GitLab 18.3: Expanding AI orchestration in software engineering

Today, GitLab is a comprehensive DevSecOps platform, unifying every stage of the software lifecycle. Building on that foundation, we're on a journey toward becoming the world's first AI-native platform for software engineering. At GitLab, we believe the future of software engineering is an inherently human and AI collaboration, and we want to bring the very best AI capabilities to every GitLab user.

This transformation is happening at three distinct layers that go beyond what other AI dev tools are doing:

First, we are a system of record. Our unified data platform holds your most valuable digital assets. This includes your source code and intellectual property, as well as a wealth of unstructured data spanning project plans, bug backlogs, CI/CD configurations, deployment histories, security reports, and compliance data. This creates a treasure trove of contextual data that remains securely within your GitLab environment, unavailable to generic agents or large language models.

Second, we act as your software control plane. We orchestrate your most critical business processes through Git repositories, REST APIs, and webhook-based interfaces that power your end-to-end software delivery. Many of our customers consider this a tier-0 dependency that their critical business processes rely on daily.

Third, we deliver a powerful user experience. We deliver an integrated interface that helps eliminate the costly context-switching that slows down most engineering teams. With complete lifecycle visibility and collaboration tools in one platform, over 50 million registered users and our vast community depend on GitLab to get their work done. This expertise positions GitLab uniquely to pioneer intuitive human-to-AI collaboration that amplifies team productivity while preserving the workflows that our users know and trust.

Extending our platform with AI natively integrated at every layer

GitLab Duo Agent Platform integrates and extends all three of these layers. It is designed for extensibility and interoperability, enabling customers and partners to build solutions that create even more value. Our open platform approach emphasizes seamless connectivity with external AI tools and systems while being deeply integrated into our existing stack at all three layers.

• First, we're extending our unified data platform with a Knowledge Graph, which indexes and stitches together code with all of the rest of your unstructured data, specifically optimized for agentic access. AI thrives on context, and we believe this will not only accelerate reasoning and inference by agents but also deliver lower-cost and higher-quality agentic outcomes.
• Second, we're adding an important Orchestration Layer to our existing Control Plane in three distinct parts: enabling agents and flows to register as subscribers for GitLab SDLC events, building a new orchestration engine that allows for purpose-built, multi-agent flows, and exposing GitLab tools, agents, and flows via MCP and standard protocols for unparalleled interoperability.
• Finally, we're extending the GitLab experience to deliver first-class agents and agent flows across the entire software development lifecycle. You'll be able to assign async tasks to agents, @ mention them in comments, and create custom agents with context specific to your workflows — but more importantly, GitLab is shipping native agents for every stage of development while unlocking a rich ecosystem of third-party agents. This creates true human-to-AI collaboration where agents become as natural to work with as your human teammates.

Watch this video to see what's coming in 18.3 and beyond, or read on.

What's new in GitLab 18.3

With 18.2, we introduced specialized AI agents that work alongside developers across the software development lifecycle, plus our Software Development Flow — a powerful feature that gives users the ability to orchestrate multiple agents to plan, implement, and test code changes end-to-end.

GitLab 18.3 introduces expanded integrations and interoperability, more Flows, and enhanced context awareness across the entire software development lifecycle.

Expanded integrations and interoperability

We're delivering comprehensive AI extensibility through both first-party GitLab agents and a rich ecosystem of third-party agents, all with full access to project context and data. This approach maintains native GitLab workflows and governance while providing the flexibility to choose preferred tools through highly integrated orchestration between these agents and GitLab's core platform. Teams gain enhanced AI functionality while preserving key integration, oversight, and user experience benefits.

• MCP server - Universal AI integration: GitLab's MCP (Model Context Protocol) server enables AI systems to securely integrate directly with your GitLab projects and development processes. This standardized interface eliminates custom integration overhead and allows your AI tools — including Cursor — to work intelligently within your existing GitLab environment. See our docs for a full list of tools included with 18.3. This is only the start; additional tools are planned for 18.4.

