VectorDash: A full Windows search replacement incorporating Embeddinggemma-2

Inspired by DigUp for macOS, VectorDash aims to be a full Windows 11 search replacement, powered by EmbeddingGemma 2.

• Full indexing: search by meaning or exact words across photos, screenshots, audio, video, PDFs and Office documents. Find “images of dogs,” a podcast discussion, or a specific PDF clause—with page and timestamp matches. Supports multilingual and optional code search. This uses EmbeddingGemma 2.
• Quick indexing: rapidly index filenames and folders across entire drives, without EmbeddingGemma 2. Combine both modes and filter by source or file type.
• Universal launcher: find and launch Windows apps, open Windows settings, and calculate math directly in search.
• Quick search: customizable global hotkey, thumbnails, previews, and familiar file actions.
• Flexible setup: one-click model/tool setup, GPU, iGPU or CPU inference, and optional remote llama-server.
• Background indexing: scheduled updates, CPU/GPU idle detection, pause/cancel controls, and exclusions for Windows cache folders.
• Agent access: optional read-only MCP server with LAN access
• Local storage: disk-backed SQLite, encrypted by default

Around 200 MB of app RAM in testing with roughly 3,500 fully indexed files and a million quick-indexed files. Options between low-memory, balanced, and performance for search performance. See performance measurements.

Models and processing helpers use additional memory and resources.

File support · Security details

Yep, vibe-coded this. Tested extensively, and I'm just happy I never have to use Windows search again! Feel free to send me bug reports (especially for different GPUs and CPUs) or feature suggestions!