[cozymori]

First, we don't actually vectorize everything.

We use a hybrid approach to keep costs low and performance high:

Function Metadata: Vectorized only once (for semantic search).

Execution Logs: Stored as standard structured data without embeddings.

Since we don't embed every log, the storage cost is virtually the same as using Postgres. The real value of VectorWave is automating this 'Vector for Search