“Bringing GitLab workflows directly into Cursor is a critical step in reducing friction for developers. By minimizing the need for context switching, teams can check issue status, review merge requests, and monitor pipeline results without ever leaving their coding environment. This integration is a natural fit for our shared customers, and we look forward to a long-term partnership with GitLab to continue enhancing developer productivity.”

- Ricky Doar, VP of Field Engineering at Cursor

“GitLab's MCP server and CLI agent support create powerful new ways for Amazon Q to integrate with development workflows. Amazon Q Developer can now connect directly through GitLab's remote MCP interface, while teams can delegate development tasks by simply @ mentioning Amazon Q CLI in issues and merge requests. The robust security and governance capabilities built into these integrations give enterprises the confidence to leverage AI coding tools while preserving their development standards. Our partnership with GitLab demonstrates AWS' ongoing commitment to expanding our AI ecosystem and making intelligent development tools accessible wherever developers work."

- Deepak Singh, Vice President of Developer Agents and Experiences at AWS

• CLI agent support for Claude Code, Codex, Amazon Q, Google Gemini, and opencode (Bring Your Own Key): 18.3 introduces integrations that enable teams to delegate routine development work by @ mentioning their agents directly in issues or merge requests. When developers mention these AI assistants, they automatically read the surrounding context and repository code, then respond to the user's comment with either ready-to-review code changes or inline comments. These integrations require you to bring your own API key for the respective AI providers and keep all interactions natively within GitLab's interface while maintaining proper permissions and audit trails.

  Note: Third-party agents is a GitLab Premium Beta feature and only available to GitLab Duo Enterprise customers for evaluation.

“Bringing Claude Code directly into GitLab puts AI assistance where millions of developers already collaborate and ship code daily. The ability to mention Claude directly in issues and merge requests removes friction while maintaining quality with human oversight and review processes. This update brings Claude Code's capabilities to more places where teams work, making AI a natural part of their developer workflow.”

- Cat Wu, Claude Code Product Lead, Anthropic

“With GitLab's new agent integration in 18.3 you can use opencode within your existing workflows. You can @mention opencode in an issue or merge request and it'll run your agent right in your CI pipeline. This ability to configure and run opencode the way you want is the type of integration we know the open source community really values.”

- Jay V., CEO, opencode

• Agentic Chat support for Visual Studio IDE and GitLab UI available to all Premium and Ultimate customers: With 18.3, you no longer need to context-switch between tools to access GitLab's full development lifecycle data. Our enhanced integrations bring the complete power of GitLab Duo into the GitLab UI as well as IDEs — expanding support from JetBrains and VS Code to now include Visual Studio. This helps developers stay in flow while accessing rich project context, deployment history, and team collaboration data directly within their preferred environment.
• Expanded AI model support: GitLab Duo Self-Hosted now supports additional AI models, giving teams more flexibility in their AI-supported development workflows. You can now deploy open source OpenAI GPT models (20B and 120B parameters) through vLLM on your datacenter hardware, or through cloud services like Azure OpenAI and AWS Bedrock in your private cloud. Additionally, Anthropic's Claude 4 is available on AWS Bedrock

New automated development flows

GitLab Flows coordinate multiple AI agents with pre-built instructions to autonomously handle those time-consuming, mundane tasks so developers can focus on the work that matters most.

GitLab 18.3 comes with two new Flows:

• Issue to MR Flow enabling automated code generation from concept to completion in minutes: This Flow automatically converts issues into actionable merge requests (MRs) by coordinating agents to analyze requirements, prepare comprehensive implementation plans, and generate production-grade code that's ready for review — helping you turn ideas into reviewable implementations in minutes, not hours.

• Convert CI File Flow built for seamless migration intelligence: Our Convert CI File Flow streamlines migration workflows by having agents analyze existing CI/CD configurations and intelligently convert them to GitLab CI format with full pipeline compatibility. This helps eliminate the manual effort and potential errors of rewriting CI configurations from scratch, enabling teams to migrate entire deployment pipelines with confidence. 18.3 includes support for Jenkins migrations. Additional support is planned for future releases.

Intelligent code and search

AI point solutions typically operate with limited visibility into isolated code snippets, but GitLab's Knowledge Graph provides agents with environment context to help inform faster and more intelligent responses.

