GitLab Inc. released version 19.4 of its DevSecOps platform on September 17, 2026, introducing advanced agentic automation tools and cost management controls designed to help engineering teams scale AI workflows securely. According to company announcements, the update brings the new `/goal` command in the GitLab Duo CLI to public beta, alongside Model Context Protocol (MCP) server tools and GitLab-hosted open weight models aimed at reducing operational expenses.
The /goal Command Enables Autonomous Local Execution
According to reports from MarketScreener, the new `/goal` slash command in GitLab Duo CLI allows developers to delegate open-ended objectives rather than supervising individual tasks. Operating in public beta, the workflow runs locally while a separate model verifies progress at each step. The system evaluates whether the stated objective is met or an iteration limit is reached, allowing developers to stop a run, revise the goal, and restart.
Chief Product and Marketing Officer Manav Khurana noted that the update transitions agentic automation from individual use cases to scalable organizational deployments.
Open Weight Models Add Cost-Efficiency Levers
To address computing expenses, the GitLab Duo Agent Platform now incorporates three GitLab-hosted open weight models: Kimi K3, MiniMax M3, and GLM 5.3. These models operate alongside existing frontier models to provide teams with flexible choices based on workload requirements.
According to official release details, these hosted open weight models deliver up to four times more calls per GitLab Credit compared to standard frontier options. Group owners retain governance over these choices, maintaining the ability to set default models for specific features and curate available options across child groups and projects.
Model Context Protocol Integration Expands Automation Reach
Version 19.4 also introduces public beta tools for the GitLab MCP server. These tools allow agents operating within any Model Context Protocol client to execute tasks end-to-end across GitLab without manual intervention between applications.
- Trigger CI/CD pipelines and read traces from failed jobs.
- Manage merge requests from opening through review to final merge.
- Search and update work items.
- Triage and manage software vulnerabilities.
Administrators govern these interactions using pre-existing group and project settings for the GitLab Duo Agent Platform. Read-only tools default to “Always Allow,” while write and delete operations default to “Always Ask” to provide mandatory review checkpoints.
Governance and Usage Visibility
Security and tracking form core components of the 19.4 release. The platform applies existing code permissions directly to agents, ensuring that every credit expenditure traces back to a specific user account without requiring a separate audit trail.
Additionally, GitLab Credits usage visibility has reached general availability. This feature gives platform owners direct tools to measure, control, and attribute agentic automation capacity across individual departments.
Related reading