The data giant open-sourced its agent governance layer and introduced an architecture that merges transactional and analytical workloads on a single copy of data
Databricks just made a play to become the operating system for enterprise AI agents. At its Data + AI Summit 2026, held June 15 to 18 at San Francisco’s Moscone Center, the company rolled out two products that aim to solve problems the data industry has been tripping over for decades: how to govern autonomous AI agents across an enterprise, and how to stop copying the same data into seventeen different databases just to run different workloads.
The two headliners are Omnigent, an open-source governance harness for AI agents, and LTAP, short for Lake Transactional/Analytical Processing, a new architecture that unifies transactional and analytical workloads on a single governed copy of data. Around 30,000 attendees from 174 countries were there for the reveal.
Omnigent: a universal remote for AI agents #
Omnigent is a meta-layer that sits on top of whatever agent framework a company is already using. It doesn’t replace those frameworks. It wraps around them, enforcing policies, enabling interoperability between different models, and allowing live sessions to be shared across agents.
Databricks open-sourced Omnigent under the Apache 2.0 license on or around June 13, 2026, a couple of days before the summit officially kicked off. The managed beta is already available on the Databricks platform for customers who want the hosted version with enterprise support baked in.
LTAP: killing the data copy problem #
LTAP was announced on June 16 and tackles the artificial separation between transactional processing (OLTP) and analytical processing (OLAP). Most enterprises today maintain separate databases for running their business and for analyzing it, which means copying data between systems, building complex ETL pipelines to keep everything in sync, and paying for storage and compute multiple times.
LTAP eliminates that duplication by running both transactional and analytical workloads on a single copy of data stored in open formats like Delta Lake and Iceberg. Databricks claims there are no performance compromises between the two workload types on this shared storage infrastructure.
LTAP also integrates streaming workloads alongside the transactional and analytical ones, meaning real-time data doesn’t need yet another separate pipeline.
The agentic ecosystem play #
Omnigent and LTAP are part of a broader push toward what Databricks calls the “agentic enterprise.” Other pieces of that ecosystem include Genie One, Unity AI Gateway, and Lakehouse//RT, an engine designed for real-time data serving. Previous Lakehouse innovations and Lakebase were referenced as foundational elements that these new tools build upon.
What this means for investors #
Databricks remains private, so there’s no stock ticker to watch. If LTAP delivers on its promise of unified transactional and analytical processing without performance degradation, it directly threatens traditional database vendors who have built their businesses on the assumption that enterprises need separate systems for separate workloads.
The open-sourcing of Omnigent creates a potential standard for agent governance. Databricks’ advantage with LTAP is that it’s building on open formats rather than proprietary storage, which lowers switching costs. Previous hybrid systems, often called HTAP, have been attempted by companies like SAP HANA and Oracle with mixed results.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our