Rummage searches files by meaning on Windows and Android via EmbeddingGemma 2

⚡ The gist in 5 seconds - Rummage has been released on GitHub — a local multimodal semantic file search tool built on EmbeddingGemma 2 from Google DeepMind. - Users choose which folders to search themselves: documents, photos, screenshots, audio recordings, videos, plus an image-based search mode. - Platform limitation — only Windows and Android; installation and setup are done from the CookSleep/Rummage repository. ### 🔍 What was found The CookSleep/Rummage repository hosts a tool for local semantic file search built on the EmbeddingGemma 2 embedding model from Google DeepMind. Supported platforms are Windows and Android. All processing happens locally: files are not sent to the cloud, and the search is limited to the folders the user has specified themselves. The logic revolves around describing content rather than file names: a single phrase about what's inside is enough, and Rummage finds matches among documents, photos, screenshots, audio recordings, and videos in the selected directories. A separate image-based search mode is also announced — "image by image." Technically, this is embedding-based search: file contents are converted into vector representations by the EmbeddingGemma 2 model, and a text or visual query is matched against them by semantic proximity rather than literal string matching. ### 💡 Why it matters EmbeddingGemma is an open family of embedding models from Google DeepMind, and the emergence of practical wrappers around it shows where local AI is heading: not chatbots, but utilitarian things like searching your own drive, where privacy is solved architecturally rather than through promises. For a geek, it's an alternative to cloud-based "ask AI about your files" services and the OS's built-in search, which clings to names and extensions. The project's maturity, license, and timeline for stable builds are not specified in the source — for now, judgments can only be based on the repository description.