter low-signal matches with --min-confidence
• 🗂️ Named profiles — focus on aws, k8s, secrets, or network
• 📝 Custom patterns — add your own regex rules via a local JSON file
• 👁️ Watch mode — automatically sanitize clipboard whenever it changes (all platforms)
• ⌨️ Windows hotkey — press Ctrl+Alt+S to sanitize clipboard on demand
• 🖥️ System tray — runs quietly in the background (Windows)
• 📋 CLI — pipe any text through it from the terminal
• 📦 PyPI — install with a single pip install scrub-ai

---

Install

Standard install

Includes secrets, cloud, and network detection, profiles, and custom patterns. Works on Windows, Linux, and macOS.

pip install scrub-ai

With PII detection (optional, ~400 MB)

Adds detection of emails, phone numbers, and person names using Microsoft Presidio and spaCy.

# Step 1 — install the package with PII dependencies
pip install "scrub-ai[pii]"

# Step 2 — download the spaCy language model (required for PII to work)
python -m spacy download en_core_web_lg

If you skip Step 2, scrub-ai will still run — PII detection will silently do nothing.

---

Usage

Basic — pipe or file

# Pipe any text through it
cat error.log | scrub-ai

# Sanitize a file
scrub-ai --file crash.log

# See what would be detected without changing the output
scrub-ai --dry-run --file logs.txt

# Sanitize and copy the result to clipboard
scrub-ai --file logs.txt --copy

Filtering — profiles and confidence

Use profiles to focus on a specific category and ignore noise from others.

# Focus on AWS credentials only (ignores IPs, hostnames, etc.)
scrub-ai --profile aws --file logs.txt

# Focus on Kubernetes-related secrets
scrub-ai --profile k8s --file logs.txt

# Only mask high-confidence detections (0.0–1.0 scale)
scrub-ai --min-confidence 0.85 --file logs.txt

# Combine profile and confidence threshold
scrub-ai --profile secrets --min-confidence 0.90 --file logs.txt

Available profiles: aws, k8s, secrets, network

Each profile activates only the detectors relevant to that context. For example, --profile aws runs only AWS credential and ARN patterns — it will not mask IP addresses or internal hostnames.

Watch mode — automatic clipboard sanitization

Watch mode monitors your clipboard continuously. Every time you copy something, scrub-ai checks it and masks any sensitive content automatically before you paste.

Works on Windows, Linux, and macOS.

scrub-ai --watch

• Starts polling the clipboard every 500ms
• If sensitive content is detected, the clipboard is silently replaced with the clean version
• If nothing sensitive is found, the clipboard is left unchanged
• Press Ctrl+C to stop

Hotkey + system tray (Windows only)

For a manual, on-demand workflow on Windows. Runs as a background service with a system tray icon.

scrub-ai --start

• Icon appears in the system tray (bottom right)
• Copy any text with Ctrl+C as normal
• Press Ctrl+Alt+S to sanitize the clipboard
• Paste the clean text with Ctrl+V
• Right-click the tray icon to toggle the hotkey on/off, or to quit

--start is Windows only. For automatic clipboard sanitization on all platforms, use --watch instead.

---

PII Detection

When installed with pip install "scrub-ai[pii]" and the spaCy model is downloaded, scrub-ai automatically detects:

Type	Example input	Masked as
Person names	John Smith	[PERSON]
Email addresses	john@example.com	[EMAIL_ADDRESS]
Phone numbers	555-867-5309	[PHONE_NUMBER]

echo "Call John Smith at 555-867-5309 or john@example.com" | scrub-ai
# → Call [PERSON] at [PHONE_NUMBER] or [EMAIL_ADDRESS]

PII detection runs automatically alongside secrets, cloud, and network detection — no extra flags needed.

---

Custom Patterns

You can add your own regex rules to catch internal identifiers that scrub-ai doesn't know about.

Create the patterns file at:

• Linux/macOS: ~/.config/scrub-ai/patterns.json
• Windows: %APPDATA%\scrub-ai\patterns.json

[
  {
    "pattern": "ticket-\\d+",
    "replacement": "[TICKET]",
    "label": "internal_ticket",
    "confidence": 0.95
  }
]

Field	Required	Description
pattern	✅	Python regex string
replacement	✅	What to replace matches with
label	✅	Name shown in the detection summary
confidence	❌	Score from 0.0–1.0 (default: 1.0). Used with --min-confidence

Custom patterns are loaded on every run — no restart needed.

---

Example

Input:

ERROR 2024-01-15 14:32:01 - Connection failed
  host: db01.prod.internal
  password: myS3cretP@ss123
  aws_access_key_id: AKIAIOSFODNN7EXAMPLE
  aws_account_id: 123456789012
  ip: 10.0.1.45

Output:

ERROR 2024-01-15 14:32:01 - Connection failed
  host: [INTERNAL_HOST]
  password: [REDACTED]
  aws_access_key_id: [AWS_ACCESS_KEY]
  aws_account_id: [AWS_ACCOUNT_ID]
  ip: [IP_ADDRESS]

Detection summary (stderr):

Detected 5 sensitive value(s): aws_access_key=1, aws_account_id=1, internal_host=1, ipv4=1, password=1

---

What Gets Detected

Category	Examples
AWS credentials	Access keys, secret keys, session tokens
AWS infrastructure	Account IDs, ARNs, S3 URLs
GCP credentials	Service account keys, project IDs
Azure credentials	Subscription IDs, connection strings
Generic secrets	API keys, bearer tokens, JWTs, private keys, hex tokens
Passwords	password=, passwd=, pwd= key-value patterns
Network	IPv4, IPv6, internal hostnames, internal URLs
PII (optional)	Person names, email addresses, phone numbers

---

Roadmap

• Project setup
• v1.0 — CLI + secrets + cloud + network detection + Windows hotkey + system tray
• v1.1 — PII detection (Presidio) + confidence scoring + profiles + custom patterns
• v1.2 — Watch mode (automatic clipboard monitoring, all platforms)
• v2.0 — VS Code extension
• v2.1 — Browser extension (warns before pasting into ChatGPT)
• v3.0 — Team policies + audit log

---

Contributing

Contributions are welcome! Please read CONTRIBUTING.md first.

# Clone
git clone https://github.com/rajwindermarwaha/scrub-ai
cd scrub-ai

# Create and activate a virtual environment
python3 -m venv .venv
source .venv/bin/activate        # Linux/macOS
.venv\Scripts\activate           # Windows

# Install with dev dependencies
pip install -e ".[dev]"

# Optional: also install PII dependencies
pip install -e ".[pii]"
python -m spacy download en_core_web_lg

# Run tests
pytest

---

License

MIT — see LICENSE

---

Author

Built by @rajwindermarwaha

Built this because I had to put in the extra effort of copying everything into Notepad first and manually scrubbing it before sharing with AI tools. Figured others do the same.