What's new in GitLab 18.4

Sep 23, 2025
Past release

New CI/CD workflows, a library of specialized agents, GitLab Runner improvements, richer Knowledge Graph for navigating complex codebases, enhanced AI governance, and much more.

AI-native development with automation and insight

With GitLab 18.4, we're evolving the way humans and AI work together with custom agents, code-aware accuracy, and automated pipeline fixes to keep developers in flow.

Read CEO's blog

AI Catalog (Beta)

Manage

Plan

Code

AI

A shared library of reusable and specialized agents, allowing developers to:

  • Create custom agents with defined behaviors and tools.

  • Share agents project-wide or across the organization.

  • Test agents before rollout to ensure predictable performance.

  • Build common use cases like security scanning, docs drafting, or deployment validation.

Enables AI agents to act as teammates, allowing developers to:

  • Start fresh conversations or resume previous ones with custom agents.

  • Run agents synchronously or asynchronously.

  • Access session views with logs, user info, and tool metadata for visibility.

Developers and AI agents use the Knowledge Graph to accelerate large codebase navigation and quickly answer complex questions. Developers can:

  • Utilize real-time indexing to map files, routes, and references across projects.

  • Use go-to-definition, reference tracking, and in-chat search.

  • Ask precise questions like "show me all route files" or "what does this change impact?"

  • Accelerates onboarding, deep research, and confident refactors.

New flow keeps software development pipelines functional by balancing technical fixes and business priorities. To keep developers in the flow state, it is designed to:

  • Detect and prioritize failures based on business importance.

  • Perform root-cause analysis across logs, dependencies, and recent changes.

  • Suggest and apply fixes aligned with deadlines and priorities.

  • Automatically create merge requests with business context.

Ensure consistent and compliant AI use across features and namespaces. GitLab 18.4 allows teams to:

  • Set model defaults at org or feature level.

  • Apply consistent preferences across namespaces.

  • Support GPT-OSS and GPT-5.

  • Note: Model selection is not available for gitlab.com customers, and GPT models are not supported on gitlab.com.

When group-level model selection is not active, developers can choose their preferred model in Agentic Chat. It allows for:

  • A dropdown selection in Agentic Chat to easily switch models.

  • The selection to persist between conversations.

Helps developers protect sensitive information by controlling the context shared with AI models. It makes it possible to:

  • Exclude specific files (e.g., secrets, proprietary algorithms).

  • Apply path-based rules by directory or file type.

  • Configure at project level with audit visibility.

New CI/CD workflows

Plan

Deploy

Operate

Developers can accelerate their workflows even further with new workflows to:

  • Simulate pipelines against any branch to test and validate changes before commit.

  • Utilize CI/CD job tokens to authenticate Git push requests with fine-grained permissions.

New security capabilities

Secure

Analyze

Operate

In GitLab 18.4, new security capabilities allow developers to:

  • Speed up secret detection scans and reduce noise by excluding low-signal files.

  • Quickly trace original pipeline IDs for resolved vulnerabilities in case they reappear.

Additional flexibility for developers with newly added models for Duo Enterprise self-hosted deployments:

  • GPT-5 on Azure OpenAI.

  • GPT-OSS 20B/120B via vLLM and Azure.

GitLab Runner 18.4

Deploy

Operate

Developers can run pipelines with increased reliability thanks to these improvements in GitLab Runner:

  • FIPS startup fixes.

  • New fastzip flag support.

  • Improved long-polling in Kubernetes.

DevOps teams can now deploy GitLab Dedicated in more regions with enterprise-grade availability and disaster recovery (DR) thanks to following improvements in 18.4:

  • io2-backed storage and disaster recovery.

  • Availability of all AWS regions in Switchboard.