What LTAP and Lakehouse//RT Are, in Plain Terms
Lake Transactional/Analytical Processing (LTAP) and Lakehouse//RT are Databricks’ new architectures that merge real-time transactional data, historical analytics, and streaming workloads on one shared lakehouse foundation, so enterprises no longer maintain separate operational and analytical databases or the fragile pipelines that connect them. For decades, organisations split online transactional processing (OLTP) and online analytical processing (OLAP), then copied data through ETL, replicas, and change data capture jobs. Databricks now proposes a single governed storage layer where operational tables, analytical models, and event streams live together in open formats, while specialised compute engines handle different workload patterns. In this model, Lakehouse//RT provides millisecond-level lakehouse real-time analytics, while LTAP defines how transactions, analytics, and agents interact with that unified data. The goal is a data lakehouse consolidation that suits AI agents as the primary consumers of the platform.
From Dual Databases to Unified Database Architecture
Databricks’ LTAP proposal attacks the long-standing split between OLTP and OLAP by unifying data at the storage layer rather than in a single engine. Historically, operational systems wrote rows quickly, analytical systems scanned columns for reporting, and teams wired them together with extraction, loading, and transformation. Hybrid transactional and analytical processing systems tried to do both in one engine, but often sacrificed isolation and created expensive proprietary stacks. Zero-ETL services hid pipelines instead of removing them, still leaving multiple copies of data that could go stale. LTAP keeps a single copy of data in open table formats such as Delta and Iceberg on cloud object storage, while Lakebase and the lakehouse engines scale independently. According to Databricks, “transactional and analytical workloads scale independently with full performance and strict isolation,” which is central to its unified database architecture story.
Lakehouse//RT: Real-Time Analytics Without a Separate Serving Layer
Lakehouse//RT is the real-time analytics side of this design, aimed at removing dedicated serving systems from the stack. It runs a new vectorised compute engine, Reyden, directly on Delta Lake and Apache Iceberg tables, avoiding proprietary formats or extra ingestion pipelines. Databricks says Lakehouse//RT can expose almost any lakehouse table for real-time querying within minutes and achieve latency below 100 milliseconds while processing 12,000 queries per second, serving tens of thousands of concurrent users and AI agents. This performance level previously required separate key-value or specialised serving layers tuned for millisecond queries, along with complex change data capture and synchronisation. By running lakehouse real-time analytics on the same governed tables, organisations can phase out those sidecar systems, simplifying governance and reducing the risk of inconsistencies between transactional and analytical views.
AI Agents as First-Class Users of the Data Lakehouse
A key design assumption behind LTAP and Lakehouse//RT is that AI agents, not humans, will be the main users of enterprise data platforms. Databricks argues that agents that write code, make calls, and run loops at machine speed need fresh transactional context and rich historical analytics in one place. Lakebase, the Postgres-based operational engine at the heart of LTAP, already handles 12 million database launches per day, and now adds features such as native vector search, full-text search, and Git-style branching of databases so agents can experiment on isolated copies. In this world, transactional analytical processing is optimised for continuous, automated reasoning rather than scheduled dashboards. Agents can read operational events, query years of history, and trigger actions through the same unified database architecture, without waiting for ETL jobs or manually tuned replicas to catch up.
Cost, Complexity, and the Shift to Lakehouse Consolidation
By collapsing OLTP, OLAP, and real-time serving into a single LTAP lakehouse, Databricks targets two chronic enterprise problems: operational complexity and infrastructure cost. Maintaining many systems for transactions, batch analytics, and low-latency serving leads to duplicated storage, overlapping governance rules, and brittle pipelines that break under change. With LTAP, all operational, analytical, and streaming data sits in one governed storage layer, while Lakehouse//RT handles lakehouse real-time analytics for high-concurrency workloads. Databricks contends that removing separate serving layers also removes the need for change data capture jobs, synchronisation pipelines, and vendor-specific formats that lock organisations into proprietary stacks. The result is data lakehouse consolidation: fewer moving parts to manage, a single source of truth to secure, and a platform shaped around continuous AI-agent consumption rather than periodic human queries.
