Initial release.
Features
- Automatic pet tagging in Immich using CLIP embeddings (ViT-B-16, openai weights).
- Web enrollment UI: add pets, search your Immich library, assign reference photos.
- Logistic regression classifier trained locally on ref vs. negative embeddings.
- Poller runs every 5 minutes, classifies new photos, writes face entries to Immich.
- Negative samples ("not my pets") to improve rejection of non-pet photos.
- Clickable ref thumbnails linking back to Immich.
- Scan status panel showing last run time and per-pet stats.
- DRY_RUN mode for testing without writing to Immich.
- Configurable confidence threshold, poll interval, and CLIP model via environment variables.