DigUp is a free desktop search app that finds files on your own computer by what is inside them rather than by their names, so you can type a description of a photo, a PDF clause or a moment in a video and jump straight to it. It is built for ordinary computer owners with messy folders of screenshots, scans, recordings and downloads, and it doubles as a readable Swift and Tauri project for developers curious about running a multimodal embedding model locally.
What it does
You pick the folders you want searched. DigUp reads the pictures, PDFs, documents, recordings and videos inside them, then waits for you to describe what you are after in plain words. The examples the README and its demo clip put forward:
- "zebra in a video" opens the clip at the moment the zebra shows up
- "where they talk about sleep" jumps to that minute of a podcast
- "the clause about pets in the lease" shows the PDF page
- "a dog on the beach" finds the photo, and "payment declined error" the screenshot
The demo also shows a query typed in English pulling up notes written in Bengali and Arabic, and a code result opening in the Zed editor. The cross-language part comes from the model: the README says it understands more than 100 languages. All of this is the project's own material, shown in its own demo recording. Nobody independent has published a measurement of how well the ranking holds up on a real, cluttered disk.
How it works
The engine is EmbeddingGemma 2, the multimodal embedding model Google DeepMind released in October 2026. It maps text, images, audio and video into one shared space, so a photo of a dog on a beach lands near the words "a dog on the beach", and a recording of someone discussing sleep lands near "where they talk about sleep". The README puts the model at 740M parameters, Apache 2.0 licensed, and says that is small enough to run on a laptop: the project's claim, not a benchmark.
DigUp wraps that into an app rather than a script. The repository's own topics name llama.cpp for inference, Swift for the Mac menu bar app, and Tauri for the Windows and Linux builds. Indexing and search both happen on your machine: the README's line is that everything runs on your computer, and nothing in the materials describes a server, an account or a cloud call.
Getting started
It runs on macOS, Windows and Linux. The Mac build is the finished one and lives under the repository's latest release; the Windows and Linux builds are tagged separately as v0.7.0-beta.2. Both are plain downloads from the Releases page at github.com/ARahim3/DigUp rather than a clone-and-build step, and the source is MIT licensed if you would rather compile it yourself.
On cost: the app is free, and because the model is local there is no API key and no paid service behind it. What you pay instead is your own hardware, the CPU or GPU time to index your folders and the disk the index occupies. The project does not list hardware requirements: no stated minimum RAM, no required chip, no VRAM figure, and no estimate of how long a large folder takes to index. Treat the "runs on a laptop" line as an argument about the model's size, not a tested spec for your machine.
When to use it / when not
The main catch is maturity. The Windows and Linux builds are explicitly beta, which both the README and the channel's own write-up flag, and the repository itself is new: created on 8 October 2026 and last pushed on 18 October, with roughly 5,200 stars collected in that short window. That is enthusiasm, not a track record.
Use it instead of Spotlight or Windows Search when the thing you remember is the content of a picture, a sentence someone spoke, or a scene in a video, which name-and-text indexes do not reach. Don't use it if you need something you can point at a shared network drive or a team server, if you want a settled tool with years of bug reports behind it, or if your machine has no spare headroom for a local model to chew through a large library.
DigUp is worth a serious look if you keep a personal archive you can no longer navigate: a decade of photos, lecture recordings, contracts, screenshots named Screenshot 2026-10-18 at 14.02. Mac users get the smoother path today; on Windows and Linux you are an early tester by the project's own admission, so keep your expectations and your backups in order. For developers, it is also a compact example of shipping EmbeddingGemma 2 inside a desktop app with llama.cpp underneath, which may be the most reusable thing in the repository.