What Lakehouse//RT and LTAP Change in the Databricks Lakehouse
Lakehouse real-time analytics with Lakehouse//RT and the LTAP architecture is an approach where transactional, analytical and streaming workloads share one governed copy of data in the lakehouse, enabling millisecond-level queries, consistent governance and AI-ready access without separate serving systems or ETL pipelines. Databricks’ new Lakehouse//RT layer brings a real-time analytics engine directly to Delta Lake and Apache Iceberg tables, while LTAP (Lake Transactional/Analytical Processing) unifies transactional analytical processing across operational and analytical use cases on the same storage. Together they aim to replace the long-standing split between OLTP databases and OLAP warehouses. Instead of copying data into dedicated real-time stacks, enterprises can query fresh records where they already live, under a single governance model, on the Databricks lakehouse platform. This architecture is aimed at both human users and AI agents that demand low-latency, high-concurrency access.
Lakehouse//RT: Real-Time Analytics Without a Separate Serving Layer
Lakehouse//RT introduces a real-time analytics engine, powered by the new Reyden compute engine, directly into the Databricks lakehouse platform. Reyden is designed for tens of thousands of concurrent users and AI agents, while keeping query latency in the millisecond range. Databricks reports that on standard analytical benchmarks, Lakehouse//RT can achieve sub-100 millisecond latency at 12,000 queries per second and deliver up to 16x better performance than specialized real-time serving stacks. The key design choice is that Reyden queries governed Delta Lake and Apache Iceberg tables in place, with no proprietary formats, no extra CDC or sync pipelines and no separate permissions layer. That means nearly any table can be made available for real-time analytics within minutes, while Unity Catalog continues to enforce governance, reducing operational complexity and avoiding new silos in real-time data serving.
LTAP Architecture: Unifying Transactional and Analytical Databases
LTAP architecture extends the lakehouse story beyond analytics into transactional analytical processing, aiming to unify OLTP and OLAP on a single copy of data. Instead of connecting separate operational databases and data warehouses with fragile CDC pipelines, Databricks LTAP brings both workloads to the lake storage layer. Operational data is written using Lakebase, a Postgres-compatible transactional engine that runs directly on object storage, the same layer that backs the lakehouse. That data is immediately available for analytics without ETL, replicas or dedicated streaming pipelines. According to Databricks, LTAP allows transactional and analytical workloads to scale independently with strict isolation, while sharing one storage layer and governance model. Because LTAP is built on open standards like Postgres protocols and table formats such as Delta and Iceberg, it aims to avoid proprietary lock-in while serving as a common foundation for applications, BI tools and AI systems.
From Human-Centric Queries to Agent-Centric Database Design
Both Lakehouse//RT and LTAP are framed as responses to an “agentic era” in which AI agents become primary database users. Traditional architectures assumed human-paced interactions, where nightly ETL and minute-level latency were acceptable. In contrast, AI agents run continuous reasoning loops and need to read, reason and act on data in near real time. Lakehouse real-time analytics directly on Delta and Iceberg tables means agents can work over fresh, governed data without waiting for copies to sync. LTAP strengthens this by ensuring operational transactions and analytical views share the same underlying storage, so agents do not have to reconcile inconsistent sources. As Ali Ghodsi notes, complicated data infrastructure had become a tax teams were forced to pay; now that agents can generate many more applications, that tax turns into a bottleneck the platform aims to remove.
Consolidated Data Infrastructure and the End of ETL-Centric Stacks
For enterprises juggling multiple transactional, analytical and real-time systems, the combined impact of Lakehouse//RT and LTAP is consolidation. Previously, achieving lakehouse real-time analytics often meant operating a separate serving layer, with its own infrastructure, governance model and data copies. Databricks argues that this led to higher infrastructure costs, fragmented governance and data that was never fully real time. By bringing millisecond-level serving directly to the Databricks lakehouse platform and unifying OLTP and OLAP at the storage layer, organizations can retire many CDC pipelines, ETL jobs and replicas that existed only to bridge systems. Every Lakehouse//RT query runs under Unity Catalog governance, while LTAP keeps operational and analytical workloads isolated in compute but aligned on storage. The result is a single, open data foundation that reduces operational complexity and prepares architectures for AI agents as first-class, high-volume database clients.
