How to Use Google Antigravity

Google Antigravity is the most autonomous AI coding tool ever released. Unlike traditional code assistants that suggest completions, Antigravity deploys autonomous agents that plan, code, test, and verify entire features across your editor, terminal, and browser.

Sounds powerful? It is. But with great power comes great configuration complexity.

In this tutorial, I’ll walk you through:
– Installing Antigravity (5 minutes)
– Configuring security settings (10 minutes)
– Spawning your first agent (5 minutes)
– Building a real project (10 minutes)
– Pro tips and best practices

By the end of this 30-minute guide, you’ll have Antigravity running and understand how to leverage autonomous agents without breaking your system.

Let’s get started.

---

Prerequisites

Before installing Antigravity, make sure you have:

✓ Operating System: macOS, Windows, or Linux (Ubuntu/Debian)
✓ Gmail Account: Personal Gmail required (not Google Workspace)
✓ Storage: 500MB free disk space
✓ RAM: 4GB minimum (8GB recommended)
✓ Browser: Chrome or Chromium-based browser
✓ Internet: Stable connection for agent API calls

Programming Knowledge: Basic familiarity with terminal commands and your chosen programming language. Antigravity works best if you can review and debug generated code.

Optional:
– Existing VS Code setup (Antigravity can import settings)
– Cursor or Windsurf IDE installed (for comparison)

---

Step 1: Download Google Antigravity

1.1 Visit the Official Download Page

Navigate to the official download page:

URL: antigravity.google/download

1.2 Select Your Operating System

Click the download button for your OS:

• macOS: Antigravity-darwin-arm64.dmg (Apple Silicon) or Antigravity-darwin-x64.dmg (Intel)
• Windows: Antigravity-win32-x64.exe
• Linux: Antigravity-linux-x64.deb (Debian/Ubuntu) or .rpm (Fedora/RHEL)

File Size: Approximately 300-400MB, depending on the platform.

1.3 Verify Download (Optional but Recommended)

For security, verify the download’s SHA-256 checksum:

macOS/Linux:

shasum -a 256 Antigravity-darwin-arm64.dmg
Code language: CSS (css)

Windows (PowerShell):

Get-FileHash Antigravity-win32-x64.exe -Algorithm SHA256
Code language: CSS (css)

Compare the output with the checksums listed on the download page.

---

Step 2: Install Antigravity

2.1 Installation on macOS

1. Open the DMG file:
   open Antigravity-darwin-arm64.dmg
2. Drag Antigravity to the Applications folder
3. First launch security:
4. macOS may block the app (unsigned)
5. Go to System Settings → Privacy & Security
6. Click “Open Anyway” next to the Antigravity warning
7. Launch Antigravity:
8. Open from the Applications folder
9. Or use Spotlight: Cmd + Space, type “Antigravity”

2.2 Installation on Windows

1. Run the installer:
   Antigravity-win32-x64.exe
2. Follow the setup wizard:
3. Accept the license agreement
4. Choose installation directory (default: C:\Program Files\Antigravity)
5. Create desktop shortcut (optional)
6. Windows Defender warning:
7. If SmartScreen blocks, click “More info”
8. Click “Run anyway”
9. Launch Antigravity:
10. Desktop shortcut or Start Menu

2.3 Installation on Linux (Debian/Ubuntu)

1. Install via dpkg:
   sudo dpkg -i Antigravity-linux-x64.deb
2. Install dependencies if needed:
   sudo apt-get install -f
3. Launch Antigravity:
   antigravity

Linux Note: If you encounter permission issues, ensure your user has access to /usr/share/antigravity/.

---

Step 3: Initial Configuration

When you first launch Antigravity, you’ll go through a setup wizard. Here’s what to expect:

3.1 Sign In with Google

Screen: “Welcome to Google Antigravity”

1. Click “Sign in with Google”
2. Select your personal Gmail account (Google Workspace not supported in preview)
3. Grant permissions:
4. Access to Gemini API
5. Cloud storage for artifacts
6. Browser integration

Why is Gmail Required? Antigravity uses Google Cloud for agent orchestration and requires API authentication.

