How to Use Google Antigravity

To use Google Antigravity, download the desktop application, command-line interface (CLI), or editor extension from Google, authenticate with a Google account, and initiate tasks using plan mode. The agent framework coordinates planning, code generation, and terminal command execution across macOS, Windows, and Linux environments.

Because the autonomous agent modifies files and executes shell commands directly, review the plan and artifacts before you let it run. Setup requires selecting an operational surface, configuring directory permissions, and managing rate limits across free or paid subscription tiers.

How to Use Google Antigravity, Step by Step

Using Google Antigravity requires selecting an operational client, authenticating your Google profile, opening a target project directory, and managing execution through plan mode. Google provides multiple ways to interact with the agent, but every workflow follows the same sequence from initial prompt to reviewed code changes.

Setup starts from Google's own Antigravity blog and docs, linked in step 1. After installing your chosen surface, you authenticate through an interactive browser window and grant the agent access to a local repository folder.

Once a project is active, entering /plan directs the agent to generate an implementation plan before touching local code. Review this plan and the artifacts before the agent runs, so you can catch a destructive command before it executes.

• 1. Download and install a client surface from the Google Antigravity 2.0 release or the documentation overview.
• 2. Sign in using a standard Google account through the automated browser sign-in prompt.
• 3. Open a project workspace folder to grant the agent filesystem access to your repository.
• 4. Submit your initial development task in plan mode by typing the /plan command in the prompt input.
• 5. Inspect the generated implementation plan, proposed terminal operations, and file artifacts before approving execution.
• 6. Authorize the agent to apply changes, verify code diffs, and run commands inside the terminal sandbox.

Selection of an Antigravity Surface

Developers can access Google Antigravity through four primary surfaces: the standalone 2.0 desktop application, the integrated development environment (IDE), the command-line interface (agy), or editor extensions. Choosing the right surface depends on whether you prefer full agent orchestration, an all-in-one editor, or lightweight terminal commands.

The Antigravity 2.0 desktop application released on May 19, 2026, functions as a standalone application without an embedded code editor, succeeding the IDE's Agent Manager. The IDE itself is not deprecated; Google recommends developers keep the IDE, though Google announced the Agent Manager will be removed from the IDE in an upcoming release. Editor extensions released on August 20, 2026, bring Antigravity capabilities to VS Code, Visual Studio, Zed, JetBrains 2026.2.1+, and Xcode 27 beta 6+ across all plan tiers.

Detailed surface breakdowns appear in our Google Antigravity 2.0 guide, Antigravity CLI guide, and Antigravity vs VS Code comparison.

Installation on Windows, macOS, and Linux

Google Antigravity supports native installation across macOS, Windows, Linux, and Googlebook systems for both graphical user interface (GUI) applications and command-line tools. The CLI installs with a one-line script per OS, listed below.

The Antigravity CLI, executed as agy, is written in Go and uses the same agent harness and settings as the 2.0 desktop application. Conversations started in the CLI open in the desktop app.

Google provides no native mobile application for iOS or Android operating systems (OS). Mobile management is supported exclusively through the Remote Control web interface introduced on August 21, 2026, which allows developers to drive an active session running on their primary workstation from any mobile browser.

• macOS, Linux, and Googlebook CLI: Run curl -fsSL https://antigravity.google/cli/install.sh | bash in your terminal.
• Windows CLI: Run irm https://antigravity.google/cli/install.ps1 | iex inside Windows PowerShell.
• Headless Environments: Authenticate by exporting GEMINI_API_KEY and setting "modelProvider":"gemini" in settings.
• Desktop app and IDE: Start from the Google Antigravity 2.0 launch post; both run on macOS, Windows, Linux and Googlebook.
• Mobile Access: Use Remote Control to drive a session running on your own machine from any web browser. It also sends push notifications.

Task Execution with Plan Mode and Slash Commands

Your first task goes in as a prompt. Start it in plan mode so the agent proposes before it acts.

Using /plan initiates plan mode, producing an implementation plan. You read that plan before the agent changes anything.

The slash-command list also includes /goal and /grill-me; check the documentation overview for what each does before relying on it. For recurring work, /schedule sets up automated tasks via cron, and subagents execute parallel routines while inheriting the main agent's permission boundaries. Antigravity also supports multi-folder projects with independent folder permissions and features live voice transcription for verbal prompting.

