DigUp searches files on a Mac by what is in them
DigUp is a free Mac app that lets you describe what is in a photo, video, recording or document on your own machine, and finds the file; it runs Google DeepMind's EmbeddingGemma 2 locally, picked up 108 stars in two days and reads more than 100 languages.
Image: GitHub
Why it mattersA finder's search box matches file names and a line or two of text. A model that embeds every page, frame and 30 seconds of audio into one shared space lets a non-technical reader find the lease clause, the podcast moment, and the zebra in the video from one short description.
A Mac user who remembers that a zebra turns up somewhere in a video, or that the pet clause is on one page of a long lease, has no finder command to open it. DigUp, a free Mac app the developer ARahim3 released on 8 October 2026, closes that gap. It reads the pictures, PDFs, documents, recordings and videos in folders of your choice, puts their contents in Google DeepMind's EmbeddingGemma 2, and lets you find a match by typing what you remember. The repository picked up 108 stars in two days under the MIT licence.
EmbeddingGemma 2, released by Google DeepMind on 8 October 2026, puts text, images, audio and video into one shared 768-dimensional space. The DigUp README says this is why a search for "a dog on the beach" finds the photo and a search for "where they talk about sleep" finds that minute of a podcast. The model has 740 million parameters, is open under Apache 2.0, and understands more than 100 languages. DigUp runs the 8-bit build through llama.cpp on the Mac's GPU, in a helper process that quits when the pass is done, and keeps every vector in one SQLite file.
What a search looks like
A press of Shift-Command-Space opens a search bar in any app. The README lists real examples: "zebra in a video" opens the clip at the moment the zebra shows up, "the clause about pets in the lease" shows the PDF page with the words marked, "payment declined error" finds the screenshot. Exact strings also match, so an invoice number or an error code finds the file it is written on. English words find notes written in Bengali or Arabic, and the other way around.
What DigUp reads depends on the folder. For pictures and screenshots it looks at what they show and the text inside screenshots. For PDFs it takes a page at a time, scans included, up to 5,000 pages a file. For audio it listens in overlapping 30-second windows and points at the moment the match sits in. For video it looks at a frame every few seconds plus the soundtrack the same way. The first pass takes a few minutes for a few hundred files and up to an hour for a big Downloads folder; after that a new file is searchable a second or two after it lands.
What it does not read, and what leaves the Mac
The README names what DigUp refuses to touch: keys, certificates, password files, hidden folders like ~/.ssh, app bundles and caches. It skips folders that look like datasets, and never downloads iCloud files that live only in the cloud. Inside a code project (a folder with .git or package.json) it reads only screenshots unless the user turns code search on in Settings.
Nothing a user indexes leaves the Mac. DigUp goes online once to download the 865 MB model, and once a day to check for updates; the update check sends nothing about the files and can be turned off. There is no account and no telemetry. The app runs on Apple Silicon Macs with macOS 14 or later, is tested on macOS 26, and installs through a DMG from Releases or through a Homebrew tap.
Source
Primary source: ARahim3/DigUp on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.