GitHub just turned Copilot into its own desktop operating system for AI agents. The new GitHub Copilot app, announced at Microsoft Build 2026, is a dedicated application that replaces scattered chat windows with a unified control center for managing multiple agents working in parallel across repositories. The app introduces a "My Work" view that consolidates active agent sessions, issues, pull requests, and background automations into a single dashboard. Every agent session runs in its own isolated Git worktree, meaning parallel agents can operate on the same codebase without conflicts.
GitHub describes this as the "agent-native desktop experience built on GitHub." Available today in technical preview for Windows 11, Windows 11 on Arm, Mac, and Linux. The app requires a Copilot Pro, Pro+, Business, or Enterprise subscription. A waitlist is open for Copilot Free users who want access later.
Beyond the dashboard, GitHub is introducing "canvases", bidirectional work surfaces where developers and agents interact directly with the work itself. A canvas can display a plan, pull request, browser session, terminal output, or deployment state.
Agents update the canvas as they work; developers can edit, reorder, approve, or redirect that work on the same surface.
"This is the beginning of agent experience (AX) in the Copilot app," GitHub's team wrote. "Chat is where you instruct, discuss, and reason through ambiguity.
Canvases are where that intent becomes visible work you can inspect, steer, and verify." The app supports both local and cloud sandboxing. Local sandboxes run in isolated environments on the developer's machine with restricted filesystem and network access, configurable through centrally enforced policies.
Cloud sandboxes run in fully isolated, ephemeral Linux environments hosted by GitHub, letting developers pick up sessions from any device.
Agent Merge carries pull requests through review, CI checks, and merging, monitoring for passing checks, tracking required reviewers, and addressing failures. Developers choose how much autonomy Copilot gets: drive CI back to green, address feedback, or merge when conditions are met. On the code review side, Copilot now offers a "medium" tier that routes pull requests to a higher-reasoning model for better precision and recall.
Admins can set per-repository guidelines to "low" or "medium," assigning lighter models for low-risk code and reserving more capable models for high-impact repos. The /security-review skill adds a dedicated path for security-focused evaluation, and the /rubberduck skill is now generally available for critiquing implementations across multiple model families. Azure DevOps users get native Copilot code review with the same inline comments and committable fix suggestions.
Developers who prefer the terminal can use the redesigned Copilot CLI, which adds voice input via on-device speech-to-text and an experimental mode with tabbed access to pull requests, issues, and gists. The /every command schedules recurring prompts and background tasks. The GitHub Copilot SDK is now generally available in Node.js/TypeScript, Python, Go, .NET, Rust, and Java, exposing the same agentic runtime that powers the app.
Partner-built agent apps from LaunchDarkly, Amplitude, Sonar, PagerDuty, and Miro integrate directly into the Copilot workflow. The numbers show why GitHub is investing in infrastructure. On GitHub alone, commits nearly doubled year over year, crossing 1.4 billion per month, with over 2 billion GitHub Actions minutes consumed weekly. As Paul Thurrott noted, the app functions as a "central dashboard for managing AI agents and interacting with GitHub", a recognition that the old model of AI-assisted coding no longer fits how developers actually work when agents outnumber humans in the build process.
