This project currently exposes a command-line interface (CLI) rather than a stable importable class-based API.
Use the CLI to discover, extract, and evaluate privacy policies, and consume the JSON it prints to stdout.
# Using uv
uv run python src/main.py [OPTIONS]
# Or module form
python -m src.main [OPTIONS]
Required: --url (site URL for auto-discovery or a direct privacy policy URL)
--url TEXT
Input URL. If it’s a homepage or non-policy page, the tool will try to resolve a likely policy URL.
--model TEXT
Override the OpenAI model (defaults to OPENAI_MODEL or gpt-4o).
--chunk-size INT (default: 3500)
Character-based chunk size for splitting long policies.
--chunk-overlap INT (default: 350)
Overlap between chunks.
--max-chunks INT (default: 30)
Hard cap for analyzed chunks; remaining tail chunks are merged.
--report {summary|detailed|full} (default: summary)
Output verbosity level.
--fetch {auto|http|selenium} (default: auto)
Extraction method. auto tries HTTP first and can fall back to Selenium.
--no-discover
Analyze the given URL as-is (skip auto-discovery).
The CLI prints JSON to stdout.
summary{
"status": "ok",
"url": "https://example.com",
"resolved_url": "https://example.com/privacy",
"model": "gpt-4o",
"chunks": 12,
"valid_chunks": 11,
"overall_score": 82.5,
"confidence": 0.9,
"top_strengths": [["user_rights_and_redress", 8.7], ["security_and_breach", 8.2], ["transparency_and_notice", 7.9]],
"top_risks": [["cross_border_transfers", 5.1], ["retention_and_deletion", 6.0], ["secondary_use_and_limits", 6.2]],
"red_flags_count": 2
}
detailedAdds:
category_scores: { [category]: { "score": number (0–10), "weight": number, "rationale": string } }red_flags: string[]recommendations: string[]fullAdds:
chunks: raw per-chunk model outputs (including per-chunk scores, rationales, and optional red_flags/notes)Each category is scored 0–10 per chunk by the model; scores are averaged and combined with the weights below to form the 0–100 overall score.
lawful_basis_and_purpose — weight 12
Clarity of purposes/justifications; purpose limitation; consent/choice clarity where relevant.
collection_and_minimization — weight 10
Specificity/necessity of collected data categories; proportionality/minimization.
secondary_use_and_limits — weight 8
Limits on additional/compatible uses; avoidance of vague blanket purposes.
retention_and_deletion — weight 8
Concrete periods or clear criteria; deletion/archiving language; avoiding indefinite retention without justification.
third_parties_and_processors — weight 12
Processors/third parties, categories & purposes, and role clarity (controller/processor/joint).
cross_border_transfers — weight 8
Destinations (if any) and plain-language safeguards for international transfers.
user_rights_and_redress — weight 14
How rights can be exercised (access/rectification/erasure/restriction/portability/objection), timelines, contacts, escalation paths.
security_and_breach — weight 12
Organizational/technical measures; breach handling language; security/DPO contact if applicable.
transparency_and_notice — weight 8
Plain language, structure, contact details, change/version notices, cookie/consent pointers where relevant.
sensitive_children_ads_profiling — weight 8
Handling of sensitive/special categories; children data statements/age gates; selling/sharing for ads and opt-out/limit choices; automated decisions/profiling.
OPENAI_API_KEY (required)OPENAI_MODEL (optional; default model if --model is not set)0: success (JSON printed)Analyze a homepage (auto-discovery):
uv run python src/main.py --url https://example.com
Analyze a known policy URL directly:
uv run python src/main.py --url https://example.com/privacy-policy --no-discover --report detailed
Force Selenium for a client-rendered page:
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