Think back to how often you run into this annoying everyday situation: you know for a fact that six months ago you saved a great video breaking down an architecture, an important PDF handbook, or a screenshot of a unit-economics spreadsheet onto your computer. But what was that file actually named? Was it IMG_4821.png, draft_v2_final.pdf, or a downloaded archive with a meaningless string of digits from Telegram?
You open Windows Search or Mac Spotlight, type in keywords—and get zero results. Traditional OS search relies on exact character matches in filenames or literal string matching in text. If the video title doesn't contain the word "unit-economics," your computer will never find it in a million years.
The team at The AI Advantage introduced a compact open-source utility called DigUp that fundamentally changes how you work with local files. It lets you search documents, presentations, and even video clips not by filename, but by vague human memories.
The Helplessness of Traditional Name-Based Search
Human memory works by association. We don't memorize file hashes or exact folder paths on drive D:. We remember images, context, and emotional details: "that video where a guy in a black hoodie drew microservices architecture with a marker on a whiteboard," "that deck where slide three had a chart showing a sharp conversion drop," or "the contractor agreement discussing penalties for delayed server delivery."
When you try to explain that to a standard OS file explorer, it throws up its hands. You end up manually opening dozens of folders, scrubbing through videos, and flipping through document pages, wasting precious hours of work time.
How DigUp's Local Semantic Index Works
DigUp solves this problem through vector semantic search (Semantic Embeddings) running completely locally on your computer:
- Background content indexing: the app scans selected folders and uses a lightweight local multimodal neural network to convert file contents into numerical vectors of meaning (embeddings). Text documents are broken into semantic paragraphs, while video files are sliced into key still frames;
- Natural-language queries: in the search bar, you can type a completely imprecise, fuzzy query like "guy explaining caching diagram on white background" or "server cost comparison in a table";
- Instant vector matching: the algorithm converts your query into a vector and within fractions of a second finds the closest matching files by meaning, even if they don't share a single exact word with your query.
The best part of DigUp is total privacy. All computation runs on your local CPU and GPU, the vector database lives right on your SSD, and your personal photos, financial reports, and work notes never leak to third-party cloud services. It is a textbook example of practical AI solving an everyday human pain point quietly, reliably, and without unnecessary fanfare.

