@Nate Are you using TurboPuffer cloud service?

yes!

Very cool. Can you share any other alternatives you've considered?

chromadb is my go to for almost everything! its open source and they also recently started hosting a managed solution ive primarily used chromadb over time and dabbled in turbopuffer, pinecone (not a fan) and pgvector i'd like to spend more time with pgvector

So why not chromadb?

well I do use chromadb in a lot of places currently. i haven't updated the slackbot implementation in quite a while so tpuf was just the vectorstore i was checking out when I deployed the slackbot i will eventually rewrite the slackbot using controlflow or pydantic ai, I just haven't gotten around to it. imo the RAG performance bottleneck is not "who gives the best cosine similarity between documents" but instead how to • enrich documents with metadata at ingest time • empower the AI to filter by this metadata at query time • combine traditional search with semantic search who gives me cosine similarity is not the most consequential in my eyes, they're all quite similar

Got it. Thank you for that!