---

Watcher is a personal movie & TV show tracker with local AI-powered recommendations. It uses Tauri 2 (Rust + vanilla web frontend) and runs entirely offline after initial setup — the only network dependency is fetching metadata from TMDB and optionally downloading the recommendation model.

Built on a whim for personal use. This space is already crowded with excellent apps, so this is just my take — contributions are welcome if you'd like to spend more time on it.

Features

• Track plays — log when and where you watched a movie or TV episode, with optional star ratings
• TMDB metadata — automatic poster, synopsis, credits, keywords, and technical details
• Watchlist — maintain a want-to-watch list for movies and TV series
• Import / Export — full JSON import/export for portability
• Local AI recommendations — describes what you're in the mood for in plain text; a local BERT model ranks TMDB candidates by semantic similarity to your prompt
• Favourite directors & actors — boosts recommendations for creators you rate highly
• Dashboard — quick stats on total plays, unique titles, favourite media, and top places

Screenshots

Coming soon

Getting Started

Prerequisites

• Rust toolchain
• A free TMDB API key (set via TMDB_API_KEY environment variable or at build time)

Build & Run

# Run in development mode
TMDB_API_KEY=your_key_here cargo run --manifest-path src-tauri/Cargo.toml

# Build a release binary
cargo build --release --manifest-path src-tauri/Cargo.toml

# Build a .deb package
npx tauri build --bundles deb

# Build an AppImage
npx tauri build --bundles appimage

Alternatively, use the top-level Makefile:

make deb   # TMDB_API_KEY=your_key_here make deb
make run   # TMDB_API_KEY=your_key_here make run

The first launch will download the all-MiniLM-L6-v2 sentence-transformer model (~90 MB) from HuggingFace Hub. Subsequent launches use the cached copy.

Status

Work in progress — not yet ready for everyday use. Things that are missing or rough:

• The frontend is vanilla HTML/CSS/JS with no reactive framework — functional but minimal
• No packaged distribution beyond .deb / AppImage
• The recommendation pipeline is naive: TMDB search + local BERT similarity. Quality varies.
• No push notifications or background sync
• Linux-only at the moment (tested on Ubuntu)

Built With

• Tauri 2 — desktop application framework
• TMDB API — movie & TV metadata
• Candle — local BERT inference (no GPU required)
• sentence-transformers/all-MiniLM-L6-v2 — embedding model
• tokenizers — HuggingFace tokenizer
• rusqlite — SQLite bindings
• hf-hub — model downloading from HuggingFace

Credits

• Author / Maintainer: BlancoBAM
• All movie and TV metadata is provided by The Movie Database (TMDB). This product uses the TMDB API but is not endorsed or certified by TMDB.
• The recommendation model is sentence-transformers/all-MiniLM-L6-v2 from HuggingFace, running locally via Candle.

License

MIT — see LICENSE for details.

Watcher