• /plan: Switches to plan mode, which produces an implementation plan for you to review.
• /schedule: Configures cron-based background jobs for automated tasks.
• Subagents: Run in parallel and inherit the main agent's permissions, so anything the main agent may do, they may do too.
• Multi-Folder Workspaces: Manages repositories that span separate filesystem paths, applying unique access rules to each directory.

Using the Browser Agent (and the Browser Extension Question)

Google's documentation lists a browser for the agent that runs in a separate Chrome profile with allow and deny lists, and /browser is on the slash-command list. Google's announced Antigravity extensions (August 20, 2026) are for code editors: VS Code, Visual Studio, Zed, JetBrains 2026.2.1+ and Xcode 27 beta 6+.

The separate Chrome profile keeps your own logged-in sessions out of the agent's reach: it is not browsing as you.

The allow and deny lists limit which sites the agent can open. For developers asking if Google Antigravity runs entirely inside a browser, Google provides no browser-only web application; web-based interaction is limited to driving a local desktop instance through Remote Control.

Artifact Inspection and Execution Safeguards

Reviewing generated artifacts and execution plans before the agent runs gives you a chance to catch a destructive command. Google's documentation lists artifacts, which you review, and a terminal sandbox.

Hooks allow developers to execute JSON-configured shell scripts before and after model and tool invocations, with workspace-level hooks taking precedence over global configurations. These guardrails provide programmatic checks against unexpected agent operations.

Two reported incidents show why you should review before you run. In December 2025, The Register reported that a command run in Turbo mode wiped a user's drive. Additionally, in April 2026, CyberScoop reported a Pillar Security finding in which prompt injection led to code execution even in strict mode (fixed February 28, before the April disclosure). In February 2026, VentureBeat also reported account bans for users who connected the third-party agent OpenClaw.

• Inspect Artifact Diffs: Read every proposed file change before the agent applies it, looking for unexpected deletions.
• Verify Terminal Commands: Read shell commands generated by the agent before authorizing execution, avoiding blind approval in Turbo-mode.
• Configure Hooks: Add JSON-configured shell scripts before and after tool and model calls as guardrails.
• Don't Lean on the Sandbox Alone: Pillar Security got code execution even in strict mode, so keep reading commands.
• Skip Third-Party Agents: Google banned accounts that connected OpenClaw in February 2026.

Plans, Rate Limits, and Quota Management

Google Antigravity organizes usage across free Individual accounts, paid Google AI Pro and Ultra subscriptions, and Google Cloud business plans. The Individual plan provides basic weekly rate limits alongside unlimited Tab completions and Command requests at $0 per month, with quotas refreshing weekly.

The Google AI Pro tier costs $20 per month and refreshes rate limits every 5 hours until a broader weekly cap is reached. The Google AI Ultra tier offers a $100 per month option with five times Pro's token rate limit, and a top $200 per month tier (reduced from $250) providing twenty times Pro's allowance. Both Ultra tiers refresh every 5 hours with the highest weekly limits.

Google does not publish exact numeric token allowances, request counts, credit-pack prices, or student discounts for any Antigravity tier. Confirm current plan structures directly on the Google Antigravity pricing page and plan documentation.

Users should monitor account limits carefully. In January 2026, Google acknowledged a bug on its developer forums where 5-hour refresh cycles caused multi-day lockout loops. Users also encounter HTTP 429 RESOURCE_EXHAUSTED errors resulting in agent execution terminations. The AI Credit Overages setting decides what happens when you reach a limit: Never stops usage there, and Always continues on overage billed at Gemini Enterprise consumption pricing. Google offers no bring-your-own-key option for extra limits. Review our Google Antigravity pricing guide for detailed tier comparisons.

Integration of Skills and Model Context Protocol

Custom skills and Model Context Protocol (MCP) servers extend Google Antigravity with specialized development workflows and external tools. If you are moving from Gemini CLI to the agy CLI, skills move from .gemini/skills to .agents/skills, MCP servers move to mcp_config.json, and GEMINI.md and AGENTS.md files still apply. Further implementation details appear in our Google Antigravity skills guide.

Google Antigravity is not suitable for developers who only want simple, non-agentic inline code autocompletion inside their existing editor. Those developers are better served by standard editor extensions or standalone code editors like VS Code, as examined in our Cursor vs VS Code comparison.

Next step: install the agy CLI with the one-liner for your OS (or the desktop app via the 2.0 launch post), sign in with your Google account, open one small project folder, and run your first task with /plan.