3.2 Choose Setup Flow

Screen: “Import Settings or Start Fresh?”

Option 1: Import from VS Code / Cursor
– Migrates extensions, keybindings, and themes
– Recommended if you’re switching from another IDE

Option 2: Start Fresh
– Clean slate with Antigravity defaults
– Recommended for new users

My Recommendation: Import from VS Code if you have 10+ extensions. Otherwise, start fresh and add extensions as needed.

3.3 Select Editor Theme

Choose your preferred color scheme:
– Dark+ (default)
– Light+
– Monokai
– Solarized Dark/Light
– (Or install custom themes later)

3.4 Configure Agent Autonomy (Critical Step)

This is the most essential configuration. You’ll set how much freedom agents have.

---

Step 4: Agent Autonomy & Security Settings

Antigravity’s power comes from autonomous agents, but unrestricted agents can be dangerous. Configure wisely.

4.1 Terminal Execution Policy

Question: “How should agents execute terminal commands?”

Options:

1. Off (Most Secure)
– Agents never auto-execute terminal commands
– You manually run suggested commands
– Best for: Beginners, production systems, paranoid developers

2. Auto (Recommended)
– Agents request permission before executing
– You see the exact command before it runs
– You can allowlist safe commands for auto-execution
– Best for: Most developers (balance of safety and speed)

3. Turbo (Dangerous)
– Agents always execute unless the command is denylisted
– Fast but risky
– Best for: Sandboxed environments, throwaway projects

My Setup: I use Auto mode with this allowlist:

# ~/.gemini/antigravity/terminalAllowlist.txt
npm install
npm run dev
npm run build
npm test
git status
git add
git commit
git push
pytest
pip install
Code language: PHP (php)

Critical Denylist Commands (Always Block):

# ~/.gemini/antigravity/terminalDenylist.txt
rm -rf
sudo
curl | bash
wget | bash
dd if=
mkfs
Code language: PHP (php)

4.2 Browser Execution Policy

Question: “How should agents interact with web browsers?”

Options:

1. Off
– Agents can’t open browsers
– Disables web scraping, UI testing

2. Auto (Recommended)
– Agents can open a browser with permission
– URL allowlist required

3. Turbo
– Agents freely navigate any URL
– High security risk (prompt injection attacks)

My Setup: I use Auto mode with URL allowlist:

# ~/.gemini/antigravity/browserAllowlist.txt
localhost:*
127.0.0.1:*
*.github.com
*.stackoverflow.com
*.npmjs.com
*.pypi.org
Code language: PHP (php)

Why Allowlist Matters: Malicious prompts could trick agents into visiting phishing sites or exfiltrating data. Restrict to trusted domains.

4.3 Review Policy

Question: “When should agents request your review?”

Options:

1. Always Proceed
– Agents work fully autonomously
– You review only the final results

2. Agent Decides (Recommended)
– Agent requests review for complex/risky tasks
– Gemini determines what’s “risky”

3. Request Review
– Agent asks before every major decision
– Slowest but most controlled

My Setup: Agent Decides (trusts Gemini’s judgment)

---

Step 5: Install Browser Extension

Antigravity needs a Chrome extension for browser integration.

5.1 Automatic Installation Prompt

When you first spawn an agent that needs browser access, you’ll see:

“Google Antigravity Browser Extension Required”
[Install Extension]

Click “Install Extension”

5.2 Manual Installation (If Needed)

If auto-install fails:

1. Open Chrome
2. Navigate to: chrome://extensions/
3. Enable “Developer mode” (top right)
4. Click “Load unpacked”
5. Select: ~/.gemini/antigravity/chrome-extension/
6. Pin the extension to your toolbar

Verify: You should see the Antigravity icon next to your address bar.

---

Step 6: Your First Agent (Hello World)

Let’s spawn your first autonomous agent and build a simple project.

