Comment by halb
9 hours ago
i feel that most of my problems in this space would be solved with better "fuzzy" search integration in obsidian. Some sort of local rag on my daily notes would be very effective.
do you know if such a project already exists?
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