GitHub made the GitHub Copilot app generally available for Windows, macOS and Linux on June 17, 2026, about two weeks after showcasing it at Microsoft Build 2026. The desktop application is built for agent-driven development: it runs several Copilot agent sessions side by side, each in its own Git worktree and branch, and pulls issues, pull requests and automations into one workspace. For developers who were juggling agents across an editor, a terminal and the browser, it gives Copilot's agents a dedicated home outside the IDE.

Key Facts#

  • Technical preview: on May 29, 2026, GitHub expanded the app's technical preview to all Copilot Pro, Pro+, Business and Enterprise customers.
  • General availability: June 17, 2026, on Windows, macOS and Linux.
  • All plans: from July 7, 2026, the app is available on every Copilot plan, including Copilot Free and GitHub Education.
  • Foundation: GitHub's documentation says the app is built on Copilot CLI and integrates natively with GitHub.
  • Isolation: each session can run in a new worktree with its own branch, in your local repository or in a cloud sandbox.
  • Autonomy modes: Interactive, Plan and Autopilot, with a choice of models, adjustable reasoning effort and bring-your-own-key (BYOK) support.
  • Governance: for Copilot Business and Enterprise, a dedicated app policy, enabled by default, is managed separately from the Copilot CLI policy.

What Happened#

GitHub describes the Copilot app as a desktop application purpose-built for agent-driven development. Instead of a chat panel attached to an editor, the app is organized around sessions. You pick a project from a local folder, from GitHub or from a Git URL, then choose where the session runs: in a new working tree, directly in your local repository or in a cloud sandbox. Because each session gets its own isolated workspace, several agents can make progress on different tasks at the same time without overwriting each other's changes, and active sessions are grouped by repository in the sidebar.

Each session has a mode that sets how much autonomy the agent gets. Interactive mode keeps the developer steering, Plan mode has the agent propose a plan for approval before it acts, and Autopilot lets it work through a task end to end. Developers can pick from multiple models, including their own provider's models through BYOK, and tune reasoning effort per session. Coverage of the launch highlighted a single "My Work" view that brings together active sessions, open issues, pull requests and background automations across connected repositories.

Around the core session model, the app adds automations for recurring tasks, local sandboxes that restrict which resources an agent's tools can reach, customizable canvases for human-agent collaboration, and a /chronicle command for looking back through session history. It also respects organization and enterprise content exclusion policies.

Background#

The app is the latest step in GitHub's shift from an autocomplete assistant to a platform for coordinating agents. In November 2025, VS Code 1.106 introduced Agent HQ as a single view to start, monitor and review agent sessions, whether local or remote and whether from Copilot or OpenAI Codex. The Copilot CLI brought the same agents to the terminal, and on June 2, 2026, GitHub made the Copilot SDK generally available so developers can embed the same runtime in their own tools (see GitHub Copilot SDK general availability).

As agents became capable of longer, more autonomous tasks, the bottleneck moved from writing code to supervising work. German outlet heise summed up the pitch at launch as an app meant to tame the chaos of agent-driven development. A desktop app built around parallel, isolated sessions is GitHub's answer to that supervision problem.

Why It Matters for Developers#

Parallel agents change how you plan work. Instead of one assistant helping with one task, a developer can hand three well-scoped issues to three sessions, review their branches and merge the good results. That only works if tasks are genuinely independent and the codebase has fast, trustworthy tests, because review becomes the limiting factor. Our guide to AI-assisted .NET development covers how to scope tasks for agents, and unit testing in .NET covers the safety net they depend on.

There are practical details for .NET solutions in particular:

  • Worktrees mean separate builds. Each worktree has its own bin and obj folders, so parallel sessions multiply restore and build work on large solutions. The shared NuGet global packages folder helps, but plan for disk space and CPU on your machine.
  • Match autonomy to risk. Use Plan mode for schema migrations, security-sensitive code or public API changes, and reserve Autopilot for well-tested, low-blast-radius work.
  • Lean on CI. Because the app integrates with GitHub branches and pull requests, agent output should flow through the same pipelines and required checks as human changes. See CI/CD for .NET with GitHub Actions.
  • Use sandboxes deliberately. Local sandboxes that restrict tool access, together with content exclusion policies, are the main controls for keeping agents away from secrets and sensitive code.

For administrators, the separate app policy matters. Organizations can allow the Copilot CLI while blocking the desktop app, or the reverse, which makes staged rollouts easier.

What's Next#

GitHub widened access quickly after GA, opening the app to every Copilot plan on July 7, 2026. In August, GitHub said its Agent Plugins 1.0 format lets developers build a plugin once and use it across VS Code, the Copilot CLI, the Copilot SDK and the Copilot app, and it has since said the app and CLI honor content exclusions across agentic workflows. The open questions are about scale: how teams will review the output of many parallel agents without creating a new bottleneck, and how pricing and usage limits will evolve as agent sessions run longer.

Sources#