Onehouse Lakegres adds vector search to existing Hudi and Iceberg tables
Start here
Onehouse has added vector search to Lakegres, its Postgres-compatible query layer, so embeddings stored in existing and new Apache Hudi and Iceberg tables can be searched through a Postgres endpoint without copying the lake data into a separate vector database. The vector index is derived metadata kept in the table's metadata and updated as the table is written; the underlying Parquet files are left unchanged. On a 247-million-vector Cohere Wikipedia table, Onehouse reports a warm p50 latency of 63.5 ms, against 655 ms for Databricks AI Search, with Recall@10 of 0.95 versus 0.952. It also reports an index build time of 18 minutes, against about 3 hours 10 minutes for Databricks. These figures are Onehouse's own and have not been independently verified.
Generated from reports · updated 1 d ago
Report timeline
Follow the reports to see the event from different sides.
- Hacker News · AIOnehouse Lakegres adds vector search on existing Hudi and Iceberg tables
Onehouse announced vector search in Lakegres, with a native index stored inside existing and new open tables and queried through a Postgres endpoint, so embeddings can be searched without copying lake data into a separate database. The index is derived metadata kept in the table's metadata and updated with table writes, leaving the underlying Parquet files unchanged. On a 247M-vector Cohere Wikipedia table, Onehouse reports warm p50 latency of 63.5 ms versus 655 ms for Databricks AI Search, with Recall@10 of 0.95 versus 0.952, and an index build of 18 minutes versus about 3 hours 10 minutes.
Heat for this event
- Current, comparable range
- 5
- Peak, comparable range
- 10November 19 02:00
- Change in 24 hours
- -50%
The trend compares the same participants over hours observed continuously, so its range can be narrower than the current heat count. Hover or tap the chart for hourly heat; the arrow keys move through it.