pgvector in Production: What Nobody Tells You
I've run pgvector in production across multiple real projects. Here are the hard-earned lessons nobody puts in their README, from index tradeoffs and embedding lock-in to Supabase gotchas and similarity threshold calibration.









