There are a few directions I know but not sure:

“semantic search via embeddings” plugins (Obsidian Semantic Search is one example)

local-first assistants via Ollama (e.g. obsidian-local-gpt)

full “RAG over your vault” projects (I’ve seen ObsidianRAG-style setups)

My takeaway so far is similar to what others said here: reliability + noise + UX are the hard parts, not just “having embeddings”. What would “good enough” look like for you: speed, recall quality, or tight integration into daily notes?

Something like Postgres embedding search works out of the box, but don’t forget any search engine today also needs reranking

https://github.com/with-logic/intent