policyboom 0.1.0

Enterprise legal risk intelligence CLI - analyze Terms of Service and Privacy Policies with multi-domain scanning

PolicyBoom 💥

Enterprise legal risk intelligence CLI for analyzing Terms of Service and Privacy Policies

PolicyBoom automatically discovers, analyzes, and scores concerning legal clauses across multi-domain company policies using a fluent dot-notation API.

Quick Start

# Install
pip install policyboom

# Scan a company's policies
policyboom exec "scan('slack.com').summarizeHigh()"

# Filter by category
policyboom exec "scan('stripe.com').summarizeHigh().category('arbitration')"

# Get all findings
policyboom exec "scan('example.com').summarizeAll()"

Features

• 🤖 AI-Powered Extraction - Uses Meta Llama 3.3 70B via Together AI for intelligent clause understanding
• 🌐 Multi-Domain Scanning - Discovers policies across root domain, subdomains, and product paths
• 🤹 Bot Evasion - User-agent rotation and request delays to avoid detection on restrictive sites
• 📱 Mobile/AMP Fallbacks - Automatically tries mobile and simplified versions for better coverage
• 🔍 Clause-Level Analysis - Extracts individual clauses with unique IDs and metadata
• ⚖️ Severity Scoring - Categorizes findings as High, Medium, or Low risk
• 🏷️ Category Tagging - Identifies arbitration waivers, data sale, tracking, COPPA violations, etc.
• 🔗 Clickable Verification - Source URLs with text fragments auto-scroll and highlight clauses in browser
• 📅 Policy Dating - Automatically extracts "last updated" dates from documents
• 💾 Local Storage - SQLite database for caching results (no cloud required)
• ✨ Beautiful Output - Rich terminal formatting with colors and tables
• 🔄 Fluent API - Chain commands with dot-notation for powerful queries

Installation

pip install policyboom

Setting Up AI Extraction

PolicyBoom uses Meta Llama 3.3 70B for intelligent clause extraction via Together AI's free tier:

1. Get your free API key from https://api.together.ai/settings/api-keys
2. Set the environment variable:

   export TOGETHER_API_KEY="your-api-key-here"
   

3. Run PolicyBoom - it will automatically use AI extraction when the API key is present

Fallback: If no API key is set, PolicyBoom uses regex-based extraction (less accurate but works offline).

Usage

CLI with Dot-Notation

The primary interface uses a fluent camelCase API:

# Basic scan
policyboom exec "scan('company.com').summarizeHigh()"

# Filter by severity and category
policyboom exec "scan('slack.com').summarizeMedium().category('dataSharing')"

# Get metadata with policy URLs and dates
policyboom exec "scan('stripe.com').summarizeAll().metadata()"

# Get full evidence for legal documentation
policyboom exec "scan('example.com').summarizeHigh().withEvidence()"

# Get detailed findings with all metadata
policyboom exec "scan('slack.com').summarizeAll().detailed()"

As a Python Library

from policyboom import scan

# Scan and analyze
result = scan("slack.com").summarizeHigh().category("arbitration")

# Access findings
for finding in result.findings:
    print(f"{finding.severity}: {finding.text}")
    
# Export results
result.export("output.json", format="json")

Categories

PolicyBoom identifies these concerning clause types:

• arbitration - Forced arbitration / class action waivers
• dataSale - Third-party data selling
• tracking - Advertising and behavioral tracking
• location - Location data collection
• retention - Data retention policies
• childrenData - Children's data handling (COPPA)

Reliability & Bot Evasion

PolicyBoom is designed to successfully scan most websites while avoiding bot detection:

User-Agent Rotation

Every HTTP request uses a random, realistic browser user-agent from a pool of:

• Chrome on Windows, macOS, Linux
• Firefox on Windows, macOS, Linux
• Safari on macOS and iOS
• Microsoft Edge on Windows

This makes requests indistinguishable from normal web browsing.

Human-Like Behavior

• Request Delays: Random 1-3 second pauses between requests mimic human browsing patterns
• Browser Headers: Complete header sets including Accept, Accept-Language, Accept-Encoding

Mobile & AMP Fallbacks

When standard URLs fail or return incomplete content, PolicyBoom automatically tries:

• Mobile sites (m.domain.com)
• AMP versions (?amp=1)
• Simplified/print views (?print=true)

This provides ~60-70% success rate across major websites, including sites with basic bot detection.

Limitations

Some sites still require headless browsers or authentication:

• Amazon, Kick.com: Advanced bot detection systems
• Facebook, LinkedIn: Policies behind login walls
• Some news sites: JavaScript-heavy rendering

For these sites, consider using headless browser support (adds ~10x latency and +700% memory/CPU cost).

Commands

# Get help
policyboom --help

# Run interactive guide
policyboom guide

# View examples
policyboom examples

# Export scan results
policyboom export <scan_id> --format json

Development

# Clone repository
git clone https://github.com/policyboom/policyboom
cd policyboom

# Install in development mode
pip install -e ".[dev]"

# Run tests
pytest

License

MIT License - see LICENSE file for details

Support

• Documentation: https://policyboom.dev
• Issues: https://github.com/policyboom/policyboom/issues
• Discussions: https://github.com/policyboom/policyboom/discussions