Offline, deterministic CLI that turns an OpenAI privacy export into an auditable interaction profile (user.json) plus an explanation report. No network, no LLMs, no embeddings, no telemetry, no probabilistic inference.
python3 -m venv .venv
. .venv/bin/activate
pip install -e .
upfe analyze --input /path/to/export_zip_or_dir --output ./out
Outputs user.json and profile_report.md (deterministic; no timestamps).
Tests are written first to enforce airgap, determinism, typos/multilingual handling, and false-positive controls.
python3 -m unittest discover -s tests
src/upfe/core library (ingest, normalize, features, synthesize, report, validate, cli).src/upfe/lexicons/JSON lexicons for compression, pushback_tolerance, tone, formats.tests/fixtures/sample exports (no personal data) for coverage.
- No network/telemetry/LLMs/embeddings/probabilistic models.
- Deterministic output for identical input.
- Explanations live in
profile_report.md;user.jsonis minimal per schema.