DigUp: Search files on Mac by what's in them, using EmbeddingGemma 2 (github.com)

Google DeepMind has launched DigUp, a groundbreaking tool that enables Mac users to search for files based on their content rather than just filenames or metadata, powered by the newly released EmbeddingGemma 2 model. This multimodal embedding model can process text, images, audio, and video within a unified framework, allowing users to find relevant content from various file types simply by describing what they're looking for in natural language. For instance, searching "zebra in a video" will jump directly to the moment the zebra appears, transforming how users interact with their digital files. This development is significant for the AI/ML community as it illustrates the potential of advanced embedding models in practical applications beyond traditional search capabilities, emphasizing the importance of multimodal understanding in AI. With its ability to understand over 100 languages and handle diverse file formats, DigUp offers a powerful enhancement over existing search tools like Spotlight by indexing files based on intrinsic content, including exact matches for words and phrases. The implementation is efficient enough to run locally on Apple Silicon Macs, ensuring privacy by keeping user data on the device and minimizing online interaction, setting a new standard for file search technology.