GitHub - allenv0/SCM: Deep AI search for every photo and every frame of video in any folder on macOS

SCM is a local-first macOS app for searching photos, video scenes, visible text, and spoken dialogue with on-device AI.

Visit Website
GitHub - allenv0/SCM: Deep AI search for every photo and every frame of video in any folder on macOS

Introduction

Overview

SCM, also called Screen Memories, is a local-first media search app for macOS. It indexes photos and videos from selected folders, then lets users find visual memories by describing them in everyday language.

Media stays on the Mac: model inference runs locally, there are no accounts or uploads, and downloaded model weights can be used offline.

Search Capabilities

SCM separates discovery into five focused modes:

  • Files ranks complete photos and videos by visual meaning, with filename and phrase boosts.
  • Scenes searches moments within videos and opens the matching timecode.
  • OCR finds literal text visible in images and video frames, with English and 35 additional language toggles.
  • Dialogue locates exact spoken lines in videos through Whisper transcription.
  • Optional local LLM search answers questions from extracted dialogue, OCR, and filenames with citations.

Library Management

Watched folders are imported automatically, while content hashes prevent renamed files from becoming duplicates. Saved queries can remain open as tabs, and dedicated Screenshots and Email tabs can be toggled as needed.

When the selected model changes, SCM re-embeds the library in the background without blocking search. Video processing segments footage into searchable scenes rather than treating each video only as a single file.

Requirements and Installation

SCM is packaged for macOS, with menu-bar and tray behavior specific to that platform. The documented Homebrew installation supports Apple Silicon on macOS 12 or later.

The first use of the default CLIP model downloads roughly 435 MB of weights. After that initial download, the app can perform its search workflow fully offline.