GitHub Integrates Claude and Codex into Agent HQ Public Preview
GitHub has announced the integration of Anthropic's Claude and OpenAI's Codex AI coding agents into its Agent HQ initiative. This public preview enables developers with Copilot Pro Plus or Copilot Enterprise subscriptions to access these agents directly within GitHub, GitHub Mobile, and Visual Studio Code. The integration aims to provide a unified workflow for developers, allowing them to manage and compare outputs from various AI tools without extensive context switching.
Introduction to Agent HQ
Agent HQ serves as a central platform designed to embed AI agents directly into developers' daily workflows. Its objective is to address the increasing complexity of AI tool stacks and reduce the friction caused by developers frequently switching between multiple AI tools.
"Its objective is to address the increasing complexity of AI tool stacks and reduce the friction caused by developers frequently switching between multiple AI tools."
Within Agent HQ, developers can run, track, and review work generated by different agents, streamlining their interaction with diverse AI capabilities.
Availability and Access
Currently, Anthropic's Claude and OpenAI's Codex are available in public preview. Access is provided exclusively to users holding a Copilot Pro Plus or Copilot Enterprise subscription. These agents can be utilized across GitHub, GitHub Mobile, and Visual Studio Code environments, with planned support for Copilot CLI in the future.
Prior to use, Claude and Codex must be explicitly enabled in settings. Each session with an agent counts as a premium request under the applicable Copilot plan.
Core Functionality and Workflow Integration
Agent HQ empowers developers to select from Copilot, Claude, Codex, or other custom agents for various tasks. Developers can initiate and manage agent sessions from an 'Agents' tab within a repository where agents are enabled. Work generated by agents typically runs asynchronously, allowing developers to monitor progress or review completed sessions later. Agent outputs, such as comments, draft proposals, and proposed code changes, are produced as artifacts that can be reviewed through standard team processes.
In Visual Studio Code (version 1.109 or later), users can interact with agents via an 'Agent sessions' view. This view offers options for interactive assistance, cloud tasks, and background asynchronous work (though background work is currently limited to Copilot).
The integrated agents can also be assigned to collaborative features like issues and pull requests. Teams can assign issues to Copilot, Claude, Codex, or even multiple agents to compare their respective solutions. Agents are capable of submitting draft pull requests for review, and developers can assign agents to existing pull requests for modifications or analysis. Follow-up work can be prompted by mentioning @Copilot, @Claude, or @Codex in pull request comments. Activity logs are designed to ensure that all agent output remains traceable within existing review workflows.
"GitHub has noted that agents can still make mistakes, positioning human review and comparison as integral components of this approach."
A central feature of Agent HQ is the ability for developers to run multiple agents against the same task and compare their responses. This facilitates using different agents for various stages of work, from initial exploration to implementation and review. The platform also suggests utilizing multiple agents in parallel for tasks such as testing logic, reviewing design choices, and proposing minor code changes, aiming to allow developers to focus on higher-level decisions rather than code syntax.
Administrative Controls and Future Plans
Agent HQ includes comprehensive administrative controls for teams, such as central policy management to determine which agents and models are permitted across an organization. This is complemented by audit logging and access management features. Related previews, including GitHub Code Quality for evaluating code changes and a Copilot metrics dashboard for tracking usage, are also available.
This integration aligns with GitHub's broader strategy of incorporating various AI models and agents to enhance its Copilot offering, which already provides access to models from Anthropic, Google, xAI, and OpenAI. GitHub plans to expand access to Claude and Codex to additional GitHub Copilot subscription tiers in the future. The company is also collaborating with other partners, including Google, Cognition, and xAI, to introduce more specialized agents across GitHub, Visual Studio Code, and Copilot CLI workflows. Separately, Microsoft has been internally evaluating Anthropic's Claude Code tool, comparing its performance with GitHub Copilot to identify areas for improvement in the latter.