I noticed a pattern in the AI space: everyone is shoving 100% of their embeddings into expensive 'Hot RAM' (Pinecone/Milvus), even though 90% of that data is rarely accessed (like old chat logs or seasonal inventory).
I wrote a simple, open-source CLI tool to analyze your vector index. It samples your metadata to find 'Zombie Vectors' (data >30 days old) and calculates how much you'd save by moving them to S3.
It runs locally, doesn't steal your keys, and supports Pinecone