ivankuria/tafuta

Tafuta is a macOS app written in Swift that lets you search through your personal video collection using plain-language descriptions. Instead of browsing files named things like "IMG_0423.mov" or hunting by date, you type something like "woman in a white blouse in a green meadow" and the app jumps directly to that specific moment in your video. It works entirely on your Mac, with no internet connection required after a one-time model download of about 108MB. The app uses Apple's MobileCLIP technology, which understands the visual content of video frames rather than just file names or metadata. When you add a folder of videos, Tafuta samples roughly one frame per second, converts each frame into a numerical fingerprint, and stores those fingerprints locally. When you search, your text query is converted into the same kind of fingerprint and compared against all stored frames to find the closest visual match. Results show exact timestamps inside your videos, not just a list of files. You click a result and playback starts right at that moment. The app also includes a global keyboard shortcut launcher so you can search from anywhere on your Mac without switching windows. The project is open source under the GPLv3 license. It requires macOS 14 or later on Apple Silicon hardware, which is the chip type Apple uses in newer Mac models. A prebuilt download is available on the releases page, signed and notarized by Apple. Building from source requires Xcode and a small utility called XcodeGen. The roadmap mentions future additions like searching spoken words in audio tracks, searching by image query, and an optional cloud tier for faster indexing of very large collections. It is designed for content creators, filmmakers, and anyone who shoots a lot of video and needs to find specific moments inside a large collection.