GitLab Inc. released GitLab 19.4, introducing a set of agentic automation features aimed at expanding AI-assisted workflows across engineering teams, according to a company statement.

A new /goal command in GitLab Duo CLI, now in public beta, allows developers to assign an open-ended objective to an automated flow that runs locally. A separate model verifies progress at each step, and the developer can stop, revise, or restart the run at any point.

The GitLab Duo Agent Platform now includes three GitLab-hosted open weight models — Kimi K3, MiniMax M3, and GLM 5.3 — alongside existing frontier models. GitLab states the new models offer up to four times more calls per GitLab Credit compared to comparable frontier models.

New tools for the GitLab Model Context Protocol (MCP) server, also in public beta, allow agents operating in external MCP clients to automate tasks across CI/CD pipelines, merge requests, work items, vulnerabilities, and projects. Write and delete tools default to an approval checkpoint, while read-only tools are set to proceed without interruption.

GitLab Credits usage visibility is now generally available, giving platform owners per-user consumption data and usage exports down to the billable event. Developers can also view their own usage for the first time.

Additional features in the release include an experimental GitLab Duo Agent Platform integration for Slack, available to Premium and Ultimate customers, which allows users to mention @GitLab in a thread to search GitLab, open issues, or retrieve project information. A redesigned session details panel and general availability of model selection for the Developer Flow were also included.

"This release takes agentic automation from something individual developers use to something an organization can scale at speed and under the controls already in place," said Manav Khurana, chief product and marketing officer at GitLab.