Google’s Antigravity coding app can now keep working after you shut your laptop, and you can talk to it while it works. Version 2.29 of the Antigravity App, released on November 3, adds remote agents that run in Google’s cloud and a voice mode built on Gemini Live. Both are rolling out gradually, and remote agents need a paid Google AI Pro or Ultra plan on a personal Google account.
If you already use Antigravity for side projects, remote agents are the bigger change. Voice mode is the more fun one, and it arrived with a sensible set of guardrails.
What remote agents in Antigravity actually do
Until now, an Antigravity agent ran on your own machine. Close the lid or lose Wi-Fi and the task stalled. Remote agents move the whole session, including the tools and commands, into an isolated environment that Google manages. Google says they keep working on long tasks even when you close your browser, put your laptop to sleep or switch between your desktop and phone.
A remote agent does not see your local files. It works with resources you attach to a remote project: GitHub repositories, Google Drive files or files you upload. In practice that makes it a good fit for repository work, such as fixing a failing test, tidying a backlog item or preparing a pull request, and a poor fit for anything that depends on a local dev server or uncommitted changes on your disk.
You can start one in three places:
- Antigravity App: click + next to Projects, choose New Remote Project, optionally attach a GitHub repo, Drive files or uploads, name it and send a prompt. You can also pick Remote, then GitHub Repositories, from the project selector in the message box.
- The web: go to antigravity.google.com, sign in with an eligible personal account and start a conversation or create a project. This also works from a phone browser.
- Antigravity CLI: update to version 1.4.12 or later with
agy update, then runagy --remote-agent. Add--project="name"to use an existing remote project, or--new-projectto create one named after the current folder.
A conversation started on one surface can be picked up on another, so you can kick off a job from the CLI and check on it from your phone later.
Remote agents vs Remote Control
Antigravity already had a feature called Remote Control, which can be confusing. Remote Control lets you drive a session that is still running on your own computer from any browser, so that computer has to stay powered on and online. Remote agents need no machine of yours at all. If your task needs local hardware, local toolchains or a checkout that is not pushed yet, Remote Control is still the right tool. If the code lives on GitHub, a remote agent is simpler.
Who can use remote agents
Google’s documentation is specific here. Remote agents are available to Google AI Pro and Google AI Ultra subscribers who sign in with a personal Google account, and the feature is rolling out gradually. It is not available on the free plan, on Google AI Plus or for business accounts on Google Workspace or Google Cloud.
That matters in India, where many developers get Google AI Plus or a free tier, and many others use a company Workspace login. Neither gets remote agents for now. If you are on Pro or Ultra and still do not see New Remote Project, the staged rollout is the likely reason.
How voice mode in Antigravity works
Voice mode turns a conversation into a spoken back-and-forth with the agent. It runs on the Gemini Live model, which hears your audio directly rather than working from a transcript, and it answers out loud in real time. You can interrupt it just by talking.
The agent still does the work in the conversation. It reads code, edits files and runs commands, and everything it writes appears on screen as usual, so you can listen while you look at your code and read the details when you need them. Voice mode also continues the conversation you are already in, including messages you typed earlier.
To start it in the app, clear the message box and click Start Voice mode. The first time, you pick a voice. In the CLI, version 1.5.0 or later, run /voice live or press F6.
Some things you can say once it is running:
- Give a task, such as asking it to find the failing test and fix it.
- Steer mid-task: “just check it, don’t change anything yet.”
- Ask “What are you doing?” to hear a status update.
- Say “stop” to end the running task, or “hold on” when you need a moment to read.
- Say “keep me posted” for a short update roughly every 30 seconds. By default it stays quiet and speaks only when it has an answer or needs you.
Google offers nine voices and playback speeds from 0.75x to 2x. Option+M on macOS or Alt+M on Windows and Linux mutes your microphone, and Shift+Option+M or Alt+Shift+M mutes the agent so you can read instead. If you switch to another conversation, voice mode follows you in a small floating dock.
Voice approvals have limits, and that is good
When the agent needs permission for an action, it reads the request aloud with its recommendation and waits. You can say “allow it just this once,” “allow it for this conversation” or “no, run the tests first.”
Destructive actions are the exception. Deleting files, running commands with elevated privileges and force-pushing to a Git remote cannot be approved by voice. You have to click Allow on screen, though you can still deny them by voice. With an agent that can run shell commands, a misheard “yes” should never be enough to wipe a folder, so this is the right call.
Voice mode limitations to know
- In the app, voice mode works only in local conversations. It is not available in Remote Control sessions or in conversations that run in the cloud, so you cannot talk to a remote agent yet.
- It can be on in only one conversation at a time, and not in a subagent’s conversation or a side question.
- In the CLI, it does not work in print mode or in remote agent sessions.
- The Start Voice mode button disappears when you are signed out or have hit your usage limit.
On privacy, Antigravity streams your microphone audio to Google’s servers, where Gemini Live processes it. Transcripts appear in a voice panel during the call and in a Voice transcript card saved on your computer afterwards. If you work on client code under strict rules, check whether sending spoken audio to Google is allowed before you turn this on.
Other changes in Antigravity 2.29
The release also redesigns the new-conversation page with the message box pinned to the bottom, reorganizes Automations into Scheduled, Always on and Paused lists, opens Jupyter notebooks natively and lets you customize keyboard shortcuts. The new default look uses the Onyx theme and Google Sans font, which you can change under Settings, then Appearance. It arrives a few weeks after Google added Gemini 4 Argon for Pro and Ultra users in Antigravity.
Should you try it?
If you pay for Google AI Pro or Ultra and keep your code on GitHub, remote agents are useful for long test runs and refactors that you want to check from another device. Review the changes before merging them. If your project depends on local files or a dev server, stick with Remote Control.
Voice mode is easier to judge. Google does not restrict it to Pro and Ultra the way it does remote agents, it keeps risky approvals on screen, and it suits debugging sessions where you would otherwise be talking to yourself anyway. The main gap is that you cannot use it with remote agents yet, which is the combination most people will want next.
