GitLab 19.4 Adds Agentic AI Automation With Cost Controls

GitLab has released version 19.4 of its DevSecOps platform with expanded agentic AI capabilities designed to help organizations scale automation while maintaining cost control and governance.

The release, announced by GitLab Inc., introduces three major features aimed at moving AI agents from individual developer tools to enterprise-wide automation: a new command-line interface for delegating complete objectives to AI agents, access to lower-cost open-weight models, and granular usage tracking for platform administrators.

Automating complete objectives, not just tasks

The centerpiece of the release is the /goal command in GitLab Duo CLI, now in public beta. Unlike previous AI coding assistants that handle discrete tasks, this feature lets developers hand off an entire objective to an agentic workflow that runs locally, implements the work, and verifies its own output at each step.

A separate model checks whether the stated goal has been met or if iteration limits have been reached, giving developers an independent verification layer. The workflow operates under existing project permissions and governance rules, eliminating the need for separate security models when deploying AI agents.

Lower-cost model options expand flexibility

GitLab 19.4 adds three GitLab-hosted open-weight models to its Duo Agent Platform: Kimi K3, MiniMax M3, and GLM 5.3. These models join existing frontier model options and deliver up to four times more API calls per GitLab Credit compared to some frontier alternatives, according to the company.

Group administrators can set default models for specific features and curate which models teams can access, with settings cascading to child groups and projects. GitLab vets each model against internal performance standards and evaluates hosting vendors through its third-party risk management process.

New governance tools for enterprise deployment

The release includes new Model Context Protocol (MCP) server tools, now in public beta, that let AI agents automate complete workflows across CI/CD pipelines, merge requests, work items, vulnerabilities, and projects. Administrators govern these tools through the same rules already configured for GitLab Duo Agent Platform.

Read-only tools default to "Always Allow" for routine operations, while write and delete operations default to "Always Ask," creating approval checkpoints before agents make changes.

GitLab Credits usage visibility, now generally available, gives platform owners per-user consumption data and billable event exports. Developers can also view their own credit usage for the first time, enabling teams to self-manage automation capacity.

Why it matters

As enterprises move beyond experimental AI coding assistants to production-scale agentic automation, governance and cost predictability become critical adoption barriers. GitLab's approach embeds AI agent controls within existing DevSecOps permissions and audit trails rather than creating parallel systems. The combination of lower-cost model options and granular usage tracking addresses two key concerns CIOs face when authorizing broader AI deployment: budget predictability and accountability for how automation capacity is consumed across engineering organizations.

"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.

Additional features in 19.4 include GitLab Duo Agent Platform integration with Slack as an experiment, generally available model selection for the Developer Flow, and community-contributed MCP server tools.

These details were first reported by GitLab Inc. in a company announcement.