Show HN: Omni – Local-first multimodal file search on macOS
6 hours ago
- #semantic-search
- #privacy-focused
- #on-device-AI
- Omni allows text, code, PDFs, images, audio, and video to be embedded into a single vector space, enabling text queries to match all types, including scanned pages.
- It supports cross-language search, so a query in one language can find files in another, such as English matching German, Chinese, or Japanese notes.
- Indexing and search run entirely on-device on Macs with Apple silicon and macOS 14+, ensuring files never leave the device, with no accounts, telemetry, or network required at query time.
- The app features list and gallery views, QuickLook thumbnails, drag-and-drop, and filters by kind, folder, and score, all designed for privacy.
- Omni uses a native MLX-Swift port of the jina-embeddings-v5-omni model, which downloads once and works offline on the Apple silicon GPU.
- It exposes search via a local, token-guarded endpoint for agents like Hermes and OpenClaw to query files semantically on the machine, with no cloud round-trip.
- Additional APIs provide raw vector embeddings compatible with OpenAI, Jina, Cohere, and Gemini.