Research & Papers

Apple Silicon gets on-device omni-modal search with omni-macos

Researcher builds a privacy-first search engine that runs entirely on your Mac

Deep Dive

Independent researcher Han Xiao has developed omni-macos, a groundbreaking on-device search engine designed to run entirely on Apple Silicon Macs without ever sending files, queries, or vectors to external servers.

The system integrates a background indexer and interactive search interface within a single memory budget. It intelligently manages computational resources by re-encoding only modified content, processing smaller data units during user input, and using quantized replicas for efficient querying. Tested across five Macs with diverse hardware configurations (spanning 8x accelerator width and 32x memory range), the engine maintains consistent performance by dynamically allocating unified memory resources based on user settings.

Key Points
  • omni-macos performs omni-modal search (text, code, images, audio, video) entirely on-device on Apple Silicon Macs
  • System dynamically manages GPU/memory resources, re-encodes only changed content, and uses quantized replicas for efficient querying
  • Tested successfully across five Macs with 8x accelerator width and 32x memory range variations

Why It Matters

Privacy-focused search without cloud dependency for professionals handling sensitive data locally.

📬 Get the top 10 AI stories daily