DigUp

DigUp is a MIT-licensed desktop search application that indexes selected local folders with Google DeepMind's EmbeddingGemma 2. Instead of relying only on filenames and keywords, it can search a photo, PDF page, recording, video frame, screenshot, or source file by a natural-language description. The maintained project ships a Mac release and beta packages for Windows and Linux.

What makes it useful

• Cross-modal local search: One query can match document text, a picture, a video moment, or audio speech. Exact terms such as invoice numbers and error strings are also boosted.
• On-device indexing: The repository says indexed material stays on the computer. It goes online only to fetch the model and check updates, with the update check configurable.
• Actionable results: It opens PDFs at a page, recordings and videos at a matching moment where supported, and can send code results to configured editors.
• Scoped code search: Code indexing is opt-in and excludes ignored, generated, vendored, hidden, and key-like files according to the documented safeguards.

A practical workflow

Install from the official release, choose one non-sensitive folder, and let the first index complete before judging relevance. Try a small query set: a visual description, a phrase from a document, and an exact identifier. Enable code search only for a disposable repository first, then check which files were indexed and how results open in your editor. The project documents a roughly 865 MB initial download for the model components, so plan for disk space and a potentially long first pass on large libraries.

Limits and alternatives

Semantic results are not a substitute for backup, access control, or conventional file search. Large image and video libraries can take hours to index on CPU-only systems. The project documents platform-specific gaps, including limited screenshot text recognition outside macOS. EmbeddingGemma is the underlying model entry; NotebookLM is a hosted alternative for curated source collections rather than a whole-device local index.

Getting started

Read the official README, verify the release for your platform, and start with folders whose contents you are comfortable indexing locally.