Spotlight, but it actually knows what's in your files

What it does DigUp is a menu-bar Mac app that indexes folders you pick — images, PDFs, documents, audio, video, and optionally code — and lets you search them by meaning rather than filename. Ask for “the clause about pets in the lease” and it opens the PDF at that page; ask for “where they talk about sleep” and it jumps to that minute of the recording. It runs Google DeepMind’s EmbeddingGemma 2 (740M parameters) locally via llama.cpp on Metal, storing everything in a single SQLite file.

The interesting bit All content types land in the same 768-dimensional embedding space, so one query can rank a video frame against a PDF page against a photo. Exact-word matches get a boost alongside the semantic ones, because no embedding holds an invoice number precisely — a pragmatic hybrid most semantic search tools skip.

Key highlights

• Fully on-device: nothing indexed leaves your Mac; the only network calls are the one-time model download and an optional daily update check
• Cross-lingual search — an English query can find a note in Bengali or Arabic (100+ languages)
• Code search is opt-in, skips gitignored/vendored files and anything key-like, and opens results at the line in your editor
• Careful about what it touches: never reads ~/.ssh, app bundles, caches, or cloud-only iCloud files
• New files become searchable a second or two after landing; the model runs in a helper process that quits when idle

Caveats

• Apple Silicon only, macOS 14+; the first indexing pass can take an hour on a big Downloads folder
• Exact words inside screenshots depend on Apple’s text recognition (~25 languages), and unspaced scripts like Chinese and Japanese aren’t split into words
• The model isn’t equally strong in every language, per Google’s own notes

Verdict If your Mac is a decade of untagged photos, recordings, and PDFs, this is the search tool you keep wishing Spotlight were. If you’re on Intel or Windows, or your files are all neatly named, move along.