---
title: What does GitHub Agent HQ let enterprise admins govern?
url: https://opencontroller.lyzr.com/blog/what-does-github-agent-hq-govern
published: 2026-11-18
updated: 2026-11-18
author: Lyzr
---

# What does GitHub Agent HQ let enterprise admins govern?

**Agent HQ gives GitHub Enterprise admins one place to see and control coding agents that work inside GitHub, including third-party agents: an AI Controls page for policy, an agent sessions view, agent-tagged audit log events, enterprise-wide custom agent definitions and, since September 2026, managed permissions for agent operations. It does not govern agents, tools or spend that sit outside GitHub's surfaces.** Use it as the GitHub layer of a wider control model, not the whole model.

## What is Agent HQ?

GitHub [announced Agent HQ](https://github.blog/news-insights/product-news/welcome-home-agents/) at GitHub Universe on 28 October 2025. The idea: coding agents from several vendors run inside GitHub, under one subscription and one set of controls. GitHub named agents from Anthropic, OpenAI, Google, Cognition and xAI as arriving over the following months.

Two parts matter for governance:

- **Mission control:** a command centre to assign, steer and track agent tasks across GitHub, VS Code, mobile and the CLI. It includes branch controls for when CI runs on agent-created code and identity controls for which agent works on a task.
- **Enterprise controls:** what GitHub called the agent control plane, for security policies, audit logging, access management, allowed agents and model access.

## Which enterprise controls are generally available?

GitHub's [February 2026 changelog](https://github.blog/changelog/2026-02-26-enterprise-ai-controls-agent-control-plane-now-generally-available/) moved the core controls to general availability. A later [September 2026 changelog](https://github.blog/changelog/2026-09-09-enterprise-managed-permissions-for-github-copilot-agent-operations) added managed permissions.

| Control | What it does | Status (per GitHub changelogs) |
|---------|--------------|-------------------------------|
| AI Controls page | One enterprise settings area for AI policy; replaced the old Copilot policies page | GA, Feb 2026 |
| Agent sessions view | Search agent session activity by agent (including third-party) and by organisation | GA, Feb 2026 |
| Agent audit events | `actor_is_agent` on entries; `agent_session.task` records sessions starting, finishing or failing | GA, Feb 2026 |
| Custom agent definitions | Apply enterprise-wide definitions from `.github-private/agents/*.md`; push rule to protect `.github/agents/*.md`; API support | GA, Feb 2026 |
| Delegated AI administration | Enterprise custom roles to view audit logs and sessions and manage AI Controls | GA, Feb 2026 |
| Managed permissions for agent operations | Block, require approval for, or allow shell commands, file reads/edits and network domains, per enterprise team; users cannot loosen them | GA, Sep 2026, in the Copilot app, Copilot CLI and VS Code sessions using Agent Host |
| Enterprise MCP allowlist | Restrict which MCP servers can be used | Still preview at the Feb 2026 GA announcement |

Check the changelog for the current status before you rely on a preview feature.

## How should admins set it up?

1. **Delegate administration.** Create an enterprise custom role for AI managers so platform engineering, not only enterprise owners, can run AI Controls.
2. **Decide which agents are allowed.** Enable only the third-party agents you have reviewed.
3. **Set managed permissions.** Start strict: block network access except approved domains, require approval for shell commands, and loosen per team.
4. **Standardise custom agents.** Publish approved definitions from `.github-private/agents/` and protect `.github/agents/*.md` with the push rule. See [governing GitHub Copilot custom agents](/blog/github-copilot-custom-agents-governance).
5. **Stream audit events.** Send events with `actor_is_agent` and `agent_session.task` to your SIEM. See [GitHub Copilot audit log events](/blog/github-copilot-audit-log-events).
6. **Control MCP.** Use the MCP allowlist where available and review servers. See [controlling MCP servers in GitHub Copilot](/blog/control-mcp-servers-in-github-copilot).

## What still sits outside Agent HQ?

Agent HQ governs work that runs through GitHub's surfaces. Typical gaps:

- **Coding agents used outside GitHub surfaces.** A developer running another vendor's CLI agent directly with a personal or team API key, or a different IDE.
- **Local MCP servers and skills on laptops** that never touch a GitHub-managed session.
- **Model spend outside Copilot billing**, such as direct provider API keys used by agents.
- **Non-coding agents.** Agents in CRM, ERP, data platforms and cloud AI services are out of scope.
- **Cross-platform posture.** The same identity or key may be used by agents elsewhere; GitHub sees only its side.

For the device side, read [governing coding agents on laptops with MDM](/blog/govern-coding-agents-on-laptops-with-mdm). For the cross-vendor view, see the pillar [how to govern AI agents across platforms](/blog/how-to-govern-ai-agents-across-platforms).

## How Opencontroller does this

Opencontroller Edge extends control to developer devices, coding agents, skills and local MCP servers, and is installed by IT via MDM, so it covers the laptop-side gaps. The main platform keeps a governed registry of agents, versions, tools and MCP servers and auto-discovers agents, models and API keys across AWS, Azure and Google Cloud. It works alongside GitHub's own controls rather than replacing them: control without migration. [Book a demo](https://www.lyzr.ai/book-demo/) to map your coding-agent estate.

## Frequently asked questions

**Does Agent HQ cost extra?**
GitHub said third-party coding agents would be included in a paid Copilot subscription. Check your plan for current terms.

**Can Agent HQ require human approval for agent shell commands?**
Yes. Managed permissions for agent operations, generally available since September 2026, can block, require approval for or allow shell commands, file operations and network domains.

**Does Agent HQ see coding agents run outside GitHub?**
No. It governs agents working through GitHub's surfaces. Agents run directly against provider APIs or in other tools need separate controls.

## Key facts

GitHub Agent HQ (announced 28 Oct 2025) lets Enterprise admins govern coding agents inside GitHub. GA in Feb 2026: AI Controls page, agent sessions view, audit events with actor_is_agent and agent_session.task, enterprise custom agent definitions, delegated AI admin roles. GA in Sep 2026: managed permissions to block/require approval/allow shell commands, file operations and network domains. Enterprise MCP allowlists were still preview at Feb 2026. Gaps: agents outside GitHub, local MCP servers, non-Copilot spend. Opencontroller Edge covers developer devices via MDM. Pricing starts from $250K per deployment per year.

## Related

- [How do you govern custom agents in GitHub Copilot?](https://opencontroller.lyzr.com/blog/github-copilot-custom-agents-governance.md)
- [Which GitHub audit log events show Copilot policy and agent activity?](https://opencontroller.lyzr.com/blog/github-copilot-audit-log-events.md)
- [How do you control which MCP servers GitHub Copilot can use?](https://opencontroller.lyzr.com/blog/control-mcp-servers-in-github-copilot.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
