GitLab Duo

GitLab Duo, centered on the GitLab Duo Agent Platform, is GitLab's AI orchestration layer for the software lifecycle. It provides specialized agents and multi-step flows that work from GitLab context such as issues, merge requests, pipelines, and security findings. The product page states it is available to Premium and Ultimate customers on GitLab 18.8 or later.

What It Is

GitLab Duo Agent Platform is an AI assistance and automation layer built into the GitLab DevSecOps platform. Agents are specialized assistants for tasks such as planning, coding, and security analysis. Flows chain agents together into guided sequences that automate multi-step work. GitLab positions it as picking up after coding agents produce a merge request, carrying the change through review, tests, security, and pipelines.

Agents and Flows in Practice

Documented use cases include turning ideas into structured plans, analyzing vulnerabilities and dismissing false positives, automating merge request code review, diagnosing and fixing failing CI/CD pipelines in a new merge request, converting existing pipelines to GitLab CI/CD, and transforming issues into merge requests. Agentic Chat supports generating code and tests and explaining, fixing, or refactoring code. Teams can build custom agents and custom flows, and manage them through an AI Catalog. Automation can be triggered by GitLab events, schedules, or the tools a team already uses. External agents such as Claude Code and Codex can be integrated into GitLab workflows.

Governance, Models, and Extensibility

Policy controls determine which agents and flows are allowed, where they can operate, and which models they can use. Agent actions are traceable to a user identity and permissions, with sessions, tools, models, and approvals in one trail. Self-hosted models can run on GitLab Self-Managed, and GitLab Dedicated supports bringing your own model through a single-tenant AI Gateway. Two-way MCP integration lets tools supply context to GitLab and lets external coding agents call into GitLab. GitLab says it offers 10+ frontier and open-weight models and that it does not train generative AI models on private data.

Context and Measurement

GitLab Orbit is described as a lifecycle context graph connecting code, merge requests, pipelines, and deployments so agents can reason across the full lifecycle. The page also lists Impact Analytics, in early access, for measuring merge request cycle time and Duo-created merge requests by team, task, and model. The supported languages listed are CSS, Go, HTML, Java, JavaScript, Markdown, Python, Ruby, and TypeScript.