BerriAI / litellm
github.com·3d ago
Developer Tools
Databricks Blog·14d ago·5 min read
Lakebase Search: State-of-the-art full text and vector search for Postgres
TL;DR
Lakebase Postgres introduces two powerful extensions for efficient full-text and vector search, significantly outperforming traditional search engines in speed and cost.
✦ Why It Matters
Consider integrating Lakebase extensions into your Postgres setup to enhance search capabilities immediately.
Key Takeaways
- Explore lakebase_vector for efficient vector search in your applications.
- Utilize lakebase_text for advanced full-text search capabilities.
- Benchmark your current search performance against Lakebase's metrics.
- Consider migrating to Lakebase on AWS or Azure for cost savings.
Full Summary
Traditional Online Transaction Processing (OLTP) systems struggle with the search demands of AI agents, often requiring separate search engines and complex ETL pipelines. Lakebase Postgres addresses this by integrating two extensions: lakebase_vector for scalable approximate neighbor search and lakebase_text for BM25 full-text search.
These extensions are now available on AWS and Azure. Lakebase_vector excels in vector search, achieving twice the throughput of its closest competitor on the VectorDBBench 100M benchmark while being four times cheaper than cloud Postgres vendors using pgvector.
It maintains a P99 latency of 71 milliseconds with a 97% recall rate, ensuring high accuracy in retrieving true nearest neighbors. This integration allows users to run search workloads directly within their OLTP database, simplifying architecture and improving performance.