Tagasaurus is a privacy-first media-manager that lives entirely on your desktop. Drop photos, videos, GIFs, PDFs, or audio files into the app and Tagasaurus builds a fully local library you can search like magic. Type a natural-language description ('sunset over the vacation 2019', 'college picnic 2012') or let the camera roll: Tagasaurus can find every shot of your old friend Jack by comparing face embeddings even if you never wrote a single tag. Because everything runs offline, none of your memories ever leave your machine. It can handle storing audio, pdfs, gifs and video.
https://github.com/mantzaris/Tagasaurus/
Under the hood the app is a tight Electron JS bundle with a modern Svelte 5 + Routify frontend. The renderer is written in TypeScript with runes-based reactivity and Tailwind-style utility classes, while the main process handles heavy lifting in Node. Media ingest pipelines lean on fluent-ffmpeg (pointed at statically bundled ffmpeg/ffprobe binaries) to extract thumbnails, duration metadata, and frames for face detection. Faces are embedded with an ArcFace ONNX model; text search uses a lightweight sentence-embedding model—both executed through onnxruntime for fast, CPU-friendly inference. Vectors, file hashes, and canonical paths land in a libSQL (SQLite) database, giving you ACID guarantees and blistering full-text + nearest-neighbour queries without shipping a server.
Tag Your Planet!
Top comments (0)