Local Search for Photos, Videos, and Code Using Natural Language

·2026.10.16 17:30

Key point

DigUp is a desktop app released under the MIT license that uses EmbeddingGemma 2 to search photos, PDFs, videos, and code locally without the cloud.

Details

DigUp runs Google DeepMind's EmbeddingGemma 2 model via llama.cpp to convert text, images, and audio into a 768-dimensional vector space for cross-modal search. During search, it loads only the ~250MB text portion into memory to optimize speed, while the full model file is 865MB (310MB for text, 555MB for the encoder). The Mac version is a stable release, while Windows and Linux are in beta. The app is licensed under MIT, and the model is licensed under Apache 2.0.

multimodal llama.cpp embeddinggemma local-search digup

Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.