gitlab Introduction to GitLab Duo Agent Platform (opens in new tab)
GitLab Duo Agent Platform introduces an AI orchestration layer designed to move beyond simple code generation into full software development lifecycle (SDLC) automation. By utilizing specialized agents and asynchronous flows, the platform enables teams to delegate complex tasks like code reviews and pipeline fixes to AI "team members" who possess full context of the project. This transition from linear workflows to multi-agent collaboration allows developers to maintain oversight through detailed session logs while focusing on high-level innovation. ### Core Functionality and SDLC Context * The platform acts as an orchestration layer that enables asynchronous collaboration between human developers and specialized AI agents. * It utilizes deep SDLC context, pulling data from issues, epics, merge requests, CI/CD logs, wikis, and security scans to inform AI actions. * Automation is designed to understand and adhere to specific organizational standards, practices, and compliance requirements. ### Agent Interaction and Interface Methods * **GitLab Duo Agentic Chat:** Provides a real-time, synchronous interface via a persistent panel in both the GitLab Web UI and supported IDEs. * **Triggered Foundational Flows:** Users can invoke pre-built GitLab workflows, such as "Fix CI/CD Pipeline" or "Convert Jenkins to GitLab CI/CD," directly within the platform. * **Custom and External Flows:** Automated workflows can be triggered asynchronously by @mentioning agents or assigning reviewers in issue and merge request comments. * **External Agent Support:** The platform supports third-party models like Claude Code and OpenAI Codex, executing them on GitLab platform compute via runner execution. ### Distinguishing Agents from Flows * **Agents:** These are specialized assistants defined by unique system prompts and toolsets; they are best suited for interactive tasks and instant feedback within the chat interface. * **Flows:** These are autonomous, multi-step workflows designed for complex background tasks, such as multi-file refactoring or event-driven automation. * **Execution Environment:** While agents are interactive, flows run asynchronously on platform compute, triggered by specific GitLab events or user assignments. ### Platform Management and Transparency * **AI Catalog:** A centralized library for discovering, creating, and sharing custom agents and flows across an entire organization. * **Automate Hub:** A management center used to configure triggers, monitor active flows, and manage agent permissions. * *Sessions:** Every interaction creates a session log that provides a transparent "decision trail," including agent reasoning, tool calls, and pipeline execution status. * **Model Selection:** Starting with GitLab 18.4, users can select specific foundational models for their conversations within the Web UI to better suit the task at hand. Teams looking to implement the GitLab Duo Agent Platform should begin by utilizing foundational flows for common tasks like pipeline debugging before moving toward custom agent creation. Reviewing the transparency logs in the "Sessions" view is highly recommended to refine agent prompts and ensure that automated actions align with internal development standards.