GitHub brings remote control to Copilot CLI as coding agents move beyond the terminal

GitHub has added remote control for Copilot CLI sessions on web and mobile, letting users monitor and steer a live terminal session from another device.

In public preview from April 13, the feature mirrors a running CLI session to web and mobile via GitHub, generating a link or QR code that opens a live view of the session on another device, keeping actions synced between the terminal and the web or mobile client.

Users can send follow-up instructions, review and modify plans, switch modes, respond to prompts, or stop the session altogether.

The rollout reflects a broader change in how AI coding agents are being used. Anthropic, for what it’s worth, has already been pushing in that direction with Claude Code, recently introducing remote access to live local sessions and positioning the tool as something that can run continuously in the background while remaining accessible across devices.

Developers ask, developers get

GitHub’s latest update, effectively, turns Copilot CLI into a session that can be managed from almost anywhere, rather than a tool confined to the shell where it started.

The company even suggests using the /keep-alive command within a Copilot session to support longer-running tasks, which is a telling detail in itself — it’s about supporting coding jobs that take long enough to enable a user to walk away for more pressing issues, and come back later.

In truth, this is something developers had already been asking for. In the wake of Anthropic’s Claude Code remote access rollout, users took to GitHub to ask for similar functionality, pointing to long-running Copilot CLI tasks and the need to check in and guide them from another device.

“I often have one or more Copilot CLI jobs running while I do something else or leave the machine - potentially for a while if it is running performance tests or a long build or analysis,” one user wrote, adding that they’d like a way to access an active copilot CLI session from their phone.

“It would be amazing – something like Claude Code's remote control feature for Copilot CLI,” another user responded.

Fast-forward six weeks, that request has been granted.

How to access remote control in Copilot CLI

For those looking to try it out, GitHub says users will first need to update to the latest version of Copilot CLI by running /update. From there, a remote session can be started with copilot --remote, or enabled within an existing session using /remote.

The feature currently requires the working directory to be a GitHub repository, after which a session link can be opened on web or mobile to follow along and interact in real time. As noted, GitHub also recommends using /keep-alive to ensure the machine stays active while the CLI is running.

Mobile access is available through beta releases on Google Play and iOS TestFlight, while Copilot Business and Enterprise users may need an administrator to enable remote control and CLI policies before the feature can be used.

The community reacts: ‘Only in GitHub repos’

Early reaction to the feature has been broadly positive, with many users welcoming the ability to step away from a running session and check back in remotely. But the responses also highlight some of the limits that come with it.

One sticking point is that the feature currently requires the working directory to be a GitHub repository, which rules out projects hosted elsewhere or kept local. “Oof, only in GitHub repositories :(” one user noted. Others took a more pragmatic view, with one replying that they had already moved dozens of projects over to GitHub “because this is where all the new goodies will be” — a shift that may not be realistic for teams still tied to platforms like Azure DevOps.

Elsewhere, developers are already pushing the idea further. One user said they wanted Copilot to send a text or Slack message when approval is needed, allowing them to respond without returning to the session directly. Another said that they had already built a version of this themselves, wiring Copilot into Telegram to handle approvals remotely.

That points toward a more messaging-driven model, similar to tools like OpenClaw, where agents run continuously and interact with users through the chat apps they’re already using, bet that WhatsApp, Slack, or Telegram.

This would represent a small step beyond remote control, but it strongly hints at where this is heading — less time spent watching these sessions run, and more emphasis on pulling the developer back in only when needed.