Think about the last time you looked for a photo on your Mac. You knew it existed. You knew roughly when you took it. And you still ended up scrolling through a thumbnail grid like someone flipping through a shoebox of unlabeled negatives, squinting at 80-pixel squares until your eyes gave out. Decades into digital photography, most of us still store images the way our grandparents stored slides: in a pile, sorted by date, searchable only by memory.
SCM, a macOS tool from a developer going by allenv0, is a swing at that problem. It runs AI search across every photo and every frame of video in a folder on your Mac. Local vision models do the looking. Ollama does the hosting. Nothing leaves your hardware.
What the numbers actually say
Let’s be honest about scale before anyone starts calling this the future of photo management. The Show HN thread picked up 86 points and 47 comments last week. The GitHub repo sits at 52 stars. In Hacker News terms, that’s a real signal but a modest one. It means practitioners looked, thought about it, and had opinions worth typing out. It does not mean a movement.
I bring this up because the comment-to-point ratio is the interesting part. Forty-seven comments against 86 points is unusually chatty. Projects that get waved past with a reflex upvote don’t generate that kind of discussion. Something about this one made people want to argue, ask questions, or share their own version of the same frustration. That’s a better early indicator than a star count ever is.
The part that matters more than the AI
The pitch here isn’t that AI can recognize objects in pictures. That’s been true for years, and every major cloud photo service already does it. Upload your library to any of them and you can search for “dog” or “beach” and get results.
The difference is where the processing happens. SCM analyzes your library on your machine, through local models, and renames files based on what the model sees. Your photos stay on your disk. No upload, no account, no terms of service update next March that quietly changes what the company can train on.
For a certain kind of user, that’s the entire value proposition. Journalists with source material. Lawyers with case photos. Medical professionals. Anyone who has signed an NDA. Parents who are not thrilled about a server farm indexing their kids’ faces. These people have been stuck choosing between useful search and keeping their data private, and the choice has mostly gone the wrong way because the privacy option didn’t exist in usable form.
Video frames are the sneaky ambition
Searching photos is a solved problem technically. Searching every frame of video is a different animal, and it’s the claim in this project that deserves the most scrutiny.
A single minute of 30fps footage is 1,800 frames. An hour is over 100,000. Run a vision model against each one and you are looking at real compute time and real heat coming off your laptop. Any tool making this promise has to answer practical questions: Does it sample frames or process all of them? How long does a 10GB folder take? What happens to battery life?
The verified material doesn’t tell us. That’s not a knock on the project, it’s a knock on how early-stage tools get covered. The feature list is public. The performance profile is not. And with local inference, performance is the feature. A tool that indexes your video archive in an afternoon is useful. One that takes four days and cooks your MacBook is a science project.
My honest read
SCM is pointed at a genuine gap, and the local-first approach is the right instinct rather than a marketing angle. Ollama as the model layer is a sensible choice, since it means the tool inherits improvements as open vision models get better without needing a rewrite.
What I can’t tell you yet is whether it works well enough to live in your workflow. Fifty-two stars and one Show HN thread is the very beginning of a project’s life. The gap between “the demo works” and “this handles my 400GB archive without complaint” is where most tools in this category quietly die.
If you’re on macOS, already running Ollama, and you have a folder of media you’ve given up on ever searching, this is worth an evening of your time. The downside is a wasted evening. The upside is finally being able to ask your own hard drive a question and get an answer.
If you need something dependable for client work tomorrow, wait. Let the early adopters in that comment thread find the rough edges first. That’s what they’re for.
🕒 Published: