privacy-policy

User Guide

This guide explains how to install, run, and interpret results from Privacy Policy Analyzer. The tool is currently CLI-first (no stable class-based API).

📋 Table of Contents

Installation
Basic Usage
Configuration
Analysis Methods
Understanding Results
Advanced Features
Troubleshooting
Examples
Getting Help

🚀 Installation

Prerequisites

Python 3.13+
An OpenAI API key available in your environment (or .env)
Google Chrome (required only if you plan to use Selenium fallback)
Optional: trafilatura for enhanced extraction

Using uv (recommended)

git clone https://github.com/HappyHackingSpace/privacy-policy.git
cd privacy-policy
uv sync
# optional: activate .venv if you prefer a shell-activated workflow
# macOS/Linux: source .venv/bin/activate
# Windows: .venv\Scripts\activate

Using Poetry

git clone https://github.com/HappyHackingSpace/privacy-policy.git
cd privacy-policy
poetry install
# run commands with: poetry run <command>

Using pip

git clone https://github.com/HappyHackingSpace/privacy-policy.git
cd privacy-policy
python -m venv .venv
# macOS/Linux:
source .venv/bin/activate
# Windows:
.venv\Scripts\activate
pip install -e .

Create a .env file if desired:

Verify that your API key is visible to the process:

python -c "import os; print('API key set:', bool(os.getenv('OPENAI_API_KEY')))"

🔧 Basic Usage

Run the CLI with a site URL (auto-discovery will try to resolve a likely privacy policy page):

# uv
uv run python src/main.py --url https://example.com
# or module form
python -m src.main --url https://example.com

Analyze a known policy URL directly (skip auto-discovery):

uv run python src/main.py --url https://example.com/privacy-policy --no-discover

Choose a fetch method:

uv run python src/main.py --url https://example.com/privacy --fetch selenium

Control output detail:

uv run python src/main.py --url https://example.com --report detailed

⚙️ Configuration

Environment Variables

You can configure the model and environment via variables or flags.

Required
- OPENAI_API_KEY: your OpenAI key
Optional
- OPENAI_MODEL: default model if --model is not provided (defaults to gpt-4o)

CLI flags that control analysis:

--model TEXT
Override the OpenAI model for this run.
--report {summary|detailed|full}
Select output verbosity.
- summary: overall score, confidence, top strengths/risks, red-flag count
- detailed: includes per-category details, red flags, recommendations
- full: includes raw per-chunk results
--chunk-size INT, --chunk-overlap INT, --max-chunks INT
Tune chunking for very long policies (tail chunks may be merged when --max-chunks is exceeded).
--fetch {auto|http|selenium}
Extraction mode (auto uses HTTP first and can fall back to Selenium).
--no-discover
Analyze the provided URL as-is (no policy URL resolution).

🔍 Analysis Methods

URL analysis with auto-discovery
Provide a site homepage or any page; the tool attempts to find a likely policy path (e.g., /privacy, /privacy-policy, robots/sitemap hints, or in-page links).
Direct policy URL
If you already know the exact policy page, use --no-discover to skip discovery.
Extraction
Content is fetched via HTTP/BeautifulSoup by default, optionally through Trafilatura when available, and can fall back to Selenium for client-rendered pages.
Chunking & scoring
Extracted text is split into overlapping chunks; each chunk is scored by the model with a fixed schema. Category scores (each 0–10) are aggregated with weights into a 0–100 overall score.

📊 Understanding Results

The CLI prints JSON to stdout.

Common fields:

status: "ok" or "error"
url: the input URL you provided
resolved_url: the discovered/verified policy URL (if discovery was used)
model: OpenAI model used (e.g., gpt-4o)
chunks / valid_chunks: number of chunks analyzed and number that produced valid scores
overall_score: weighted 0–100 score across all categories
confidence: coverage ratio (0–1), based on how many categories received valid scores
category_scores: per-category {score (0–10), weight, rationale}
top_strengths / top_risks: strongest/weakest categories
red_flags: unique risk indicators extracted from chunk results
recommendations: short, actionable notes
(full only) chunks: raw per-chunk outputs

Dimensions (what they mean)

Each dimension is scored 0–10; weights are applied to compute the overall score:

Lawful Basis & Purpose: Whether the policy explains clear purposes for processing and, where relevant, the legal basis or justification.
Collection & Minimization: How clearly the policy describes the types of data collected and whether collection is limited to what is necessary.
Secondary Use & Limits: Whether the policy restricts or explains additional uses beyond the original purpose.
Retention & Deletion: Clarity on how long data is kept, deletion practices, or criteria for determining retention.
Third Parties & Processors: Disclosure of processors, vendors, or third parties with whom data is shared, and their roles.
Cross-Border Transfers: Information on transfers outside the user’s country/region and safeguards in place.
User Rights & Redress: How users can exercise rights such as access, correction, deletion, or complaint, and available escalation channels.
Security & Breach: Security measures described and any statements about breach notification or handling.
Transparency & Notice: Overall clarity, structure, contact details, and how users are informed of updates or changes.
Sensitive Data, Children, Ads & Profiling: How sensitive categories are handled, rules for children’s data, use of data for advertising, and automated decision-making/profiling.

🚀 Advanced Features

Flexible fetching with --fetch auto|http|selenium
Configurable chunking via --chunk-size, --chunk-overlap, --max-chunks
Report levels with --report summary|detailed|full
Model override with --model or OPENAI_MODEL

Note: Caching, batch processing, CSV/HTML exports, and a stable importable Python API are not part of the current CLI release. See the roadmap in the contributing guide.

🔧 Troubleshooting

OPENAI_API_KEY is not set
Set the key in your environment or a .env file.
Empty or insufficient text
Allow auto-discovery (avoid --no-discover) or provide a better URL. Some pages may require Selenium (--fetch selenium) to render content.
Very long policies
Increase --max-chunks, or adjust --chunk-size/--chunk-overlap. The tool merges tail chunks to stay within limits.
Model issues
Ensure the selected model supports JSON-style responses. The tool uses temperature=0 for consistent scoring.

📚 Examples

Analyze a homepage with default settings:

uv run python src/main.py --url https://example.com

Analyze a known policy URL with detailed output:

uv run python src/main.py --url https://example.com/privacy-policy --no-discover --report detailed

Force Selenium for a client-rendered policy:

uv run python src/main.py --url https://example.com/privacy --fetch selenium

Tune chunking for very long policies:

uv run python src/main.py --url https://example.com --chunk-size 3000 --chunk-overlap 300 --max-chunks 25

🆘 Getting Help

Documentation: Check the API Reference for detailed API documentation
Issues: Report bugs on GitHub Issues
Discussions: Join community discussions on GitHub Discussions
Discord: Join our Happy Hacking Space Discord

Happy analyzing! 🔍✨

This site is open source. Improve this page.