Privacy

Lineage handles some of the most personal material a family has. Here is exactly where it goes.

What stays on your computer

Audio. Speech recognition (WhisperX) and speaker labelling (pyannote) run locally. Your recordings are never uploaded by Lineage.
Originals. audio/ and photos/source/ are never modified; their checksums are recorded so any change is noticed.
The project folder. Transcripts, story units, chapters and PDFs live in your project folder. Nothing is published unless you publish it.

Transcript text goes to a cloud language model. The writing is done with Claude (Claude Code), so transcript passages, notes and drafts are sent to Anthropic while you work. If your recordings contain things that must not leave your machine, remove them from the transcript before you start writing, or don't use the writing stages for them.
Model downloads. The first transcription run downloads the speech and diarization models from Hugging Face. The diarization model is gated: you accept its licence once, with a free account and an access token. The token is only used to download the model; no audio is sent.
Research. If you ask Claude to look up public records (rosters, newspapers, museum catalogues), those requests go to the websites concerned. The skills tell Claude never to put your name, email or other personal details in those requests.

The book is about real people, many of them alive. The style guide makes the writer flag anything about living people, legal trouble or trauma with // REVIEW:. The writer never cuts or softens it on its own: you decide what prints.
Before you share a draft, read the // REVIEW: list (grep -rn "REVIEW" content chapters).

make new writes a .gitignore that keeps audio, photos, raw web caches and outputs out of git. If you put your project on GitHub, keep the repository private: transcripts are in it.
Never commit tokens. HF_TOKEN belongs in your shell environment, not in a file in the project.