This project handles highly sensitive personal data (a full export of someone's digital life). Privacy is a design constraint, not a feature.

• No telemetry, no analytics, no phone-home. The tools read files from disk and write files to disk.
• No network calls in the core pipeline (fireconverter.py, book_compiler.py, reclaim_indexer.py, reclaim_search.py, and reclaim_answer.py in keyword mode).
• One exception, clearly marked: reclaim_semantic.py (and --mode semantic) downloads an embedding model from the model host on first run. After that it works offline. If you need strict air-gapping, pre-download the model or avoid the semantic layer.

• The .gitignore blocks committing *.db, *.jsonl, *.docx, and Takeout directories so you don't accidentally publish your own data. Do not force-add these.
• Prefer running on a machine you control (your own computer or NAS).

Open a private security advisory on the repository, or email support@quietfireAI.com. Please do not file public issues for anything that could expose user data.