GitHub’s standalone Copilot App went generally available today — no waitlist, no technical preview caveat, available to every Copilot subscriber on macOS, Windows, and Linux. The features that shipped in preview last month (parallel sessions, isolated worktrees, Canvases) are now production-grade. And there’s one genuinely new addition at GA: Cloud Automations, which lets agents run on a schedule in GitHub-hosted infrastructure without your machine.
Cloud Automations: The Feature That Wasn’t in Preview #
Every other Copilot App feature debuted in the technical preview. Cloud Automations is new at GA, and it’s the most consequential addition. The idea: schedule recurring agent tasks in the cloud so they run whether or not your laptop is on.
The mechanism is a /every
command inside a session. You describe what the agent should do and on what trigger — a schedule, a GitHub event like a new issue being opened, or a PR merge. The agent runs in an ephemeral Linux environment on GitHub’s servers. By default, it asks permission before any write action, which matters when you’re not watching.
The analogy that actually makes sense: GitHub Actions, but you describe what you want instead of writing YAML. You’re not configuring a workflow file — you’re telling an agent what to do and trusting it to figure out the steps. CI automation has been “write YAML, debug YAML, curse YAML” for a decade. Cloud Automations is the first credible alternative that doesn’t require learning another DSL.
What GA Actually Means #
The technical preview label meant “subject to change without notice.” Teams couldn’t build on it responsibly — the API surface could shift, features could disappear. GA changes that. The app now carries a stability commitment, and the parallel Copilot SDK GA — stable since early June across Node, Python, Go, .NET, Rust, and Java — means platform developers have a surface to build on without it breaking under them.
Availability is broad: Copilot Free, Pro, Pro+, Business, and Enterprise. No separate subscription. Download from github.com/github/app.
The Actual Workflow: Parallel Sessions with Worktree Isolation #
The foundational architecture hasn’t changed from preview, but it’s worth stating clearly for teams evaluating adoption. Each session runs in its own Git worktree — up to ten parallel sessions on the same repository, none touching each other’s files, conversation state, or task state. You start a session from an issue, a PR, or a plain prompt. The My Work dashboard shows everything in one view: active sessions, background automations, open PRs, issues.
The practical workflow: kick off three sessions in the morning — fix the auth bug, add dark mode, run a dependency audit. Let them run. Come back in two hours. Review the diffs. Merge what looks right, redirect what went sideways. This is a different interaction model than pair-programming with an AI assistant. It’s closer to managing a team that reports back to you.
Canvases: A New Interaction Paradigm #
Canvases remain the hardest feature to explain in a sentence. The clearest version: a Canvas is a structured surface that both you and the agent update simultaneously. Not a chat thread. Not a file editor. A Canvas might show a plan the agent wrote before executing, a PR diff it’s building, a browser session it’s testing against, or a deployment status it’s monitoring. You edit the Canvas while the agent is updating it — reorder steps, approve or reject decisions, redirect work without stopping the session.
GitHub calls this “agent experience (AX)” as distinct from user experience (UX). Canvases are designed for humans and agents sharing a work surface — not for humans navigating menus. The feature overview on the GitHub blog has useful screenshots of what Canvas states look like in practice.
Who Should Try It Now #
The Copilot App is not a replacement for Cursor or VS Code with the Copilot extension. That’s not a knock — it’s a different tool for a different task. The extension is for inline assistance while you’re actively writing. The App is for asynchronous agent sessions you set up and check back on. If you’re already running one AI coding session at a time in your editor, the App adds something adjacent rather than replacing what you have.
One cost consideration: Copilot now runs on AI Credits billing. Running ten parallel sessions at Sonnet-tier models consumes credits at ten times the rate of a single session. Start with two or three sessions, measure the credit usage, then scale. The GA announcement has the full feature list and download links.
Cloud Automations and the stable SDK are the GA additions worth taking seriously. If you’ve been waiting for the technical preview label to drop before piloting this on a real project, today is the day to download it and find out what it can actually do.