GitHub has launched a new feature called Agent HQ, designed to simplify the use of multiple AI coding tools by bringing them into one central interface. The platform aims to give developers better control over third-party AI agents from providers like OpenAI, Google, Anthropic, xAI, and Cognition.
GitHub COO Kyle Daigle told CNBC that developers have gained powerful new tools through AI, but managing each one separately has created unnecessary complexity. Agent HQ offers a way to organize that process by allowing users to assign, monitor, and coordinate tasks across agents from one location.
Agent HQ will be available soon to GitHub Copilot users, while Copilot Pro+ subscribers can already use OpenAI’s Codex model in VS Code Insiders. The system is built with enterprise security in mind. Instead of giving agents unrestricted access to code repositories, GitHub’s system uses sandboxed environments, limits each agent’s access to specific branches, and enforces strict identity protocols.
Daigle said this approach ensures that even if an agent behaves unexpectedly, it won’t be able to send data out of the system unless those protections are manually disabled.
GitHub is also introducing tools that allow companies to customize how AI agents behave. With AGENTS.md files, teams can set permanent preferences for their AI assistants, like choosing specific logging tools or writing styles, which then apply across the project without needing to be re-entered repeatedly.
The platform also now supports the Model Context Protocol (MCP), a communication standard for AI tools developed by Anthropic. Through GitHub’s new registry in VS Code, developers can discover and connect MCP-enabled tools with a single click, streamlining how agents interact with outside services.
Alongside Agent HQ, GitHub is adding new functions to VS Code. A feature called Plan Mode lets developers work with Copilot to design structured development workflows. Instead of jumping straight into writing code, the AI first asks clarifying questions and builds a step-by-step plan, which can then be executed locally or by cloud agents.
GitHub is also updating how it handles code review. The company’s CodeQL engine, which has typically been used to scan for security issues, will now help review pull requests generated by agents. This creates an added layer of quality control before the code reaches human reviewers.
