Skip to content
Remote Control
Feature overview2-0Antigravity AppFeaturesFeature overview Antigravity App FeaturesAntigravity AppAntigravity CLI
Available on:
Antigravity AppAntigravity CLI

The Antigravity App and the Antigravity CLI run the same agent, so most features work on both. This page starts with the features they share, and then covers the features that are specific to the Antigravity App and the Antigravity CLI.

Where a feature works differently in the App and the CLI, select your surface in the tabs. Your choice applies to every tab on this page and is remembered across the docs.

Agents work in projects, which define the folders an agent can access and the settings and permissions it uses.

  • Worktree support: Projects natively support Git worktrees, allowing agents to operate in isolated background folders.
  • Scoped settings: Settings are scoped, allowing you to have different security settings per project. This means you can have a more permissive setting for a trusted project and a more restrictive security setting for an untrusted folder. For the available presets, refer to Permission presets.
  • Scoped permissions: Attach permission grants to projects to control what the agents are allowed to access. Permissions manually granted during a conversation can persist, allowing the agent to learn trusted actions over time.
  • Multi-folder access: You can configure a project to work in multiple folders, allowing agents to operate across different codebases within the same conversation.

Learn more about projects.

By default, agents ask before sensitive actions and limit their file access to your project’s folders. The details differ by surface:

  • Interactive approvals: By default, agents request your explicit permission before running terminal commands outside the isolated Terminal sandbox. With the previous permission settings, agents ask before running any terminal command.
  • Bounded access: By default, your agent can only read and write within the provided folders of a project. If you broaden your permission settings (for example, the “Turbo” preset, or the “Full machine” or “Turbo mode” presets in the previous permission settings), the agent has read and write access over your full machine.

Learn more about permissions and the terminal sandbox.

The main agent can delegate work, such as research, builds, or test runs, to subagents. Each subagent starts with a clean context and runs in its own conversation, in parallel with your main conversation.

In the side panel’s Overview, open Subagents and select a subagent to follow its conversation.

Learn more about subagents.

The browser subagent drives Google Chrome to test and debug web pages:

  • On-demand: Invoke the browser subagent through the /browser command.
  • Chrome DevTools integration: Integrate natively with Chrome DevTools MCP.
  • Video recording: Record browser sessions as WebM videos.

Type / in the App’s message box or the CLI’s prompt box to run commands such as /plan, /grill-me, /goal, /boost, /schedule, and /teamwork-preview. Some commands are available on only one surface, and /boost and /teamwork-preview require a paid plan.

Browse all slash commands.

Antigravity features built-in live voice transcription, allowing you to prompt agents and leave feedback using natural speech.

  • Start or stop: Click the microphone button next to the message box to start recording, and click it again to stop. You can also press Ctrl+M to start recording, and press it again to stop.
  • Live view: As you speak, your words are transcribed in real time directly into the message box.
  • Conversational awareness: Antigravity gives the speech recognizer your recent conversation and the text around your cursor, so project-specific terms are more likely to come through correctly.
  • Where it works: Use voice in the message box to start conversations and send prompt updates, and in artifact comments to leave precise, inline feedback on plans, code diffs, and deliverables.

Remote Control allows you to drive and monitor your Antigravity App sessions and Antigravity CLI instances across multiple machines from any web browser:

  • Work from any device: Start long-running agent work on your workstation, and then monitor it or approve actions from a phone or laptop.
  • Keep your local setup: Agents keep full access to your workstation’s files, toolchains, credentials, and Git worktrees, without a separate environment.
  • Get notified: Receive browser push notifications when tasks complete or when your input is needed.

To turn on Remote Control:

In Settings, open Application and turn on Enable Remote Control. You can then enter a Device Name to identify the machine.

Learn more about Remote Control.

Extend what the agent knows and can do. These customizations work in both the App and the CLI:

  • Skills: Give the agent folders of instructions, each with a SKILL.md file, that it follows when working on specific tasks.
  • Rules: Set persistent instructions, coding standards, and architectural constraints that apply to every conversation.
  • MCP servers: Connect the agent to local developer tools, databases, and remote APIs through the Model Context Protocol.

JSON hooks allow you to execute custom local shell scripts at critical stages of an Antigravity agent’s execution cycle. You can intercept and control the agent’s behavior before tool calls, after model responses, or at loop stopping conditions—configured globally or per workspace using simple JSON files.

Learn more about hooks.

Plugins are namespaced bundles that can contain skills, agents, rules, MCP servers, and hooks as a single deployable unit.

In the left sidebar, click Customizations, and then open the Marketplace tab to browse and install plugins. Plugins that you install in the App also appear in the CLI.

Learn more about plugins and the Marketplace.

These features are specific to the Antigravity App.

Start quick, one-off conversations outside of any project. These sessions run in a local scratch folder. They have their own settings and permissions in addition to inheriting from global permissions.

