TLDR: GitLab has launched the public beta of its Duo AI Agent Platform, integrating custom AI agents and a Knowledge Graph to transform software development. This initiative aims to boost developer productivity, automate complex DevSecOps tasks, and enhance code management with deep contextual understanding and advanced security features.
GitLab has officially rolled out the public beta of its groundbreaking Duo AI Agent Platform, marking a significant evolution in the DevSecOps landscape. This new platform, announced in July and August 2025, is designed to foster asynchronous collaboration between human developers and specialized artificial intelligence agents, fundamentally reimagining the software development lifecycle.
At the core of this innovation are custom AI agents, trained to automate a diverse range of DevOps tasks and workflows. These agents can be assigned tasks both synchronously and asynchronously, offering an extensible platform with comprehensive visibility across the entire software development lifecycle (SDLC). Emilio Salvador, Vice President of Strategy and Developer Relations for GitLab, highlighted that these AI agents move beyond passive chat interfaces, becoming active members of a DevOps team capable of gathering context, clarifying ambiguities, and even making code changes.
Complementing the AI agents is the new GitLab Knowledge Graph, described as a ‘living, breathing map’ of an entire codebase. This graph goes beyond merely indexing files; it comprehends the intricate connections and dependencies within a project. It provides AI agents with the deep context necessary to offer insightful contributions, enabling them to understand how changes in one area might impact another. This contextual understanding is crucial, as an AI’s effectiveness is heavily influenced by the context in which it operates, as noted by Mitch Ashley, Vice President at Futurum Group.
The Duo AI Agent Platform introduces several key features to enhance this collaboration. Developers can orchestrate multi-agent workflows, assigning different roles such as a Software Developer Agent, a Security Analyst Agent, or a Deep Research Agent, each performing specialized tasks. Interaction with these agents is facilitated through chat interfaces embedded in popular IDEs like VS Code and JetBrains, allowing developers to delegate tasks using natural language commands such as ‘/explain’, ‘/tests’, or ‘/include’.
Security is a paramount concern, addressed by features like ‘Context Exclusion’. This allows developers to designate sensitive parts of their projects – such as files containing secrets or proprietary algorithms – as off-limits for AI access, ensuring data integrity and confidentiality. Organizations can also define custom agent rules using natural language, aligning AI behavior with established coding preferences and compliance requirements.
GitLab CEO Bill Staples emphasized the transformative nature of this release, stating, “GitLab Duo Agent Platform isn’t just another AI tool; it’s a fundamental reimagining of software development from isolated, linear processes into dynamic, intelligent collaboration.” He added that by leveraging GitLab’s position as the system of record for the entire SDLC, the platform provides AI agents with “unprecedented context and capabilities” to boost productivity, velocity, and efficiency.
Also Read:
- Secure Code Warrior Unveils Trust Agent: AI Beta for Enhanced Code Risk Management in AI-Driven Development
- Google Cloud Introduces New AI-Powered Tools to Boost Data Science Efficiency
The platform also supports Anthropic’s Model Context Protocol (MCP), enabling seamless integration with external systems and additional data sources. This positions the GitLab Duo AI Agent Platform not just as a developer productivity tool, but as an intelligent, extensible layer across the entire DevSecOps ecosystem. The public beta is available to GitLab.com Premium and Ultimate customers, as well as self-managed installations, signaling GitLab’s commitment to integrating AI deeply into the software development process.
