I mean, I don’t think we should throw the baby out with the bath-sewage.
I do really want a file system that allows semantic search via embeddings. That would give good natural language retrieval or even person based search on your photos.
You can run those entirely locally. You can train them reasonably, though the AI labs have good quality ones. They’re available for free, and I can commute thousands per second on a modest GPU.
Why not have a filesystem integration for them, with pluggable model support. You can do that with lancedb or chromadb, or pgvector, or you could build this into a SQLite plugin.
Also, making hooks into software that agents can use if needed, also enables accessibility software to use them. I’ve been using AppleScript for like 20 years to automate shit, it’s nice that now I can plug agents into it, and all sorts of macOS features are built on the same bridges.
There’s no need to default to plugging into the US based providers who (by US law) don’t respect privacy.