GitLab 18.3 starts transformation towards comprehensive AI integration

With the release of version 18.3, the GitLab team announced that the development platform is on its way to becoming the "world's first AI-native platform for software engineering". To this end, innovations are to take place on several levels. The new release not only includes enhanced AI features, but also updates for the Web IDE and for project migration using the Direct Transfer function. A total of 38 changes are included in the release.

A dedicated blog entry describes GitLab's transformation to AI-supported orchestration as part of the GitLab Duo Agent Platform. Accordingly, the unified data platform will receive a knowledge graph that is optimized for agentic access and designed to index code and link it to all other available unstructured data. This should improve reasoning and inference. In addition, an orchestration layer will be added to the control plane, which will be used to provide GitLab tools, agents, and flows via the Model Context Protocol (MCP). In addition, the GitLab team is working on extended functions for AI agents and agentic flows.

AI beta features: Visual Studio connection and hybrid model selection

The Duo Agent Platform in Visual Studio has reached beta status and is available to all GitLab users. It allows access to the platform's AI-powered features directly from Visual Studio: agentic chat – for example, for creating and editing files or searching the codebase – and agentic flows designed for more complex tasks such as comprehensive planning and the ability to turn ideas into architecture and code.

Paying customers using a self-managed GitLab instance can now also use their own language model in GitLab Duo Self-Hosted, also a beta feature. The same user group will have access to another beta feature, the ability to use a mix of models from GitLab AI vendors and privately configured, self-hosted models in GitLab Duo Self-Hosted. This hybrid model selection is designed to help admins balance security and scaling requirements.

Migration via direct transfer and web IDE updates

In addition to the AI innovations, migration via Direct Transfer is now generally available for all GitLab users. This allows GitLab groups and projects to be transferred between GitLab instances, either via the GitLab UI or the REST API. In contrast to migration using an export file, Direct Transfer should offer several advantages: It should work more reliably for large projects, support migrations with larger version gaps between source and target instances and provide better insights into the migration process and its results.

In the Web IDE, additional functions are now available in the Source Control panel. Among other things, branches can now be created and deleted there and changes can be forced directly from the interface (force push).

The GitLab blog emphasizes these and other highlights, including fine-grained permissions for CI/CD job tokens, a user-defined admin role and support for the AWS Secrets Manager for GitLab CI/CD.

(mai)