Show HN: Omni – airgapped macOS multimodal search over local files (github.com)

Omni, a new multimodal search tool for macOS, has been launched, allowing users to perform semantic searches across local files entirely offline. By indexing various file types—documents, code, PDFs, images, audio, and video—Omni enables users to search by meaning rather than just filenames. The application utilizes an embedded model called jina-embeddings-v5-omni, which runs on Apple Silicon Macs with macOS 14 or later, eliminating the need for Python or an external server during queries. Users can easily set it up to index chosen folders and execute searches directly from their device. This development is significant for the AI/ML community as it showcases advancements in local processing power and efficiency for multimodal applications. The model is built using MLX-Swift and supports cross-modal embedding, allowing different media types to coexist in a single vector space for seamless retrieval. With technical features like a batch-processing GPU pipeline for embedding and the storage of embeddings as bf16 format for memory efficiency, Omni not only aims for high accuracy with exact cosine similarity measures but also ensures quick response times during searches. The use of SQLite for metadata management and live indexing further enhances its performance, making it a compelling tool for local AI-driven file organization and discovery.