---
title: "Lakebase Search: State-of-the-art full text and vector search for Postgres"
description: "Traditional OLTP systems like Postgres face challenges meeting the low-latency, high-accuracy search demands of AI agents, often requiring standalone search engines with ETL pipelines. Lakebase introduces lakebase_vector and lakebase_text, two extensions that bring scalable approximate nearest neighbor search and BM25 full-text search directly into Lakebase Postgres. lakebase_vector overcomes pgvector’s memory-bound HNSW index by using hierarchical IVF clustering and binary quantization, achieving 71-millisecond P99 latency at 97 percent recall on the 100-million-vector LAION benchmark. The extensions decouple storage from compute for pay-per-use scaling, enable fast parallel index builds, and support native hybrid search that fuses vector similarity with keyword relevance in a single SQL query. Early adopters like Conexiom reduced infrastructure costs by three times while achieving five times higher throughput, and the service is generally available on AWS and Azure."
---

# Lakebase Search: State-of-the-art full text and vector search for Postgres

[Databricks](https://yomu.fyi/company/databricks) · Zhou Sun, Jinjing Zhou, Keming Yang, Usamoi Cui, Pranav Aurora · Sep 28, 2026

**Type:** Announcement

## Summary

Traditional OLTP systems like Postgres face challenges meeting the low-latency, high-accuracy search demands of AI agents, often requiring standalone search engines with ETL pipelines. Lakebase introduces lakebase\_vector and lakebase\_text, two extensions that bring scalable approximate nearest neighbor search and BM25 full-text search directly into Lakebase Postgres. lakebase\_vector overcomes pgvector’s memory-bound HNSW index by using hierarchical IVF clustering and binary quantization, achieving 71-millisecond P99 latency at 97 percent recall on the 100-million-vector LAION benchmark. The extensions decouple storage from compute for pay-per-use scaling, enable fast parallel index builds, and support native hybrid search that fuses vector similarity with keyword relevance in a single SQL query. Early adopters like Conexiom reduced infrastructure costs by three times while achieving five times higher throughput, and the service is generally available on AWS and Azure.

## Context

Traditional OLTP databases are not optimized for the low-latency, high-accuracy search workloads required by AI agents, often forcing developers to combine a standalone search engine with the primary database via ETL pipelines. pgvector, the standard vector search extension for Postgres, faces scalability bottlenecks: it requires the entire index to fit in RAM for fast queries, index builds are slow and block writes, and search quality trades off with performance because queries are not parallelized.

## Approach / What changed

Lakebase Search consists of two Postgres extensions: lakebase\_vector, which uses hierarchical IVF clustering and binary quantization (RaBitQ) to enable fast approximate neighbor search even when the index is on object storage, and lakebase\_text, which provides native BM25 full-text search with term scoring using global inverse document frequency. The architecture decouples storage from compute, making indexes stateless and allowing scale-to-zero, parallel index builds, and hybrid search combining vector and BM25 in one SQL query.

## Takeaways

- lakebase\_vector outperforms pgvector at scale by using hierarchical IVF clustering and binary quantization, avoiding memory-bound HNSW; it achieves 71ms P99 latency at 97% recall on 100M vectors.
- Lakebase Search decouples storage from compute, enabling pay-per-use scaling, fast cold starts (1.13s P90 after scale-to-zero for 100M vectors), and index builds that can be offloaded to distributed engines like Spark.
- Native hybrid search in Lakebase Postgres combines vector similarity with BM25 keyword relevance and standard SQL filters in a single query, allowing consolidation of operational and search workloads into one serverless database.

## FAQ

### How does lakebase\_vector achieve fast search on object storage?

It uses hierarchical IVF clustering to group vectors into contiguous blocks that can be read sequentially, and binary quantization (RaBitQ) compresses each vector to about one bit per dimension. When cached, search uses the compact codes; when cold, only the promising blocks are fetched, avoiding random reads across the entire index.

### What are the cost savings compared to pgvector at scale?

lakebase\_vector is four times cheaper than a cloud Postgres vendor using pgvector on the VectorDBBench 100M benchmark, before accounting for autoscaling. Customers like Conexiom reduced database spend by three times while achieving five times higher throughput.

### What is hybrid search in Lakebase Postgres?

Hybrid search combines lakebase\_vector's vector similarity search with lakebase\_text's BM25 full-text scoring in a single SQL query. Developers can also apply standard SQL filter predicates and join against live operational tables, replacing multi-system retrieval pipelines.

**Tags:** [AI Agents](https://yomu.fyi/topic/ai-agents), [BM25](https://yomu.fyi/topic/bm25), [Postgres](https://yomu.fyi/topic/postgres), [Serverless](https://yomu.fyi/topic/serverless), [Vector Search](https://yomu.fyi/topic/vector-search)

- Source: [Databricks](https://www.databricks.com/blog/lakebase-search-state-art-full-text-and-vector-search-postgres)
- Source URL: https://www.databricks.com/blog/lakebase-search-state-art-full-text-and-vector-search-postgres
- Ingested by Yomu: 2026-09-28T21:00:08.204Z

[Read original post](https://www.databricks.com/blog/lakebase-search-state-art-full-text-and-vector-search-postgres)