GitHub launched a desktop-native Copilot app in technical preview this week. The key feature: agent sessions tied to a repo, issue, or PR — with a full session space that s, resumes, and can drive changes into a pull request.
I spent the last two days testing it. Here's what's actually different and whether it's worth your attention.
The old way: you open VS Code, you get Copilot completions in your editor, maybe you use a chat panel. The agent is a feature inside your IDE.
The new way: you open a standalone Copilot app that treats your repository as a first-class context. You can kick off an agent session tied to a specific PR, issue, or branch. The session has a full workspace — files, conversation history, task state — that s when you close the app and resumes when you come back.
The agent can then open a PR with its changes, which you review through normal GitHub PR workflows.
Here's what I did this morning: I had a refactor I was avoiding because it touched 14 files and I was pretty sure I'd miss something. I opened the Copilot app, pointed it at the branch, described what I wanted, and let it run.
It made the changes. It opened a PR. I reviewed the diff in about 15 minutes instead of the two hours the manual refactor would have taken.
The key difference from just using Copilot in VS Code: the session has persistent context across a larger scope of work. I wasn't doing one-off completions — I was running an agent against a multi-file task and getting a reviewable artifact at the end.
Getting into the technical preview:
The setup took about 5 minutes once I had access — install the app, authenticate with your GitHub account, pick a repo to connect.
I want to be precise about this: I'm not replacing OpenClaw with GitHub Copilot. They do different things.
The workflow I'm settling into:
The agent isn't fast. For a 14-file refactor, it took about 20 minutes of compute time. If you're in a hurry, this isn't the tool.
The session workspace is persistent but not infinitely flexible. If you have a task that requires non-code artifacts (database migrations, infrastructure changes), you need to be explicit about what you want and check the diff carefully.
It doesn't replace your code review judgment. I caught one edge case it missed in my refactor — the PR diff made it obvious. This is a tool for making the mechanical parts of coding faster, not for thinking about whether the change is right.
If you're a developer who spends time on multi-file refactors, dependency updates, release note generation, or any other routine mechanical task that doesn't require deep creative decision-making — this is worth 30 minutes of your time to set up and test on one task.
If you're already using Copilot completions in your IDE and that's working fine for you, the desktop app adds most value when you have tasks that are too large for a single prompt but too mechanical for the kind of deep thinking where you'd want to control every line.
Try it: github.com/features/copilot → look for the desktop app technical preview signup. Test one routine task (dependency update, standard refactor, release notes) and measure time saved. For OpenClaw users, think of it as a focused coding agent that lives inside your GitHub workflow rather than alongside your entire digital life.