allenv0/scm is a macOS-only app that searches photos and video frames inside a folder locally with semantic search. According to the GitHub repository description and README, it offers five search modes: file, scene, on-screen text, dialogue, and local Q&A. There is no public latest release. So what is covered here is not the changes of a specific version, but the features presented in the repository.
File search combines the visual similarity of an image with file names and phrases. Scene search splits video into scenes and embeds them, so you can jump to the timecode of a result. OCR uses Tesseract and supports 35 languages besides English. Dialogue search finds text in Whisper transcripts. Local Q&A is answered by a llama.cpp sidecar that adds numbered citations based on dialogue, OCR, and file names. The default model is Qwen3 1.7B. The characteristic of this setup is that scene position, on-screen text, and dialogue search are bundled with photo and video file search in one app.
Search runs locally on macOS and requires no account or cloud upload. After downloading about 435MB of CLIP weights once, it can be used offline. OCR and dialogue search work by string matching without embeddings, so they remain usable even when the AI engine is not ready or is turned off. CJK queries are handled with overlapping bigrams. On the other hand, because of dependencies on Electron, Bun, and local models, there is no Linux or Windows support, and there is no screen capture recording feature either. With no comparative performance data against other tools, a technical advantage has not been confirmed either.