GitLab 18.9 Introduces Self-Hosted AI Models
GitLab's latest release, version 18.9, now supports self-hosted AI models within its DevSecOps platform. The feature is targeted at organizations with strict data residency, compliance, or privacy requirements, allowing them to use AI-powered code review and agentic workflows without sending data to third-party providers. The update also expands the agentic capabilities of the GitLab Duo assistant.
- The GitLab Duo Agent Platform, which powers these new AI capabilities, reached general availability in the 18.8 release in January 2026, setting the stage for this expansion into self-hosted models. This platform allows for the orchestration of AI agents across the entire software development lifecycle. - Organizations can use a "Bring Your Own Model" (BYOM) approach, connecting to models from providers like Mistral, Meta, Anthropic, and OpenAI. These can be hosted on platforms such as vLLM, AWS Bedrock, and Azure OpenAI. - A hybrid deployment option is also available, allowing teams to use their own self-hosted models for some features while utilizing GitLab-managed models for others. - GitLab's journey into AI-assisted features was significantly accelerated by its acquisition of UnReview in June 2021. UnReview's machine learning technology for automating code reviewer recommendations formed a foundational part of GitLab's AI strategy. - The 18.9 release also introduces Agentic SAST Vulnerability Resolution in beta, which uses AI agents to autonomously analyze, explain, and create merge requests to fix security vulnerabilities. - This version makes "Security Attributes" generally available, allowing security teams to categorize projects by business context, such as business impact and internet exposure, to better prioritize risks. - While GitLab is pushing further into AI, some financial analysts have expressed concerns about increased competition from AI startups and the company's ability to maintain its growth trajectory.
