Databricks Lakehouse//RT aims to make separate real-time databases redundant

Databricks has launched Lakehouse//RT, an engine designed for real-time analytics directly on Delta Lake and Apache Iceberg tables. The new technology aims to allow applications and AI agents to access data directly from the Lakehouse without the need for additional serving systems or specialized real-time databases.

The real-time analytics market has historically been dominated by specialized systems such as ClickHouse, Apache Pinot, or Apache Druid. Companies often use these in addition to their analytics platforms to ensure low-latency data delivery to applications, dashboards, or AI systems, but the data copies and synchronization processes involved increase complexity and costs.

Lakehouse//RT aims to simplify this architecture by executing queries with millisecond-range latencies directly on Delta Lake and Apache Iceberg tables, thereby removing the need for a separate real-time serving layer. Databricks justifies this move partly due to the increasing use of AI agents, which require up-to-date information with minimal latency. Lakehouse//RT enables data access directly from the Lakehouse without intermediate copying.

The system integrates with Databricks Unity Catalog, meaning governance policies and permissions apply to real-time queries as well. This eliminates the need for separate governance layers or proprietary data formats, allowing access to live data from existing Delta or Iceberg tables within minutes.