It would be very convenient if, when searching in Discover, I could type something like “program to read PDF files” in any language, and it would return PDF reader applications for me.
This would be possible thanks to semantic search: my query is converted into a vector of meaning and compared with all vectors in the database, and the most relevant results appear at the top.
We already have Ollama and llama.cpp, which are sufficiently stable and can easily run embedding models