• Knowledge Graph for real-time code intelligence: With 18.3, GitLab's Knowledge Graph now delivers real-time code indexing to enable faster code searches, delivering more accurate and contextual results. By understanding the relationships between files, dependencies, and development patterns across your entire codebase, our agents are designed to provide insights that would take human developers hours to uncover — and this is just the first step in unlocking the powerful capabilities that are planned for Knowledge Graph.

Enterprise governance

AI transparency and organizational control are critical challenges that can hold teams back from fully adopting AI-powered development tools, with 85% of executives agreeing that agentic AI will create unprecedented security challenges.

These new features in 18.3 help address concerns around data governance, compliance requirements, and the need for visibility into AI decision-making processes so organizations can integrate AI within their existing security and policy frameworks.

• Agent Insights for transparency through intelligence: Our built-in agent tracking provides visibility into agent decision-making processes. Users can optimize workflows and follow best practices through transparent activity tracking.

• GitLab Duo Code Review for Self-Hosted: This brings the intelligence of GitLab Duo to organizations with strict data governance requirements by allowing teams to keep sensitive code in controlled environments.
• Hybrid model configurations for flexible AI deployment: GitLab Duo Self-Hosted customers can now use hybrid model configurations, combining self-hosted AI models via their local AI gateway with GitLab's cloud models through GitLab's AI gateway, enabling access to various features.

• Enhanced security with OAuth support: Our MCP server now includes full OAuth 2.0 authentication support, enabling secure connections to protected resources and sensitive development environments. This implementation follows the draft OAuth specification for MCP, handling authorization flows, token management, and dynamic client registration.

Secure by Design platform: Governance that scales

True platform security requires consistent application of governance principles across every layer of the development lifecycle. The same security fundamentals that make AI adoption safe — least-privilege access, centralized policy management, proactive monitoring, and granular permissions — must be embedded throughout the entire SDLC to create a cohesive, defense-in-depth approach.

GitLab 18.3 strengthens the foundational controls that help protect your entire software supply chain with these new updates:

• Custom admin role: Provides granular, purpose-built administrative permissions, replacing blanket admin access with precise, least-privilege controls. Instead of granting blanket administrative privileges that create security risks, organizations can now create specialized roles tailored to specific functions — platform teams managing runners and monitoring, support teams handling user management, and leadership accessing dashboards and usage statistics. With complete role lifecycle management through UI and API, audit logging, and auto-generated documentation, this feature enables true least-privilege administration while helping maintain operational efficiency and improve overall instance security.
• Instance-level compliance framework and security policy management: Organizations can now designate a dedicated compliance group that has the authority to apply standardized frameworks and security policies directly to top-level groups, automatically cascading enforcement to all their subgroups and projects. This centralized approach eliminates the compliance adoption blocker of fragmented policy management while maintaining group autonomy for additional local policies.
• Enhanced violations reporting: Teams now receive immediate notifications when unauthorized changes are made to MR approval rules, framework policies lack proper approvals, or time-based compliance controls are violated. By directly linking violations to specific compliance framework controls, teams get actionable insights that tell them exactly which requirement was breached, turning compliance from a reactive checkbox exercise into a proactive, integrated part of the development and security workflow.
• Fine-grained permissions for CI/CD job tokens: Replaces broad token access with granular, explicit permissions that grant CI/CD jobs access only to specific API endpoints they actually need. Instead of allowing jobs blanket access to project resources, teams can now define precise permissions for deployments, packages, releases, environments, and other critical resources, reducing the attack surface and potential for privilege escalation.
• AWS Secrets Manager integration: Teams using AWS Secrets Manager can now retrieve secrets directly in GitLab CI/CD jobs, simplifying the build and deploy processes. Secrets are accessed by a GitLab Runner using OpenID Connect protocol-based authentication, masked to prevent exposure in job logs, and destroyed after use. This approach eliminates the need to store secrets in variables and integrates cleanly into existing GitLab and AWS-based workflows. Developed in close collaboration with Deutsche Bahn and the AWS Secrets Manager team, this integration reflects our commitment to building solutions alongside customers to solve real-world challenges.

