🤖GitLab 19.4 Adds Goal-Driven Mode to Duo CLI
GitLab's new goal-driven mode for Duo CLI
TL;DR
GitLab 19.4 introduces goal-driven mode in Duo CLI, allowing developers to set objectives and let the tool work until completion or iteration limit. The new /goal command supports hosted models and provides cost savings.
GitLab 19.4 introduces goal-driven mode in Duo CLI, letting developers set objectives and let the tool work until completion or iteration limit. This mode supports hosted models like Kimi K3, MiniMax M3, and GLM 5.3, offering up to four times more calls per GitLab Credit. Developers can now focus on high-level goals while the tool handles the details, reducing the need for constant supervision. The release also includes enhanced cost attribution features, with credit usage exports and per-user caps, helping teams better manage their budgets. GitLab Duo CLI 9.17.0 or later is required for the /goal command, available to Premium and Ultimate customers.
Key Points
GitLab 19.4 introduces goal-driven mode in Duo CLI, with /goal command available to Premium and Ultimate customers on GitLab.com, GitLab Self-Managed, and GitLab Dedicated, requiring Duo CLI 9.17.0 or later.
Goal-driven mode supports hosted models like Kimi K3, MiniMax M3, and GLM 5.3, offering up to four times more calls per GitLab Credit.
GitLab Duo Agent Platform gains support for hosted open-weight models, providing up to four times as many calls per GitLab Credit as comparable frontier models.
Credit usage exports now include a row for each billable event, helping teams manage costs more effectively.
Developers can see their own consumption, while billing managers receive warnings when on-demand use reaches 50% or 80% of a monthly cap.
Why It Matters
If you're using GitLab's Duo CLI for development tasks, the new goal-driven mode flips the script on how you manage your projects. For instance, setting a goal like 'deploy the latest version of the app' can now be automated, reducing the need for manual intervention. However, the cost savings come with a catch: the premium models are only cost-effective above ~100K read IOPS per instance, so smaller teams might not see the same benefits.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,565 builders reading daily.