What Lakehouse//RT Is and Why It Matters
Databricks Lakehouse//RT is a real-time analytics layer that lets organizations query transactional and historical data directly in a single governed data lakehouse platform, removing the need for separate real-time serving databases, ETL pipelines, and data copies while still delivering millisecond-level response times at high concurrency for both human users and AI agents. Enterprises have long split workloads: operational databases for transactions, warehouses or lakes for analytics, and a separate serving tier for low latency queries. That model introduces data movement, lag, fragile CDC jobs, and scattered governance. Lakehouse//RT attacks this by running real-time analytics on Delta Lake and Apache Iceberg tables that are already governed in Unity Catalog, so the same tables can power dashboards, agents, and applications. For data teams under pressure to support AI agents and streaming use cases, Lakehouse//RT is less another system and more a simplification of the whole lakehouse architecture.
Inside Lakehouse//RT: Real-Time Analytics Without a Separate Serving Layer
Lakehouse//RT is powered by Reyden, a new compute engine designed for low-latency, high-concurrency workloads directly on Delta and Iceberg tables. Databricks reports that, on standard analytical benchmarks, Lakehouse//RT achieves sub-100 millisecond latency while processing 12,000 queries per second and has delivered up to 16x better performance than specialized real-time serving stacks. Because queries run inside Unity Catalog’s governance framework, there is no extra permissions tier, proprietary storage format, or sync layer to maintain. Any table can be prepared for real-time analytics in minutes, so product and analytics teams can expose fresh data to applications without building new ingestion paths. Eliminating a separate serving cluster also cuts operational overhead: no more CDC pipelines to keep transactional and analytical stores in sync, and fewer moving parts for SREs to watch. In practice, Lakehouse//RT turns the governed lakehouse itself into the serving system.
LTAP: Unifying Transactional and Analytical Processing in the Lake
While Lakehouse//RT focuses on real-time analytics, LTAP (Lake Transactional/Analytical Processing) tackles a deeper architectural divide: transactional analytical processing across OLTP, OLAP, and streaming on one shared storage layer. Databricks positions LTAP as the first architecture to keep all operational, analytical, and streaming data on a single copy in the lake, instead of pushing changes through CDC pipelines into warehouses or serving stores. The foundation is Lakebase, a Postgres-compatible system that brings ACID transactions to open object storage, the same storage that underpins the lakehouse. Lakebase already serves thousands of customers and handles 12 million database launches per day. In LTAP, transactional and analytical workloads still scale independently and keep strict isolation, but they operate against the same data, removing ETL by design. Any application that speaks Postgres and any engine that reads Delta or Iceberg can share this unified lakehouse architecture.
From HTAP and Zero ETL to LTAP and the Real-Time Lakehouse
Databricks contrasts LTAP and Lakehouse//RT with earlier attempts to bridge OLTP and OLAP. HTAP engines aimed to handle both workloads in one system, but often sacrificed isolation and performance while locking customers into proprietary stacks. Zero ETL approaches hid pipelines rather than removing them, so data still moved and broke under load. In LTAP, the unification moves down to the storage layer: operational writes land in Lakebase and are immediately available in open table formats for analytics, streaming, and real-time queries. According to Databricks, “transactional and analytical workloads scale independently with full performance and strict isolation,” while still sharing one governed source of truth. Lakehouse//RT then adds a real-time analytics engine on top of this shared storage, giving teams millisecond response times on the same tables that back their applications. The result is a lakehouse architecture that serves both agents and humans without duplicating data.
What This Means for Enterprise Data Teams
For enterprise data teams, Lakehouse//RT and LTAP reframe how to design real-time analytics and transactional analytical processing. Instead of stitching together OLTP databases, change-data-capture tools, streaming buses, warehouses, and low-latency serving stores, teams can center their architecture on a single data lakehouse platform with one storage layer and one governance model. Real-time analytics can run directly on Delta and Iceberg tables, while transactional workloads sit on Lakebase, all without ETL pipelines or replicas. That reduces latency, simplifies observability, and cuts the operational burden of keeping many systems consistent. It also shortens the path from data to AI: agents can query current and historical data in one place, at high throughput, without custom sync logic. For organizations facing a surge of AI-driven applications and streaming data, the promise is less about new features and more about making the lakehouse the only system they need to run both transactions and analytics.
