GitLab has rolled out version 18.4 of its DevSecOps platform, introducing a series of AI-driven features aimed at improving developer productivity, security, and control. The update builds on earlier investments in AI collaboration and extends GitLab’s vision of seamless human-AI workflows.
Advancing AI-driven development
With GitLab 18.4, teams can now create and share their own AI agents, navigate complex codebases through the new Knowledge Graph, and fine-tune how AI models are used across projects. The release also enhances GitLab’s built-in assistant, GitLab Duo, giving users more control over which large language models power specific features.
“GitLab 18.4 aims to make the day-to-day experience of developers smoother, smarter, and more secure,” the company said, adding that the update helps teams maintain flow while balancing speed and governance.
One of the key additions is the GitLab Duo AI Catalog, a central space for organisations to build, share and collaborate on custom agents. These reusable agents are designed to streamline workflows and support greater efficiency across development teams.
Smarter navigation and pipeline reliability
Another highlight is the GitLab Knowledge Graph, which provides a connected map of an entire project. By linking files, routes, and references, it helps developers and AI agents understand complex codebases and locate information faster.
To support reliability in software delivery, GitLab has introduced the Fix Failed Pipelines Flow with business awareness. This feature combines technical error detection with organisational context, automatically analysing root causes and creating fixes that align with business priorities. It can generate merge requests and integrate review processes while adding transparency about the reasoning behind changes.
Agentic Chat, GitLab’s conversational interface, has also been upgraded. Developers can now see richer details about AI-driven actions, including job logs and tool metadata, helping teams understand and control how AI operates within their projects.
Security and customisation improvements
The release also adds Context Exclusion, giving teams more control over AI access to sensitive data. Project-level settings can now exclude certain files or paths from being analysed, allowing developers to use AI tools while protecting confidential information.
GitLab Duo model selection has moved to general availability, enabling customers to choose the large language models that best meet their performance and compliance needs. Teams can apply a single model across an organisation or tailor selections to individual features.
GitLab said the 18.4 release is part of its continued focus on integrating AI directly into the software development lifecycle. By expanding AI-driven navigation, decision-making, and automation, the company aims to help teams deliver code faster and with greater confidence, while maintaining the flexibility and governance enterprises require.