Automations (previously called scheduled tasks) let you plan ahead with your projects by sending messages to your agents on a schedule, even while you’re away. Open them from Automations in the left sidebar:

  • Repeatable: Set up time-based triggers to start conversations periodically.
  • Model tier: Each automation runs on the model tier you choose: Flash Lite, Flash (the default), or Pro.
  • Approximate timing: Runs start close to their scheduled time rather than on the exact minute. Antigravity spreads them out by up to 10% of the interval, and by no more than 15 minutes.
  • Run on demand: Run an automation right away with Run Now from its menu, or select Create with Prompt to have the agent walk you through setting up a new one.

For a timer or recurring check inside a single conversation, in either the App or the CLI, use the /schedule command.

Use the built-in browser to check your running app or read documentation in the side panel, next to your conversation, and to point the agent to exactly what you see on a page.

The Antigravity App with a conversation on the left, where the agent answers with a localhost link to a Bookmarks web app, and the built-in browser in the side panel on the right showing that app

  • Open a page: Click a web link in an agent response. To open a new browser tab, press Cmd+T (macOS) or Ctrl+T (Windows and Linux), or click + in the side panel, and then choose Browser. To open a link in your default browser instead, hold Cmd (macOS) or Ctrl (Windows and Linux) while you click it.
  • Go to a URL or search: Type in the address bar. For local servers that don’t use HTTPS, include http://, for example http://localhost:3000.
  • Navigate: Use Back, Forward, and Reload in the toolbar, or click Open in external browser to open the current page in your default browser. Right-click a link to open it in a new tab or in your external browser.
  • Keep sign-ins separate: The built-in browser keeps its own cookies and sign-ins, separate from your Chrome profile.

To point the agent to part of a page, select text and click Comment (C) to leave feedback, or Quote (Cmd+L on macOS, Ctrl+L on Windows and Linux) to add the text to your message. The agent receives the page URL and the selected text with your next message.

Text selected on a web page in the built-in browser, with a popup that offers Comment (C) and Quote (Ctrl+L)

To have the agent browse and test pages for you, use the /browser command, which runs the browser subagent in Google Chrome.

PDF files open in the side panel, so you can read a spec or paper next to your conversation and point the agent to specific passages. To open a PDF from your project, click it in an agent response, or click + in the side panel, choose Open File, and search for the PDF. Links to PDFs on the web open in the built-in browser.

A Bookmarks API guide PDF open in the side panel of the Antigravity App, with a toolbar at the bottom for page thumbnails, page navigation, zoom, and fit mode

  • Navigate: Use the toolbar at the bottom of the PDF to show page thumbnails, change pages, zoom, and fit the page or the width. To search the PDF, download it, or copy its path, use the More actions menu at the top.
  • Comment or quote: Select text and click Comment (C) or Quote (Cmd+L on macOS, Ctrl+L on Windows and Linux), the same as on a web page.

Open a terminal in the side panel: click + and choose Terminal. To show or hide the terminal at any time, press Cmd+J (macOS) or Ctrl+J (Windows and Linux).

The Antigravity App provides a Git-native Review pane for the changes in your project. Open it from any conversation by clicking Review () in the Overview panel, and then perform the following actions:

  • Compare views:
    • Uncommitted: View all staged and unstaged working tree changes.
    • Branch: Review all changes made on the current branch.
    • Agent Edits: Inspect the files that the agent modified in the conversation.
  • File actions: Stage, unstage, or discard unstaged changes per file.
  • Repo actions:
    • Commit: Create local commits, with a generated commit message if you leave the message empty.
    • Amend: Add changes to the latest commit.
    • Push: Send branch commits to a remote repository.
  • Feedback: Comment on lines in a diff to send feedback to the agent.

These features are specific to the Antigravity CLI.

Run agy to work with the agent in an interactive terminal UI. Besides the shared slash commands, the CLI has commands for managing your session, such as /resume, /rewind, /model, /context, /codesearch, /tasks, and /usage.

  • Keyboard shortcuts: Customize shortcuts with /keybindings, or edit prompts with modal Vim keybindings in Vim editor mode.
  • Execution modes: Choose an execution mode to control whether the CLI pauses to ask before it modifies files or runs commands.

Browse all CLI commands.

Run /diff to open the interactive diff viewer, where you can inspect workspace changes, commit history, and the changes from each agent turn, and add line-by-line comments to steer the agent’s next steps.

Learn more about the diff viewer.

Run the CLI non-interactively to script agent tasks, integrate with CI pipelines, and capture machine-readable output. Pass a prompt with -p (or its aliases --print and --prompt) to run it once and exit.

Learn more about headless mode.

Configure the CLI in ~/.gemini/antigravity-cli/settings.json. For example, in addition to your permission mode, you can define specific commands that the agent is allowed or denied to run:

{
  "permissions": {
    "allow": ["command(git)", "command(npm test)"],
    "deny": ["command(rm -rf)"]
  }
}

You can also pipe live agent metadata (JSON that contains the current working directory, active model, token usage, state, and more) into your own shell scripts to generate a custom status line or terminal window title.

Learn more about CLI settings.