Show HN: AI search for every photo and every frame of video on macOS

SCM, short for Screen Memories, is an MIT-licensed Electron app. It offers five search modes: Files, Scenes for timecoded video moments, OCR through Tesseract, Dialogue through Whisper transcription, and an optional local chat over extracted text. The default CLIP model downloads about 435MB of weights on first use, and everything after that runs offline. Video shot detection runs through ffmpeg. Version 0.2.5 was released on Oct. 4 through a Homebrew tap. The repository has about 366 stars and 21 forks. Reception on Hacker News was mixed. Digests of the thread note praise for the privacy-first design. One commenter argued the project looks like a single LLM prompt dumped onto GitHub without review. The same commenter flagged edge-case bugs in the metadata code. Alternatives named in coverage include Apple Photos and DEVONthink. For AI founders, the signal is that a small team can now ship multimodal retrieval with no server cost and no data-handling burden. Client assets often contain sensitive material that cannot go to cloud services. That makes on-device search a credible wedge for creative and enterprise users. The risks are a roughly 435MB model download, Apple Silicon-only support, and questions about code quality. Cloud-based media search products will have to compete on accuracy or scale, since privacy no longer favors them.