New macOS Tool Brings AI Search to Every Photo and Video Frame

A developer known on GitHub as allenv0 has published SCM, an open-source macOS application that turns a machine's photo and video collections into a queryable index. The pitch is straightforward: type a description, get matching stills and video moments back, with nothing leaving the local drive. The project appeared in a Show HN submission and had not drawn public comments when logged.

What sets SCM apart from the long list of Mac photo organizers is how it treats video. Rather than indexing clips as single opaque files, it samples and embeds individual frames, so a search can surface a specific moment inside a longer recording. Still images are handled the same way, meaning filenames, creation dates and manual tags are not the only handles a user has on their archive.

Apple's built-in Photos app does offer text search and object recognition, but its coverage of video is uneven, and libraries scattered across external drives, folder trees and old exports often fall outside what it indexes. SCM appears built for exactly those collections. The approach rests on vision-language embedding models that are now small and fast enough to run on Apple silicon, which is what makes an on-device index practical in the first place.

Keeping the pipeline local has obvious appeal for anyone holding family footage, client work or sensitive recordings, since no frames are uploaded to a third-party service. The cost shows up in two places: the first scan can take a long time, and the resulting vector store eats disk space. Scanning thousands of hours of 4K footage frame by frame is not cheap, and there is no indication yet of how SCM manages that load.

Existing open-source options split into two groups. Immich and PhotoPrism aim at self-hosted libraries and need a server running somewhere, while lighter Mac utilities mostly handle duplicate detection or metadata cleanup rather than semantic retrieval. A native, local-first search layer covering both stills and video frames would be a genuinely different entry, though its value depends on how it copes with mixed formats, HEIC files and large 4K libraries in practice.

The Show HN post is the project's only real visibility so far, and the thread carried no comments when it was recorded, so no independent testing, bug reports or accuracy measurements have surfaced publicly. For software that promises to index an entire media archive, that lack of outside scrutiny is the immediate caveat. Anyone interested should check the repository for supported macOS versions, hardware requirements and whether indexing runs as a background process that drains battery.

The next useful signals will be published benchmarks on index build speed and search precision, plus early user reports from libraries in the tens of thousands of files. If SCM can pull up a decade-old home video clip in seconds without sending data anywhere, it addresses a gap Apple has left open. Until those numbers exist, it is a repository with a strong idea and no field evidence.