# DigUp: Search Local Files with Google DeepMind EmbeddingGemma 2 This repository profile is provided by osrepos.com, an open source repository discovery platform. Source: osrepos.com Repository profile: https://osrepos.com/repo/arahim3-digup Generated for open source discovery and AI-assisted research. DigUp searches photos, PDFs, documents, audio, video and optionally code by meaning, across macOS, Windows and Linux. Built on Google DeepMind’s EmbeddingGemma 2 and llama.cpp, it indexes and searches files on-device. GitHub: https://github.com/ARahim3/DigUp OSRepos URL: https://osrepos.com/repo/arahim3-digup ## Summary DigUp searches photos, PDFs, documents, audio, video and optionally code by meaning, across macOS, Windows and Linux. Built on Google DeepMind’s EmbeddingGemma 2 and llama.cpp, it indexes and searches files on-device. ## Topics - desktop-app - local-first - privacy - cross-platform - semantic-search - multimodal-search - file-search ## Repository Information Last analyzed by OSRepos: Wed Oct 14 2026 21:28:12 GMT+0100 (Western European Summer Time) Detail views: 5 GitHub clicks: 3 ## Safety Notice OSRepos shares public repositories for knowledge and discovery only. Review source code, dependencies, licenses, and security implications before running or installing anything. ## Content ## Overview DigUp is a cross-platform desktop file search app built on Google DeepMind’s EmbeddingGemma 2 and llama.cpp. It lets you describe what you remember from a file, rather than relying only on its name or exact text, and searches the selected folders on macOS, Windows and Linux. Its appeal is finding related content across file types: a query can match a photo, a passage in a PDF, a point in a recording or video, or code when code search is enabled. Indexing and search run on your computer, which suits people who want semantic search without sending their files to a hosted service. ## Key Features - Searches images, PDFs, documents, audio and video by their content, with results that can open at a relevant page or moment. - Uses EmbeddingGemma 2’s shared text, image, audio and video embedding space to search across different kinds of files. - Combines semantic matches with exact-word matching for terms such as names, invoice numbers and error codes. - Supports cross-language search across the model’s supported languages. - Offers optional, separate code search for selected project folders, with results that can open in an editor at the matching line. - Keeps its index and model on-device; the README says files are not uploaded and there is no telemetry. - Provides a global shortcut and menu-bar or system-tray access on macOS, Windows and Linux. ## Use Cases - Find a photo or screenshot by describing what it shows, even when its filename is unhelpful. - Locate a statement in a lease or other long PDF and open the result at the relevant page. - Jump to a particular topic or event in a podcast or video without manually searching through the recording. - Search a local codebase by describing a function’s behavior, or by entering its name, when code search is enabled. - Find files across languages or formats without uploading a work or personal archive to a search service. ## Our Take DigUp is a promising, actively changing local search app, but its newness and dependence on a single contributor make it worth evaluating before relying on it for essential workflows. **Good fit if:** - You want private, on-device semantic search across media and documents on macOS, Windows or Linux. - You are comfortable trying a beta on Windows or Linux, or a fast-moving new project on any supported platform. - You want optional code search alongside ordinary file search, with an MIT-licensed project you can inspect or build. **Look elsewhere if:** - You need a mature project with a broader contributor base; the measured activity is concentrated entirely in one contributor. - You need text recognition in screenshots on Windows or Linux, or need older Word documents, RAW photos or SVG files indexed on those platforms. - You need a hosted, centrally managed search service rather than an index maintained on each computer. ## Project Health | Signal | Value | |---|---| | Status | **New**: less than six months old and changing fast | | Last commit | 2026-10-14 (today) | | Commits, last 90 days | 100+ | | Releases, last 12 months | 3 (latest v0.7.0-beta.1, 2026-10-14, pre-release) | | Contributors | 1 | | Issues closed, last 90 days | 8 (typically closed in 2 days) | | Pull requests merged, last 90 days | 0 | | Project age | 5 days | Checked on 2026-10-14 with the GitHub API. ## Project Facts - Language: Swift - License: MIT - Stars: 3.5k - Forks: 235 - Topics: embeddinggemma, embeddings, file-search, linux, llama-cpp, local-first, macos, menubar-app - Archived: no ## What You Need Detected in the repository: - Swift (main language on GitHub) - A test suite and automated checks on GitHub Actions ## Getting Started On a Mac, install with Homebrew: ```sh brew install --cask arahim3/tap/digup ``` Windows and Linux builds are marked beta in the README. See the [README](https://github.com/ARahim3/DigUp#readme) for downloads, platform-specific setup and build instructions. ## License in Plain Words **MIT** (permissive). - Commercial use: yes - Modify and redistribute: yes - You must keep: the copyright and license notice - Share your changes: no A summary, not legal advice: the LICENSE file is what applies. ## Alternatives - [lilbee](https://osrepos.com/repo/tobocop2-lilbee): lilbee searches files, code, and websites to provide cited answers through several interfaces, while DigUp focuses on semantic search across local files. | Project | Language | License | Stars | Status | |---|---|---|---|---| | **DigUp** | Swift | MIT | 3.5k | New | | [lilbee](https://osrepos.com/repo/tobocop2-lilbee) | Python | MIT | 62 | Active | ## Considerations - The project is very new and changing quickly. Recent activity includes frequent commits and three releases in the past year, with the latest tagged release marked as a beta. Measured issue handling shows issues being closed, but no merged pull requests in the last 90 days. - Development is concentrated in one contributor. That is a maintenance and continuity risk for teams depending on long-term support. - The first setup downloads an 865 MB model, and initial indexing can take substantial time for large collections. The README notes that CPU-only indexing on Windows and Linux can be slow for large photo libraries; a GPU is used when available and found to be faster, but is not presented as a requirement. - Platform support is not identical. Screenshot text recognition is currently limited to macOS, as are some older document and image formats. Windows and Linux are described as beta. - The README says indexing is limited to selected folders and excludes cloud-only files, hidden or sensitive-looking folders, and several generated or dataset-like files. Check its file-handling rules against your needs before indexing a work collection. - DigUp is MIT-licensed. The README also identifies third-party components with their own licenses, including EmbeddingGemma 2 under Apache 2.0 and FFmpeg under LGPL; review bundled license notices when redistributing packages.