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 Chat (Agentic), in the UI or your IDE, 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 and a variety of IDEs.

For more information, see:

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:

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:

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 Automate > Sessions.

For more information, see:

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 GitLab Duo Chat (Agentic) 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:

Resources