Artifact management: Securing your software supply chain

When artifacts aren't properly governed, small changes can have big consequences. Mutable packages, overwritten container images, and inconsistent rules across tools can trigger production outages, introduce vulnerabilities, and create compliance gaps. For enterprise DevSecOps, secure, centralized artifact management is essential for keeping the software supply chain intact.

Enterprise-grade artifact protection in 18.3

Building on our comprehensive package protection capabilities, GitLab 18.3 adds important new features:

• Conan revisions support: New in 18.3, Conan revisions provide package immutability for C++ developers. When changes are made to a package without changing its version, Conan calculates unique identifiers to track these changes, enabling teams to maintain immutable packages while preserving version clarity.
• Enhanced Container Registry security: Following the successful launch of immutable container tags in 18.2, we're seeing strong enterprise adoption. Once a tag is created that matches an immutable rule, no one — regardless of permission level — can modify that container image, preventing unintended changes to production dependencies.

These enhancements complement our existing protection capabilities for npm, PyPI, Maven, NuGet, Helm charts, and generic packages, enabling platform teams to implement consistent governance across their entire software supply chain — a requirement for organizations building secure internal developer platforms.

Unlike standalone artifact solutions, GitLab's integrated approach eliminates context switching between tools while providing end-to-end traceability from code to deployment, enabling platform teams to implement consistent governance across their entire software delivery pipeline.

Embedded views: Real-time visibility and reports

As GitLab projects grow in complexity, teams find themselves navigating between issues, merge requests, epics, and milestones to maintain visibility into work status. The challenge lies in consolidating this information efficiently while ensuring teams have real-time access to project progress without context switching or breaking their flow. Launching real-time work status visibility in 18.3 GitLab 18.3's embedded views, powered by our powerful GitLab Query Language (GLQL), eliminate context switching by bringing live project data directly into your workflow:

• Dynamic views: Insert live GLQL queries in Markdown code blocks throughout wiki pages, epics, issues, and merge requests that automatically refresh with current project states each time you load the page.
• Contextual personalization: Views automatically adapt using functions like currentUser() and today() to show relevant information for whoever is viewing, without manual configuration.
• Powerful filtering: Filter by 25+ fields, including assignee, author, label, milestone, health status, and creation date.
• Display flexibility: Present data as tables, lists, or numbered lists with customizable field selection, item limits, and sort orders to keep your views focused and actionable

Unlike fragmented project management approaches, we've designed embedded views to maintain your workflow continuity while providing real-time visibility, enabling teams to make informed decisions without losing focus or switching between multiple tools and interfaces.

Learn about the newest features in GitLab 18.3.

Get started today

GitLab 18.3 is available now for GitLab Premium and Ultimate users on GitLab.com and self-managed environments.

GitLab Dedicated customers are now upgraded to 18.2 and will be able to use the features released with GitLab 18.3 next month.

Ready to experience the future of software engineering? Enable beta and experimental features for GitLab Duo and start collaborating with AI agents that understand your complete development context.

New to GitLab? Start your free trial today and discover why the future of software engineering is human and AI collaboration, orchestrated through the world's most comprehensive DevSecOps platform.

This blog post contains “forward-looking statements” within the meaning of Section 27A of the Securities Act of 1933, as amended, and Section 21E of the Securities Exchange Act of 1934. Although we believe that the expectations reflected in the forward-looking statements contained in this blog post are reasonable, they are subject to known and unknown risks, uncertainties, assumptions and other factors that may cause actual results or outcomes to be materially different from any future results or outcomes expressed or implied by the forward-looking statements.

Further information on risks, uncertainties, and other factors that could cause actual outcomes and results to differ materially from those included in or contemplated by the forward-looking statements contained in this blog post are included under the caption “Risk Factors” and elsewhere in the filings and reports we make with the Securities and Exchange Commission. We do not undertake any obligation to update or release any revisions to any forward-looking statement or to report any events or circumstances after the date of this blog post or to reflect the occurrence of unanticipated events, except as required by law.