GitHub has added Anthropic's Claude agent and OpenAI Codex to its Agent HQ feature, giving developers access to multiple coding agents inside GitHub and Visual Studio Code under a single workflow.
The two third-party agents are available in public preview for users with Copilot Pro+ or Copilot Enterprise subscriptions. GitHub said the aim is to reduce context switching for developers who increasingly use several AI tools in parallel.
Agent HQ acts as a central place to run, track, and review work produced by different agents. GitHub introduced the concept last year as developers started building more complex AI tool stacks across separate services and interfaces.
The latest update adds Claude and Codex alongside GitHub Copilot. Developers can move between agents without changing surfaces, with session history and review information tied to the same repository workflow.
Where it runs
Users can run agents inside GitHub, GitHub Mobile, and Visual Studio Code. GitHub also said support is coming for Copilot CLI.
In GitHub, developers can start and manage sessions from an Agents tab within a repository, where agents are enabled. They can submit a request and choose the agent. Work runs asynchronously by default, and developers can follow progress live or review completed sessions later.
Each session produces artefacts such as comments, drafts, and proposed code changes. GitHub positions these as standard contributions that teams can review through existing processes.
In Visual Studio Code, users can work with agents through an Agent sessions view on version 1.109 or later. Options include fast interactive assistance, cloud tasks that run on GitHub, and background asynchronous work that is limited to Copilot.
Workflow integration
GitHub is also integrating agents more directly into collaboration features such as issues and pull requests. Teams can assign an issue to Copilot, Claude, Codex, or more than one agent to compare results.
Agents can submit draft pull requests for review. Developers can also assign agents to existing pull requests for changes or analysis. GitHub said users can prompt follow-up work by mentioning @Copilot, @Claude, or @Codex in pull request comments.
Activity logs are designed to make agent output traceable within normal review workflows. GitHub also flagged that "Agents can still make mistakes" and positioned review and comparison as core parts of the approach.
Comparing outputs
A central element of Agent HQ is the ability to run multiple agents against the same task and compare responses. GitHub said developers can use different agents for different stages of work, from early exploration to implementation and review.
The company also described using multiple agents in parallel to test logic, review design choices, and propose smaller code changes. It framed this as shifting developer attention from code syntax to higher-level decisions.
Anthropic highlighted how its agent will appear within GitHub's workflows, including code changes and code review interactions.
"We're bringing Claude into GitHub to meet developers where they are. With Agent HQ, Claude can commit code and comment on pull requests, enabling teams to iterate and ship faster and with more confidence. Our goal is to give developers the reasoning power they need, right where they need it," said Katelyn Lesse, Head of Platform, Anthropic.
OpenAI linked the Codex integration to the wider adoption of AI-assisted development tools across the software industry.
"Our collaboration with GitHub has always pushed the frontier of how developers build software. The first Codex model helped power Copilot and inspired a new generation of AI-assisted coding. We share GitHub's vision of meeting developers wherever they work, and we're excited to bring Codex to GitHub and VS Code. Codex helps engineers work faster and with greater confidence-and with this integration, millions more developers can now use it directly in their primary workspace, extending the power of Codex everywhere code gets written," said Alexander Emiricos, OpenAI.
Access and controls
GitHub said Claude and Codex must be explicitly enabled in settings before use. Each agent session consumes one premium request under the relevant Copilot plan.
Agent HQ also sits alongside administrative controls aimed at teams. GitHub described central policy management for which agents and models are permitted across an organisation, as well as audit logging and access management.
GitHub also referenced related previews that extend into software quality and measurement. These include GitHub Code Quality, which it said evaluates maintainability and reliability impact of changed code, and a Copilot metrics dashboard for tracking usage and impact.
GitHub said it expects access to Claude and Codex to expand to more Copilot subscription types over time. It also said it is working with additional partners, including Google, Cognition, and xAI, for more specialised agents across GitHub, Visual Studio Code, and Copilot CLI workflows.