Get started with the GitLab Duo Agent Platform | GitLab Docs Skip to main content Go to GitLab Docs homepage What's new? What's new? Get free trial Toggle menu * Use GitLab * GitLab Duo * GitLab Orbit * Extend * Install * Administer * Subscribe * Contribute * Solutions GitLab Duo Agent Platform Getting started Tutorials Availability Agentic Chat AI Catalog Agents Flows Sessions Customization Code Suggestions Models Contextual awareness Composite identity Authentication Security threats Model Context Protocol Semantic code search Troubleshooting GitLab Duo Non-Agentic Data usage * GitLab Docs / * GitLab Duo / * GitLab Duo Agent Platfor… / * Getting started Help us learn about your current experience with the documentation. Take the survey . Get started with the GitLab Duo Agent Platform The GitLab Duo Agent Platform is an AI-native solution that embeds multiple intelligent assistants ( agents ) throughout the software development lifecycle. * Instead of following a linear workflow, collaborate asynchronously with AI agents. * Delegate routine tasks, from code refactoring and security scans to research, to specialized AI agents. The Agent Platform is made up of several features, which are available in the GitLab UI and IDEs. Step 1: Access GitLab Duo Chat GitLab Duo Agentic Chat is your interface for asking questions and interacting with agents. It can provide advice, but it can also propose and implement solutions. Chat has access to your project, including issues, merge requests, commits, and CI/CD pipelines, and Chat maintains context across conversations. You can build up complexity gradually, reference previous responses, and iterate until you reach the desired outcome. GitLab Duo Chat is available in: * The GitLab UI. * A variety of IDEs. * The GitLab Duo CLI. For more information, see: * GitLab Duo Agentic Chat . * GitLab Duo CLI . Step 2: Work with agents Agents are specialized AI assistants designed for specific workflows. * Foundational agents are available by default and handle common development tasks. The GitLab Duo Agent provides general assistance for questions, explanations, and code navigation. Other foundational agents help with things like planning releases or securing code. * Custom agents are created by your organization to address team-specific workflows. You can build agents for code review standards, compliance checks, deployment automation, or any workflow that s unique to your team. * External agents integrate GitLab with AI model providers you already use. You trigger external agents from issues, epics, and merge requests. For more information, see: * Agents overview . * Foundational agents . * Custom agents . * External agents . Step 3: Use multiple agents together in a flow A flow is a combination of one or more agents working together to complete a task. Flows can help you automate multi-step workflows that would typically require manual coordination between tools or team members. For example, you can trigger a flow from a merge request, and the flow can do a security scan, review code, generate tests, and draft documentation. GitLab provides foundational flows, like the software development flow in your IDE, or flows in the UI that do things like converting or fixing CI/CD pipelines. You can also create your own custom flows. The AI Catalog is the central location where you discover and create agents and flows, and enable them for use in your projects. For more information, see: * Flows . * AI Catalog . * Triggers . Step 4: Monitor and review agent activity Actions an agent takes are tracked in a session with logs. Sessions can help aid with debugging, facilitate learning, and support audit requirements. To view sessions, go to your project and select AI > Sessions . For more information, see: * Sessions . Step 5: Extend capabilities with integrations To increase the knowledge of your AI agents, use the Knowledge Graph. It creates structured representations of your code repositories and helps agents and your team better understand relationships between files, functions, and dependencies. You can also extend the platform beyond GitLab by connecting with external tools and data sources. * Connect GitLab Duo features like Agentic Chat to external MCP servers so that other MCP clients can provide more comprehensive assistance. * MCP server works in the opposite direction: external AI tools like Claude Desktop or Cursor can securely connect to your GitLab instance, giving those tools access to your GitLab data. For more information, see: * Knowledge Graph . * MCP clients . * MCP server . Resources * Eight-part tutorial: Get started with GitLab Duo Agent Platform: The complete guide * Blog: GitLab engineer: How I improved my onboarding experience with AI * Talk recording: Agentic AI in GitLab Duo Agent Platform | Use Cases & Best Practices | DACH Roadshow Vienna 2025 ( slides ) * Transition from non-agentic GitLab Duo to the GitLab Duo Agent Platform * Step 1: Access GitLab Duo Chat * Step 2: Work with agents * Step 3: Use multiple agents together in a flow * Step 4: Monitor and review agent activity * Step 5: Extend capabilities with integrations * Resources * Facebook * LinkedIn * Twitter * YouTube Company * About GitLab * View pricing * Try GitLab for free Feedback * View page source * Edit in Web IDE * Contribute to GitLab * Suggest updates Help & Community * Get certified * Get support * Post on the GitLab forum Resources * Terms * Privacy statement * Use of generative AI * Acceptable use of user licenses *