In GitLab 18.4, teams can create their own AI agents, leverage the Knowledge Graph context, and automatically fix errors in pipelines, all of which can help developers stay focused and work in the flow.
In GitLab 18.4, teams can create their own AI agents, leverage the Knowledge Graph context, and automatically fix errors in pipelines, all of which can help developers stay focused and work in the flow.
GitLab reports this on its blog.
AI Catalog is currently an experimental feature. It’s a centralized library where teams can create, share, and collaborate on customized agents across the organization. Every team has a unique way of getting things done.
GitLab provided several examples of directory capabilities:
“These agents are not just standalone tools, they integrate into the GitLab workflow and help speed up tasks without changing your usual processes,” the company added.
GitLab Knowledge Graph, now in beta, is changing the way developers and agents understand and navigate complex codebases, the company says. It creates an interconnected map of the entire project, connecting files, routes, and links across all stages of the software development lifecycle.
Using tools like “jump to definition,” codebase search, and link tracking via chat requests, developers can ask precise questions, such as “show me all route files” or “what else will this change affect?”
“In this release, Knowledge Graph focuses on indexing local code, where the gkg CLI turns your codebase into a dynamic, embedded graph database for RAG. You can install it with a simple one-line script, parse local repositories, and connect via MCP to query your workspace,” GitLab explained.
The new Fix Failed Pipelines Flow feature is designed with business needs in mind — understanding that a pipeline failure in the main client application is a critical issue, while an error in the experimental branch can wait.
The function analyzes errors taking into account this context, finds the root of the problem, determines the root cause, and even suggests fixes by creating a merge request.
“It’s like having a senior engineer who not only finds a problem, but also understands its urgency and helps solve it,” the company notes.
GitLab also provides granular control over the selection of large language models that power AI features, ensuring you can work with providers you trust.
Context Exclusion allows you to draw red lines around sensitive parts of your project: for example, files containing secrets, passwords, or core proprietary algorithms. That is, you can simply tell the AI something like: “Access here is denied. Always.”
— and tell the AI: “Access denied. Always.” This gives you the confidence to use AI capabilities without compromising security.
“GitLab Duo Context Exclusion helps protect sensitive data while accelerating development with AI agents,” the company added.
