DigUp
In one sentence
Describe what you remember about a file’s contents; DigUp ranks matches in a shared embedding space on your machine — no account, no cloud index of your files.
What it can search (modalities)
| Modality | What DigUp indexes | Open-to |
|---|---|---|
| Photos / images | Visual content (and OCR text on Mac for screenshots / scanned pages) | File |
| PDFs | Pages as meaning + text; scanned PDFs get OCR on Mac (0.6.1+) | Page |
| Documents | txt, md, rtf, docx, odt, html; .doc mainly Mac | File / passage |
| Audio | ~30-second windows in a shared vector space | Timestamp |
| Video | Frames / moments by description | Timestamp |
| Code (optional) | Enable in settings; query with code: prefix | Line |
Win/Linux beta (0.7.0-beta.1) does not yet OCR screenshot text; skips old .doc, RAW, and SVG on that beta — see Windows / Linux guides.
How it works (high level)
- You choose folders — DigUp does not auto-index the whole disk; cloud placeholders (undownloaded iCloud/OneDrive) are skipped.
- Extract → embed — content becomes images/text/audio chunks; EmbeddingGemma 2 runs locally via llama.cpp (~865 MB first download).
- SQLite index — vectors plus keyword boost for exact strings (invoice numbers, error codes, names).
- Search — query text is embedded; nearest neighbors + keyword weights; hotkey opens the panel from any app.
More on the model: How DigUp uses EmbeddingGemma 2.
Requirements by platform
| Platform | Version (2026-10-15) | Hardware / OS | Status |
|---|---|---|---|
| Mac | 0.6.1 | Apple Silicon, macOS 14+ | Stable |
| Windows | 0.7.0-beta.1 | 64-bit Windows 10/11 | Beta |
| Linux | 0.7.0-beta.1 | amd64; .deb or AppImage | Beta |
GitHub stars ~3,716 as of 2026-10-15 — re-check the repo for live counts.
What DigUp is not
- Not a full Spotlight / Windows Search replacement (it can sit beside them).
- Not an app launcher (as of research date).
- Not a cloud sync or collaboration search product.
- Download DigUp from the home download section.