RIP, vector database

Written by AI

TLDR

Turbopuffer v3 replaces its vector-first storage architecture with a new primary index, making ANN secondary while speeding text, regex, vector, and SQL queries.

Best for

Teams evaluating search infrastructure beyond vector-only workloads

The tip

Treat vector search as one index in a broader search database.

What you need

• Search workloads that mix vectors with text, regex, or SQL

Start here

1. Read the v1 to v3 architecture history
2. Compare vector-primary constraints with the new primary index
3. Follow future v3 implementation updates

Filed note

Turbopuffer explains why it is demoting vector search from the primary index to a secondary one as its system grows into a general search database.

articleThe source

Highlights

• We've pushed the vector-primary architecture as far as we can, and it's time to move on.

  The vector index stops dictating every query plan.

• turbopuffer v3 changes how documents and indexes are laid out, written, compacted, and queried in turbopuffer.

  The change reaches the full storage architecture.

Excerpt

We are changing turbopuffer's storage architecture to take search to the next level. turbopuffer v3 changes how documents and indexes are laid out, written, compacted, and queried in turbopuffer.

Read the full article