DigUp Windows

Find anything on Windows by what it means, not only what it's called.

A native Rust desktop search tool for filenames, documents, OCR and optional local semantic search. Exact search and extraction run entirely on your machine; semantic search uses an embedding server that you configure.

_{The project page, with the search modes, the pipeline and the privacy boundary.}

DigUp is a Windows preview. Exact search and document extraction run entirely on your machine. Semantic search is optional and uses an embedding server that you configure.

Contents

1. How it works
2. Features
3. Install
4. Optional components
5. Build from source
6. Scripts
7. Privacy boundary
8. Documentation
9. Contributing
10. Acknowledgements
11. License

How it works

flowchart LR
  A[Filesystem watcher] --> B[Exact index]
  A --> C[OCR and media extraction]
  C --> D[Local embeddings]
  B --> E[Ranking]
  D --> E
  E --> F[Preview and open]
  D -. unavailable .-> B

Loading

If the embedding server is unavailable, DigUp falls back to exact search.

Features

• Case-insensitive, regex, wildcard and folder search
• TXT, Markdown, RST, PDF, DOCX and ODT extraction
• Optional OCR and multimedia indexing
• Native folder watching and result previews
• Opens PDFs at the matching page and media at the matching timestamp (VLC)
• Windows per-user install and uninstall scripts that preserve your settings and data

For semantic search, use LM Studio or an OpenAI-compatible llama.cpp embedding server. For EmbeddingGemma 2, DigUp applies the model's SearchQuery prompt to normal queries, CodeRetrieval to source search, and titled title: … | text: … inputs to documents. The configured server receives the request; use localhost to keep inference local.

Install

1. Download DigUpWindows-<version>-windows-x64-preview.zip from the latest release. Check it against SHA256SUMS.txt:

   (Get-FileHash .\DigUpWindows-*-windows-x64-preview.zip -Algorithm SHA256).Hash

2. Extract the zip, then in PowerShell from the extracted folder:

   ./check-prerequisites.ps1   # verifies x64 Windows and the VC++ runtime
   ./install.ps1               # per-user install, add -DesktopShortcut for a shortcut

To remove it, quit DigUp from its tray menu and run uninstall.ps1. Settings, vectors and models are preserved.

The preview is not code-signed, so Windows SmartScreen may warn. Verify the checksum, then choose More info → Run anyway.

Optional components

Exact search needs nothing else. Add features as you need them:

Feature	How to enable
OCR (images, scanned)	./install-dependencies.ps1 -Ocr (Tesseract, Poppler)
Audio/video indexing	./install-dependencies.ps1 -Media (FFmpeg)
Timestamp seeking	./install-dependencies.ps1 -VideoSeeking (VLC)
Local semantic search	./setup-llama.ps1, then ./start-llama.ps1

Tool locations can be overridden with DIGUP_TESSERACT, DIGUP_PDFTOPPM and DIGUP_VLC.

Build from source

Requires Windows 10/11 x64, Rust (stable), Visual Studio C++ Build Tools and the Windows SDK.

git clone https://github.com/boss477/digup-windows.git
cd digup-windows
./build.ps1      # fmt check, tests, clippy (warnings are errors), release build
./run.ps1

./build.ps1 -TestsOnly runs the checks without the release build.

Scripts

Script	Purpose
build.ps1, run.ps1	Build with all checks; run the release build
package.ps1, fetch-upstream-notices.ps1, collect-notices.ps1	Create the zip with third-party license notices
install.ps1, uninstall.ps1	Per-user install and removal
verify-package.ps1, verify-install.ps1, verify-clean-install.ps1	Validate the package and installer behavior
verify-gui.ps1, verify-ocr.ps1, verify-media.ps1, verify-gpu.ps1	Headless and hardware validation

For packaging, run ./fetch-upstream-notices.ps1 once, then ./package.ps1.

Releasing

Bump version in Cargo.toml, move the changelog's Unreleased entries under it, then tag:

git tag v0.1.3
git push --tags

The release workflow runs the full validation, then publishes the zip and SHA256SUMS.txt to a GitHub Release.

Privacy boundary

Exact indexing and extraction run locally. Semantic and multimedia search send content to the provider you configure; source files may contain secrets and .gitignore rules are not applied. Use a localhost endpoint to keep everything on your machine. See SECURITY.md.

Documentation

• Product design
• Security policy
• Changelog

Contributing

Issues and pull requests are welcome. Read CONTRIBUTING.md and the Code of Conduct first.

Acknowledgements

DigUp Windows is inspired by ARahim3/DigUp by Abdur Rahim, the cross-platform on-device search app built on Google DeepMind's EmbeddingGemma 2. Thanks for the idea and the design. This project is an independent Windows-focused Rust implementation.

License

MIT © boss477. Third-party notices ship in each release package.