EDB launches agentic Postgres AI with governance tools

EDB has launched new agentic database and converged analytics features for its Postgres AI platform, aimed at letting enterprises run AI agents and transactional workloads on the same PostgreSQL base.

The release also adds governance tools in preview, positioning controls at the data layer rather than in separate systems. The agentic database and converged analytics products, including EDB PG AI for ClickHouse, are generally available.

The announcement reflects a wider push by database and analytics suppliers to bring AI processing closer to operational data. Businesses want ways to connect live enterprise records with AI systems without moving information across multiple platforms, particularly where regulated or sensitive data is involved.

EDB said its agentic database feature is designed to turn Postgres into a self-managing system that monitors more than 200 operational and performance metrics. The software can identify issues, recommend changes and, where policy allows, apply fixes automatically.

Users can decide whether individual actions are automated, routed for human approval or delayed until a maintenance window. Each action is recorded in an audit trail.

According to EDB, the platform brings relational, JSON, time-series, geospatial and vector data together through one SQL interface. The approach is intended to support operational, analytical and AI workloads without splitting data across separate engines.

EDB said the system can make database tuning up to 10 times faster, cutting work that might take a database administrator 60 to 90 minutes to a process of minutes. It also said the same optimisation can improve application performance by up to eight times for end users.

Kevin Dallas, chief executive officer of EDB, set out the company's view of the market shift.

"The industry spent a decade telling enterprises to move everything into the lake. That's exactly backwards for agents," said Kevin Dallas, chief executive officer of EDB. "Agents act in the moment, on live data, under real rules. You don't get speed, accuracy, or sovereignty by reaching into a cloud for a copy. You get it by bringing the intelligence to the data. That's what we built. Your AI, your data, your rules, on infrastructure you own."

Analytics push

Alongside the database changes, EDB has expanded the analytics side of the platform with what it calls a zero-ETL architecture. This is intended to make operational and analytical data continuously available for real-time analysis and large-scale warehousing from the same environment.

EDB PG AI for ClickHouse is now generally available as part of the release, aimed at real-time analysis of event and log data. EDB PG AI for WarehousePG is aimed at historical analysis and complex reporting at petabyte scale, while heavier workloads can be offloaded to GPU-accelerated Spark.

Compared with legacy warehouses and cloud data platforms, EDB said the converged analytics functions can deliver up to 30 times faster single-node query performance in internal testing, rising to 99 times through GPU offloading. Customers could also see up to 52% greater scaling efficiency for high-concurrency workloads and up to 58% lower total cost of ownership, it said.

"Every other approach asks you to move your data to the intelligence. We did the opposite-we put the intelligence in the database, on infrastructure you own," said Max Romanenko, chief engineering officer of EDB. "It's the database that runs itself, on your terms. That's not a feature you bolt on. It's the foundation."

EDB cited Kyobo Book Centre, one of South Korea's largest booksellers, as a customer that reworked its analytics environment on an on-premises WarehousePG base. The retailer projected significant total cost of ownership savings while creating a data platform for AI and vector-based services, according to EDB.

AI retrieval

EDB also used the launch to emphasise vector search and retrieval for AI agents. It said the platform combines structured and unstructured data, analytics and vector search in one query layer so agents can work with authorised data without relying on a separate vector database.

The company pointed to benchmarking by McKnight Consulting Group that it said showed lower query latency and higher retrieval accuracy than rival platforms including Databricks and MongoDB. According to EDB, the testing found up to 99.4% lower query latency than Databricks, 93% lower than MongoDB, and a Recall@10 score of 0.911.

EDB also said new writes become queryable in 12 milliseconds, compared with 3.8 seconds for Databricks. It argued that this matters for workloads that rely on current data, and that the combination of speed, freshness and transactional consistency suits autonomous agents acting on live enterprise records.

NTT East, one of Japan's largest telecommunications carriers, is using EDB PG AI in AI-led network operations, according to the company. EDB said the deployment involves generative AI agents that detect, analyse and respond to network issues in a private environment where operational data remains under the carrier's control.

Governance layer

A further part of the release is a governance function now in preview. EDB said this feature uses native Postgres roles and row-level security to control agent access at the point where data is queried, rather than through a separate control plane.

An agent's identity, purpose, permissions and organisational policies are combined into a constrained query executed directly by Postgres, according to EDB. The company said this is intended to ensure that agent actions are subject to the same controls and auditing as human user actions.

"Agents don't act on copies. They act on the real thing-live, governed, right where it sits, with no separate system to secure and no lake to fall out of sync with," said Romanenko.

EDB said the platform is built on open Postgres and open table formats and can be deployed on-premises, in hybrid environments or across clouds. Its partner ecosystem includes Dell, IBM, Nvidia, Red Hat and Supermicro.

"IBM Power and EDB Postgres AI are empowering enterprises for the AI-native era by providing a secured, sovereign, and AI-ready infrastructure foundation. Together, we enable a resilient data ecosystem that supports data sovereignty," said Unnikrishnan Rajagopal, ww director for ISV ecosystem, GSIs and alliances at IBM.