pgEdge AI DBA Workbench is a four-pronged Postgres shared data store: collector, server, alert, and React client that provides dashboards and a chat panel where Ellie lives. Ellie is an agent loop that calls any LLM of your choice (Claude, ChatGPT, Ollama, or anything OpenAI compatible) with a fixed set of database tool calls. The model never queries Postgres directly, which is the point. Anomaly detection uses three stages: z-score baselines, pgvector matching against historical patterns, and LLM scaling for residual cases. Source on GitHub under the PostgreSQL license.
Workbench is four Go services and TypeScript: A collection daemon connects to each monitored instance and pulls a bunch of metric probes pg_stat_statements, pg_stat_activityreplication status, and OS-level resource utilization. Each probe writes to a dedicated time series table to which it is pointed connection_id again collection_timewith adjustable cadence per scope.
The server exposes JSON-RPC 2.0 MCP and REST endpoints via HTTPS and consumers all interactions between LLM and your databases. Alert uses anomaly detection and threshold testing. The client is React SPA.
Ellie is a client-side agent loop that communicates with a configurable LLM backend through a server. We currently support Claude, ChatGPT, and Ollama local models, as well as other providers with a compatible OpenAI API such as LM Studio or EXO. Model selection is more important than people often think because LLM never asks your database directly. Instead, it returns the tool calls that the client sends to them /api/v1/tools/executeand the server uses them under the caller’s host token against connection pools for each token so that no two users share a state.
The same agent loop calls the AI analysis button on all chart, alert, and dashboard panels. Clicking it runs a one-shot version of the loop, populated with panel data, time range, server context, and timeline events in that window; the result comes down to a mode you can read or download as Markdown, no discussion needed.
GitLab announced the offering of GitLab hosted on Google Cloud
The offering is delivered through GitLab certified managed service providers, allowing organizations to deploy GitLab in a secure, independent environment while maintaining greater control over infrastructure and compliance requirements.
“We’re in the age of agent engineering, and it’s never been easier and faster to code. That speed comes with a level of chaos that businesses can’t afford,” said Manav Khurana, chief product and marketing officer at GitLab. “Reliable events, unexpected spending, and compliance exposure to agent actions slows down organizations when they move quickly without the controls they need. GitLab is a platform where businesses already build and deploy software, which means we sit at the intersection of all the code that touches human and agent workflows, pipelines, or production. With these new capabilities, Gigentic for dynamic infrastructure is opening the dominant infrastructure, GitLab evolving into an infrastructure-driven, enterprise-grade software delivery.”
Key news points are:
- Next Generation Source Code Management, now in private beta, replaces repository clones with structured API access to project intelligence, delivering 50x faster workload execution for each agent.
- GitLab Orbit, now in public beta, is a context graph for the entire software lifecycle that enables agents to deliver 11x faster responses that require up to 4.5x fewer tokens.
- Agent Management, now in private beta, adds new AI inspection and control capabilities to meet compliance requirements.
- GitLab Flex is a one-year commitment that includes field seats, GitLab Credits, and new qualifying capabilities as they become available, with monthly reservations that can be reconfigured without contract amendment.
- GitLab and Google Cloud’s go-to-market teams work with certified service providers, including Beyond and Digital Future, to help businesses migrate to reliable, trusted DevSecOps architectures on GitLab and Google Cloud. Also, the latest versions of Google’s Gemini models, including Gemini 3.5, are now available on the GitLab Duo Agent Platform, and Google’s Gemma models, including Gemma 4, are now available for GitLab Duo Self-Hosting customers.
