I reorganized 1,388 items in my password manager using AI, without exposing secrets to the model.
Keeping the secrets local was the condition for doing it at all.
The basic pattern was:
sanitize metadata -> classify with AI -> review uncertain cases -> rehydrate by stable item IDThe model never needed the passwords. It did not need TOTP secrets, recovery codes, notes, or anything sensitive. For classification, metadata was enough: item titles, URLs, rough category hints, and stable IDs.
The stable ID is important. AI can suggest that github.com belongs under Development or that a bank login belongs under Finance. I do not want it producing the final vault. Deterministic code maps each reviewed classification back to the original item by ID.
The separation was:
- split data first,
- send only sanitized metadata for classification,
- keep secrets local,
- review uncertain results,
- merge back deterministically.
Uncertain classifications dropped from 891 to 485, a 45.6% reduction. I could inspect every suggestion before anything touched the real data.
This is the pattern I trust for sensitive AI work: AI handles the semantic mapping, and deterministic code handles the final writes. The model never gets control of the real vault.