GitLab 18.9 Released with Self-Hosted AI Models
GitLab has released version 18.9, which introduces support for running AI models locally within self-hosted deployments. The new functionality allows development teams to build and test AI features without relying on third-party cloud services. This provides greater control over data privacy and enables deeper integration of AI into the software development lifecycle, including code reviews and CI/CD pipelines.
- This functionality is delivered through a component called the GitLab Duo Agent Platform, which becomes generally available for self-managed customers with a cloud license in version 18.9. - The platform's architecture includes a self-hosted AI Gateway, which allows all data and requests to remain within the user's own environment, avoiding external API calls to third-party services. - Administrators can connect to various large language models (LLMs), including Anthropic's Claude and OpenAI's GPT models, through services like AWS Bedrock and Azure OpenAI. - GitLab's AI assistant, GitLab Duo, leverages these models to provide features like Code Suggestions in over 20 languages, automatic test generation, and AI-assisted code reviews directly within the development workflow. - A core privacy principle of GitLab Duo is that customers' proprietary code and other data are not used to train the underlying AI models. - For customers with online licenses, billing for the self-hosted AI platform shifts to a usage-based model with this release. - The AI can be integrated directly into CI/CD pipelines to help with tasks like explaining and resolving security vulnerabilities or analyzing the root cause of failed pipeline jobs. - In addition to self-hosting, GitLab provides fine-grained security controls, such as the ability to exclude specific project files or paths from being accessed by AI features to protect sensitive information.
