GitHub Copilot in VS Code: cloud agents, Agent Host and computer use
GitHub has expanded how Copilot agents run in VS Code. Cloud sessions can be picked up across VS Code, GitHub Mobile and github.com, while the new Agent Host keeps local and remote agents running after you close a window. Computer use and side-by-side model comparison arrive as preview or experimental features.
Key points
- Cloud agent sessions can be continued across VS Code, GitHub Mobile, github.com and Copilot CLI.
- The Agent Host keeps agents running after you close a window, and it also works on a remote machine.
- To control a development machine from your phone, run code agent host --github-environment from the project folder.
- The Copilot harness is now the default for new sessions in VS Code and the Agents window.
- Computer use is experimental: enable chat.agentHost.computerUse.enabled, restart VS Code and run /computer on.
- Side-by-side model comparison with a Judge model and Synthesize is in preview.
GitHub has widened the GitHub Copilot experience in VS Code around three things: which models you use, where your agents run, and how closely you follow them. Cloud agent sessions run on GitHub's infrastructure, so they keep going after you close your laptop, and you can pick the same session up in VS Code, the GitHub Copilot app, GitHub Mobile, github.com or Copilot CLI. A new Agent Host separates sessions from individual editor windows, and agents can also run on a remote machine reached through Remote SSH or dev tunnels.
The Copilot harness, built on the Microsoft Copilot SDK, is now the default for new sessions in both the editor and the Agents window, which should make behavior more consistent across clients. Several features are still early: side-by-side model comparison with a Judge model is in preview, computer use lets the agent click and type in desktop apps but is experimental and needs approval per app, and the plugin marketplace and Copilot connectors are in preview. Administrators can restrict or disable several of these through policy.
Real-world use
- Handing a long refactoring task to a cloud agent, closing the laptop and checking the result later from github.com.
- Running an agent on a stronger remote machine while continuing other work on your primary computer.
- Running the same coding task with two models in parallel and comparing their results before merging.
Why it matters
Agent sessions no longer depend on your machine staying awake or on one window staying open, so you can keep steering the same session from VS Code, your phone or the web. The shared Copilot harness should also make agent behavior more consistent across these clients.