GitLab 19.4 Enhances AI Cost Controls, Adds Goal-Driven Agents

GitLab has launched version 19.4 of its DevSecOps platform, introducing goal-driven agents and enhanced cost controls for AI spending. The update expands GitLab Duo CLI’s capabilities, adding a /goal command that accepts open-ended objectives and operates within existing project permissions.

Goal-Driven Agents in Action

The /goal command, now in public beta, allows developers to set objectives without predefined flows. A dual-model system ensures tasks are completed accurately: one model executes changes, while another verifies progress against the goal. Developers retain control, able to halt, redirect, or modify tasks as needed. This approach keeps automation within established project boundaries, ensuring scalability and security.

GitLab’s verification process relies on a second model to assess whether results align with natural-language goals, though it does not formally verify code. The system can report unresolved tasks and includes an iteration limit to prevent indefinite operation. Clear acceptance criteria remain essential, even as developers step back from direct oversight.

Cost Management for Automated Agents

As agents operate automatically, cost control becomes critical. ReleaseTEAM, a DevOps consultancy, highlights the risk of unchecked automation, where flows triggered by pipeline failures or merge requests could run frequently. They advise setting usage limits before connecting flows to high-volume triggers and recommend pilot testing to measure consumption.

Administrators can set default per-user caps and individual overrides, while developers monitor their consumption. Billing managers receive alerts when usage reaches predefined thresholds, ensuring better cost management.

Enhanced Model Support and Cost Efficiency

GitLab Duo Agent Platform now supports three hosted open-weight models: Kimi K3, MiniMax M3, and GLM 5.3. These models offer up to four times more calls per GitLab Credit compared to similar frontier models. Group owners can designate default models for specific features and restrict team access, optimizing costs by reserving more expensive models for critical tasks.

Expanding MCP Tools and Access Controls

The update also expands the Model Context Protocol (MCP) server tools, enabling agents in third-party clients to perform tasks like triggering pipelines, reviewing merge requests, and managing vulnerabilities. Read-only tools are set to “Always Allow” by default, while write and delete operations require explicit approval. Administrators can customize these rules at the group or project level, ensuring granular control over agent actions.

Most MCP tools are available across Free, Premium, and Ultimate tiers, though vulnerability management is exclusive to Ultimate. This approach aligns with industry trends, as platforms like Android Studio and GitHub Copilot also integrate MCP with enterprise-level access controls.

Cost Attribution and Administrative Insights

With automation comes the need for tighter cost management. GitLab 19.4 introduces detailed credit usage exports, breaking down billable events by user, project, session, flow type, credits, and token counts. A new settings page allows owners to set default per-user caps and individual overrides without relying on the GraphQL API. Developers can monitor their consumption, while billing managers receive alerts when usage reaches 50% or 80% of monthly limits.

These enhancements build on previous updates, such as GitLab 19.2, which introduced agents for dependency remediation and security reviews. Earlier versions, like 18.10 and 18.11, added flat-rate code reviews, free-tier access to Duo Agent Platform, and monthly credit limits. Version 19.4 extends these capabilities to longer-running tasks, providing administrators with greater visibility into associated costs.

GitLab 19.4 was released on September 17, 2026. The /goal command is available to Premium and Ultimate customers on GitLab.com, GitLab Self-Managed, and GitLab Dedicated, requiring GitLab 19.3 or later with Duo CLI 9.17.0 or later. While the command and expanded MCP tools remain in public beta, the new credit usage features are generally available.