SCM Ships MIT-Licensed, Offline CLIP Search for Mac Media
TL;DR
- SCM is an MIT-licensed Electron app for macOS that indexes photos and video frames locally and runs CLIP inference fully offline after a first-use download.
- The default model is CLIP ViT-L/14@336 at roughly 435MB, with 480-570ms per image on CPU and SigLIP variants offered as speed/detail trade-offs.
- Five search modes ship: Files, Scenes, OCR via Tesseract in 35+ languages, Dialogue via Whisper transcription, and an optional local LLMs 'Ask' chat.
SCM, short for Screen Memories, is a new MIT-licensed Electron app for macOS that indexes every photo and every video frame in watched folders on-device, so a plain-language description retrieves matching stills and video moments without a cloud call. Its GitHub README pitches the model bluntly: "Your media never leaves the machine."
The default vision model is CLIP ViT-L/14@336, downloaded once at roughly 435MB and running about 480-570ms per image on CPU, with SigLIP variants offered as speed-and-detail alternatives. Search covers five modes: Files for semantic photo and video retrieval, Scenes for timecoded video moments, OCR via Tesseract across 35+ languages, Dialogue via Whisper transcription (tiny.en at ~150MB or base.en at ~300MB), and an optional local "LLMs (Ask)" chat over extracted text. Video shot detection runs through ffmpeg, data lives under `~/Library/Application Support/scm`, and the README states that "No telemetry or external uploads occur."
It lands in a steady run of local-model launches we have been logging on the open-source AI tracker, and squarely in the edge AI lane where the Mac keeps picking up jobs that used to belong to a server.
Originally reported by github.com
Read the original article →Original headline: Show HN: SCM Ships Local-First macOS Search Across Every Photo and Video Frame via CLIP