Antigravity Remote Control: What It Does From Your Phone, and What It Does Not

Since Google shipped Remote Control with Antigravity 2.0 on August 21, 2026, "antigravity remote control" has become the second most frequent query that brings people to this site, after our own name: about 1 900 impressions in four weeks, and almost nobody clicking, because the page Google was showing them is about our remote, not Google's. That is fixed now. Here is what Antigravity Remote Control is, checked line by line against the documentation on September 22, what it does not do, and how it fits next to the remote we build.

What it is, in one paragraph

Remote Control lets you reach the Antigravity sessions running on your computer from a web browser, on any device, without SSH and without opening a port. The machine keeps its files, its build tools, its credentials and its environment variables; the browser gets a dashboard signed in with the same Google account. Google announced it for Antigravity 2.0 and for the Antigravity CLI on the same day, and the documentation covers both.

You can do five things from the dashboard: read the active conversations, start a new agent task, review an implementation plan, inspect the artifacts the agent produced, and answer what the agent is waiting for, whether that is a tool confirmation or a question. A prompt typed on the phone streams into the local terminal, and a confirmation given on the phone unblocks the terminal at the desk. Both sides stay in sync.

Turning it on

There are three ways in, and they map to three habits.

In Antigravity 2.0, the editor, it is a switch: Settings, App section, "Enable Remote Control". You can give the machine a nickname there, which is what you will see in the dashboard's instance switcher.

On the Antigravity CLI, inside a session you already have open, type /remote-control. The tunnel opens for that session and closes when the session ends, or when you type /remote-control off. The agy --remote-control flag does the same thing at launch time. The machine gets an auto-generated name per session, something like devbox-swift-falcon.

If you want the tunnel to outlive the terminal, run the daemon instead:

agy remote-control start --name office-mini
agy remote-control status
agy remote-control stop

start registers a background service: a systemd user service on Linux, a LaunchAgent on macOS, a Scheduled Task on Windows. Two details in the documentation worth reading twice: on macOS the daemon only starts at login, not at boot, and on Windows it needs an Administrator install to start at boot, otherwise it too waits for a login. On Linux it comes up with the machine. The --name flag sets the machine name; the cliRemoteControlHostname entry in the CLI config file does the same permanently.

The daemon is limited to one per machine.

Connecting from the phone

Open the Remote Control dashboard in the phone's browser and sign in with the Google account of the machine. This sign-in is separate from the editor's login, which is the first thing to check when the dashboard shows nothing. Every machine that runs the toggle or the daemon appears in the instance switcher, so a laptop and a desk machine sit side by side under the same account.

It is a web app, not a store app. On Android and on iPhone, add the dashboard to the home screen; installed that way it can send push notifications, and that is the part that makes it usable: you get a notification when an agent has finished its turn or is waiting for you, you open the dashboard from the notification, you answer, you put the phone back in your pocket.

Three limits worth knowing before you rely on it

It reaches Antigravity only. The dashboard sees Antigravity 2.0 sessions and Antigravity CLI sessions. A Claude Code session on the same machine is invisible to it, and so is a Codex, Cursor or Copilot CLI session. Claude Code has a Remote Control of its own, driven from the Claude app; we compared the two in our article on Claude Code Remote Control.

Settings stay on the CLI. You cannot flip Remote Control on or off from the dashboard, nor change the machine name from there. That is a CLI or editor action, at the machine.

It is rolled out account by account. The documentation lists Remote Control as a feature flag enabled per account or organization. If the toggle is not in your Settings, your account is not in yet. There is nothing to reinstall.

One more, which is a property rather than a limit: an interactive /remote-control tunnel dies with its terminal. If you close the laptop lid and the session ends, the phone shows the machine offline. The daemon is the answer when you want the machine reachable after you have walked away.

How we use it next to the AgentsRoom remote

We run Antigravity agents inside AgentsRoom, the same way we run Claude Code, Codex and the other CLIs: each agent is a terminal in the app, with a role, a model and a status. The AgentsRoom mobile app is a remote for those terminals. It pairs with the desktop over a relay that is end-to-end encrypted, forwards what you type into the local terminal of the agent, and streams the output back. The agent does not know a phone is involved, which is why it works for all 14 supported CLIs on the day they are supported, Antigravity included.

So for an Antigravity agent we have two remotes, and they answer different questions.

Google's Remote Control gives us the Antigravity view of the session: the conversation as the editor renders it, the plan to review, the artifacts, the confirmations, with the account-level push notifications. It is the better screen when the work is a single Antigravity session and the question is "what did it decide, and does it need me".

The AgentsRoom remote gives us the machine: the raw terminal of that agent, the Claude Code agent running next to it, the backlog ticket the task came from, the diff of what was written and the commit button, the tunnel to the dev server the agent just started. It is the better screen when several agents share the machine, or when the answer to the notification is "merge it" rather than "yes, go on". Nothing runs on the phone and nothing is re-hosted: the agent keeps executing locally, and the relay only ever sees encrypted bytes.

Both can be on at the same time on the same session. The split that makes sense to us: the Antigravity daemon on the machine that hosts the agents, so that its turns reach the phone through Google's notifications, and the AgentsRoom app for everything else on that machine. One notification stream from Google for the Antigravity turns, one from AgentsRoom for the rest of the fleet.

What to remember

• Remote Control shipped on August 21, 2026 for Antigravity 2.0 and the Antigravity CLI: a web dashboard signed in with your Google account, installable on the phone's home screen for push notifications.
• Turn it on in the editor's Settings, with /remote-control in a session, or with the agy remote-control start daemon when you want it to outlive the terminal.
• It drives Antigravity only, one daemon per machine, and its settings live on the machine, not in the dashboard. A missing toggle means your account is not in the rollout.
• For the rest of the machine, the AgentsRoom remote agent control covers every CLI at once, and the two remotes coexist on the same Antigravity session. If your question is about running two Google accounts on the CLI, that one has its own article.