privacy-gate-llm
Copyright 2026 Yauhen Bichel

This product includes software developed as part of the MoleCare project
(https://www.molecare.co.uk/).

Licensed under the Apache License, Version 2.0. See LICENSE.

--------------------------------------------------------------------------------
Third-party components
--------------------------------------------------------------------------------

This repository contains no vendored third-party code and no Python
dependencies beyond the standard library.

At run time it calls an Ollama endpoint serving an embedding model. The default
is BAAI/bge-m3 (https://huggingface.co/BAAI/bge-m3), which is distributed under
the MIT Licence by the Beijing Academy of Artificial Intelligence. That model is
neither included in nor redistributed by this repository; only the 1024-weight
logistic head in model/ is ours.

--------------------------------------------------------------------------------
Test fixtures
--------------------------------------------------------------------------------

data/gold.jsonl and data/fresh.jsonl contain strings that are deliberately
shaped like credentials and personal data. All of them are invented. Two are
published placeholders owned by others and are reproduced here for the purpose
they were published for:

  AKIAIOSFODNN7EXAMPLE
      Amazon Web Services' documentation placeholder access key ID.

  07700 900xxx
      Ofcom's reserved drama range for UK telephone numbers, which is never
      allocated to a subscriber.

No real person, patient, credential, or organisation appears in this repository.
See SECURITY.md.
