arahim3/digup is an on-device app that searches photos, documents, recordings, videos, code and other files on your computer by their content. The latest release shown on the GitHub API is v0.6.1, and the provided materials do not list what changed in this release separately. The README presents Mac as the official distribution and points Windows and Linux users to v0.7.0-beta.1. Release versions and per-platform guidance therefore need to be read as separate things.
Search uses Google DeepMind's EmbeddingGemma 2. Text, images, audio and video are converted into embeddings — comparable numeric representations — and placed in a single search space, so you find things with descriptions that do not specify a file type. The README gives, for example, searches for a zebra in a video or for people talking about water surfaces as examples. The model runs via llama.cpp, and it says it also supports cross-lingual search in more than 100 languages and exact string search. The model is Apache-2.0 licensed, and the app repository's license is MIT. The README notes that indexed content never leaves the device, and that it only connects for model download and a once-a-day update check.
This approach focuses on finding across multiple file formats inside a personal device. However, the README has no benchmark table of its own, and the only languages covered by the included evaluation files are English, Vietnamese and Arabic. So, separately from the claim of support for more than 100 languages, the evidence showing actual per-language search quality is limited.