What matters in AI.

Subscribe

Onehouse adds vector search to Lakegres on open lake tables

Lakegres keeps the vector index in the table metadata, and teams do not copy data into a different vector database.

Onehouse announced vector search in Lakegres, its tool that lets Postgres clients query data lake tables. Vector search finds items with the same topic, from lists of numbers called embeddings. Onehouse says its index is in the same bucket as the Hudi or Iceberg table, in open formats, and this removes a copy of the data.

How it works

Hudi and Iceberg are open table formats. Lakegres groups vectors in clusters and stores small codes in the table metadata. Each write to the table also updates the index, and a Postgres client can find items by topic with SQL.

The numbers

Onehouse says the HNSW graph index must keep all the graph in memory. By the AWS formula, 247 million vectors at 1,024 dimensions use about 1.15 TB of memory, and 10 billion vectors use about 46 TB. The text gives no speed or cost test results for Lakegres.

What is not known

Onehouse says this is the only open index of its type for Hudi and Iceberg tables that you have, which is its claim. The Hudi code is not in the open source project at this time, and Apache XTable is in release candidate tests. Other engines cannot use it.

Insight: Open tables now compete on who owns the search index

Onehouse says other lakehouse vendors keep indexes in closed formats, and a move to a different engine forces a new index. On 4 November 2026, Google let AlloyDB read Iceberg tables where they are. This reads as a contest for the index layer, because the company that holds the index can hold the customer. Find where your vector index can be if you change engines.

Sources

  1. Lakegres: Vector search on existing Open/Hudi/Iceberg tablesonehouse.ai
AI MATTER · NEWS · AI MATTER · NEWS ·18 NOV2026

Posted

Tags

Learn the terms in this story

More in Compute

All Compute news