Skip to content

SYSTEM Cited by 1 source

LakehouseRT (Reyden real-time analytics engine)

LakehouseRT is Databricks' engine for real-time, low-latency analytics directly over open lake storage. It is powered by a new execution engine codenamed Reyden. Together with Lakebase (the transactional side), LakehouseRT is positioned as the analytical half of the first true LTAP (Lake Transactional Analytical Processing) system — agent-native data infrastructure designed for autonomous, agentic loops. Introduced in the Databricks sponsor talk at VLDB 2026 (Ippokratis Pandis). (Source: sources/2026-08-27-databricks-building-for-the-ai-era-lakebase-streaming-and-lakehouse-innovations-vldb-2026)

What it is

  • Real-time analytics over open storage. Rather than requiring data to be loaded into a proprietary real-time store, LakehouseRT runs low-latency analytical queries directly over open lake storage (the Parquet + Delta/ Iceberg data that already backs the Lakehouse).
  • Reyden engine. A new engine built for the low-latency serving path; internals are not detailed in the announcement.
  • Half of an LTAP system. In the Databricks framing, Lakebase supplies transactional processing over open storage and LakehouseRT supplies analytical processing over the same open storage — unifying the two at the storage layer rather than in a single engine. This is the design that lets a single copy of data serve both OLTP and OLAP without a CDC/ETL hop.

Why it exists — agent-native infrastructure

Databricks argues that autonomous, agentic loops impose new demands on data infrastructure: agents branch and query state at high volume and expect transactional writes and fresh analytical reads over the same data. LakehouseRT is the real-time analytical component of that agent-native stack, alongside Lakebase's compute-storage-separated serverless Postgres.

Caveats

  • Announcement-level disclosure only. "Reyden" and "LakehouseRT" are named but the execution model, latency targets, and how it merges the recent transactional delta with materialized columnar data are not specified in the source. Treat internals as TBD until the underlying paper/design lands.

Seen in

Last updated · 766 distilled / 2,225 read