Lakebase Postgres now includes built-in full-text and vector search, generally available on AWS and Azure. With lakebase_vector for approximate nearest-neighbor search and lakebase_text for BM25 search, teams can run semantic, keyword, and hybrid retrieval alongside operational data, without maintaining a separate search system and ETL pipeline.
Key Highlights
- Search directly in Postgres: Combine vector similarity, BM25 keyword relevance, SQL filters, and joins against live operational tables in one database.
- Built for scale: Lakebase Search uses a serverless architecture that separates storage from compute, scales to zero when idle, and scales compute with query demand rather than data volume.
- Designed for agent retrieval: Support bursty, parallel retrieval workloads with usage-based scaling, low-latency queries, and governed access through standard database controls.
- Efficient vector search: lakebase_vector uses hierarchical clustering and compact quantization to search only the relevant data while balancing recall and latency. The reported 100M-vector benchmark achieved 97% recall at 71 ms P99 latency.
- Native hybrid search: Combine lakebase_vector and lakebase_text to match both semantic meaning and exact keywords, useful for agent grounding and search experiences that need both signals.
- Simpler operations: Keep OLTP and search workloads together, with index builds designed to run in parallel and away from the primary database.
๐ Read the full story here
