pgEdge AI DBA Workbench Now Supports Jev and Other Decision Models

Related stories

First Application of Decision Models Like Jev for Faster and More Economical Agentic Postgres Database Monitoring and Administration

pgEdge, the leading open source Postgres company for agentic AI and enterprise applications, announced the pgEdge AI DBA Workbench now supports decision models such as Jev, a decision model from TypeSafe AI, yielding significantly faster and more economical agentic monitoring and administration for Postgres databases. In pgEdge testing, Jev classified database monitoring alerts 15 times faster and at nearly one-fiftieth the cost of Claude Haiku 4.5. This is among the first practical applications of Jev and other decision models for use with the open source Postgres database.

The pgEdge AI DBA Workbench is an agentic database monitoring and administration tool for Postgres that dramatically increases the number of databases a human DBA can monitor and manage, while keeping them in the loop for actual problem resolutions. It relies on a three-tier anomaly detection system that leverages both statistical analysis and AI models to detect, classify and analyze database issues across a wide range of Postgres health and performance metrics.

Jev and other decision models can be used in place of large language models (LLMs) for decisions and classification, at much greater speed and lower cost. When presented with a decision or a block of text to classify, a decision model rapidly returns answers, each ranked with a rated probability. However, there is no textual response to provide reasoning behind the answers. The AI DBA Workbench can be optionally configured to use Jev (or equivalent) for classification of monitoring alerts, with an on-demand automatic fallback to an LLM to provide the reasoning text for an alert.

Also Read: Factory Expands Microsoft Collaboration for Autonomous Software Engineering

pgEdge tested the AI DBA Workbench running the Jev decision model against Claude Haiku 4.5, across 30 days of monitoring data. The Jev results agreed with those generated with Haiku 89% of the time, and in no case did Jev suppress an alert that Haiku had classified as such. Furthermore, the cost per 1,000 decisions with Jev was $0.04 versus $1.93 with Haiku, an almost 50 times cost improvement. Latency was 200 milliseconds with Jev versus three seconds for Haiku, 15 times faster.

The significant cost improvement available with decision models means that organizations that were reluctant to adopt agentic Postgres database monitoring because of token costs can move forward without the risk of incurring high AI bills to monitor their estate.

Beyond Jev, decision models supported by the AI DBA Workbench include OpenAI models accessed via the OpenAI Decisions API, with successful testing using GPT-6 Luna. According to OpenAI, the Decisions API returns results up to 10 times faster than the Responses API on the same model.

“As someone who has been developing Postgres administration tools for more than 20 years, I’m extremely impressed with how much more effective decision models are for managing and monitoring key Postgres metrics,” said Dave Page, CTO of pgEdge. “Adding the ability to fall back to a full LLM for reasoning about specific alerts is a very powerful combination and applies the right model to the right task.”

Subscribe

- Never miss a story with notifications


    Latest stories