GitHub Ships Standalone Copilot Desktop App in Technical Preview, Stepping Onto Cursor and Claude Code's Turf

Overview

GitHub on May 14 opened a technical preview of a new standalone Copilot application that runs as a native desktop program rather than as an IDE extension, the company announced on its changelog. GitHub describes the product as “a GitHub-native desktop experience to start agentic development from the work in front of you, keep it isolated, steer it as it goes, and land the change through pull request review.”

The move places GitHub in the same product category as Anthropic’s Claude Code, OpenAI’s Codex CLI, and Cursor — tools that house coding agents in dedicated, stateful surfaces rather than inside Visual Studio Code or other editors.

What We Know

The app is a desktop program rather than a web service. According to the GitHub documentation page, it “is a desktop application for agent-driven development that brings parallel workstreams, GitHub integration, and PR lifecycle management into one place,” and runs on macOS, Linux, and Windows. The project’s GitHub repository, which hosts the public releases and issue tracker, lists recommended builds for Mac (Apple Silicon), Windows, and Linux, with alternate downloads for Mac (Intel) and Windows (ARM).

Each Copilot task lives in its own isolated workspace. The docs state that users can “Run multiple isolated agent sessions simultaneously, each with a dedicated git worktree and branch” — a design that lets several agents run against the same repository without their checkouts colliding. The GitHub Copilot app product page summarises the same idea as “Run multiple agent sessions across repos, each isolated and tracked in real time.”

Sessions can be driven in one of three modes the documentation lists as “Interactive (collaborative),” “Plan (agent plans, you approve),” and “Autopilot (fully autonomous).” Users can “Select from multiple LLMs and adjust reasoning effort per session,” and “Save recurring agent tasks and run them on a schedule or on demand,” per the same page.

Workflows are anchored to GitHub objects rather than to local files. The product page describes the loop as: “Pick up an issue or PR from your inbox, put agents on it, review the diff, and merge.” The docs add that users can “Browse and find issues, start sessions from them, create and close pull requests, review pull requests,” and “Search for issues or pull requests across your repositories directly from the app.” Extensibility comes from “MCP servers and your own custom skills,” according to the product page.

Access is gated by Copilot tier. The changelog states that “GitHub Copilot Pro and Pro+ subscribers” can “sign up for early access as the technical preview expands,” while “GitHub Copilot Business and Enterprise subscribers will get access as availability rolls out through the week.” The docs confirm that Business and Enterprise users can download directly from the GitHub Copilot app repository, while Pro and Pro+ users must “join the waitlist to request access.” The product page notes that the app “requires an active paid GitHub Copilot subscription.”

Releases have been shipping at a brisk cadence since the preview opened. The github/app releases page lists v0.2.0, whose release notes read simply “Technical Preview for the GitHub app,” as the first public build on May 12, followed by v0.2.1 through v0.2.5 over the next four days — six releases in total, with v0.2.5 published on May 15. Subsequent builds have focused on accessibility improvements, queued follow-up prompts during active sessions, a foldering tree for workspace navigation, integrated browser previews under an opt-in experiment, and bug fixes around worktree lifecycle and merge availability detection.

What’s New About It

GitHub already offered Copilot as an editor extension, as a chat interface inside github.com, and as a cloud agent that opens pull requests from issues. The desktop app collapses those surfaces into a single program and adds the things competing standalone coding-agent UIs have built: parallel sessions, worktree-level isolation, and an agent loop that persists outside any IDE.

The app builds on a cloud-side capability GitHub already shipped. In April, the company introduced a “Fix with Copilot” button on github.com that resolves merge conflicts via the Copilot cloud agent. “Click the button, and a comment is prepopulated in the comment box asking Copilot to resolve conflicts. When you submit the comment, Copilot fixes the conflicts, checks that the build and tests still pass, and pushes,” the changelog explained. The desktop app brings that style of agent-driven PR work into a local interface where multiple such jobs can run in parallel.

What We Don’t Know

GitHub has not disclosed when the desktop app will leave technical preview, whether it will be priced separately from existing Copilot subscriptions, or how usage will be metered against the per-plan limits the company began enforcing earlier this year. The preview is gated by waitlist for Pro and Pro+, and the company has not said how many seats are being admitted.

The app’s relationship to other GitHub-owned surfaces — the existing GitHub Desktop client, the github.com web UI, the GitHub CLI, and Copilot’s IDE extensions — also remains underspecified. Documentation describes the new app as the home for agent-driven workstreams, but none of the older clients are being deprecated, and GitHub has not published a positioning statement about which surface to use for which workflow.

Finally, the company has not detailed performance, telemetry, or how the local app interacts with the Copilot cloud agent that does much of the existing in-PR work. The first month of releases will likely make those tradeoffs more visible as the preview expands.