6.1 Open Agent Manager

Keyboard Shortcut: Cmd + Shift + M (macOS) or Ctrl + Shift + M (Windows/Linux)

Or:

Menu Bar: View → Agent Manager

You’ll see: Empty “Mission Control” interface with “+ New Agent” button.

6.2 Create a Simple Task

Click “+ New Agent” and enter this prompt:

Create a Python script that fetches current weather for San Francisco using the OpenWeatherMap API and displays it in a formatted table.

Why this task?
– Tests API integration
– Requires external package installation (requests)
– Involves terminal execution
– Simple enough to complete in 2-3 minutes

6.3 Choose Planning Mode

Prompt: “Use Planning Mode or Fast Mode?”

Select Planning Mode (recommended for your first agent).

Click “Start Agent”

6.4 Review Task List

The agent generates a task list. You’ll see something like:

Artifact: Task List
1. Research OpenWeatherMap API documentation
2. Create a Python virtual environment
3. Install requests library
4. Write an API wrapper function
5. Format output as an ASCII table
6. Add error handling for API failures
7. Write usage documentation

Review the plan:
– ✓ Looks good? Click “Approve”
– ✕ Need changes? Comment on specific tasks

6.5 Watch the Agent Work

Once approved, the agent starts executing:

Terminal Activity:

Creating virtual environment...
python3 -m venv .venv
source .venv/bin/activate

Installing dependencies...
pip install requests tabulate

Editor Activity:
– Creates weather.py
– Writes API integration code
– Adds error handling

Artifacts Generated:
– Code diff showing changes
– Terminal output logs
– Implementation notes

Total Time: ~2 minutes

6.6 Review Results

Agent completes and shows the final artifact:

Artifact: Implementation Summary
– ✓ Created weather.py
– ✓ Installed dependencies
– ✓ Tested with sample API call
– Screenshot of formatted output

Test the code yourself:

python weather.py
Code language: CSS (css)

Output:

San Francisco Weather
---------------------
Temperature: 62°F
Condition: Partly Cloudy
Humidity: 68%
Wind Speed: 12 mph

Congratulations! You just completed your first autonomous agent task.

---

Step 7: Building a Real Project

Now let’s tackle something more complex: a full-stack To-Do app.

7.1 Spawn Multi-Agent Workflow

Prompt:

Build a full-stack To-Do application:
- React frontend with TypeScript
- Node.js Express backend
- PostgreSQL database
- User authentication (JWT)
- CRUD operations for tasks
- Responsive design with Tailwind CSS
Code language: JavaScript (javascript)

Agent Response:

“This is a complex task. I’ll spawn 3 parallel agents:
– Agent 1: Backend API + Database
– Agent 2: React Frontend
– Agent 3: Authentication & Security

Continue with parallel execution?”

Click “Yes, Spawn Agents”

7.2 Monitor Parallel Agents

You’ll see three agents in Agent Manager:

Agent 1 (Backend):
– Creates Express server
– Sets up PostgreSQL schema
– Writes API routes (GET/POST/PUT/DELETE)

Agent 2 (Frontend):
– Scaffolds React app
– Creates components (TodoList, TodoItem, AddTodo)
– Styles with Tailwind

Agent 3 (Auth):
– Implements JWT generation/validation
– Creates login/signup endpoints
– Adds protected route middleware

Progress Bars: Each agent shows completion percentage.

7.3 Review & Provide Feedback

While agents work, review artifacts:

Example Feedback:

On Agent 2’s UI component artifact:

“The Add Task button should be green, not blue. Also, add a dark mode toggle.”

Agent 2 incorporates feedback without stopping and regenerates the component.

7.4 Agent Collaboration

Agents communicate through shared knowledge base:

Agent 3 (Auth):

“I’ve added authentication middleware. Updating Agent 1’s API routes to use it.”

Agent 1 (Backend):

“Received auth middleware. Protecting /api/todos routes.”

This autonomous coordination is Antigravity’s killer feature.

7.5 Final Results

Total Time: 12 minutes (vs 3-4 hours manually)

Deliverables:
– Working on a full-stack app
– PostgreSQL migrations
– Comprehensive README
– Environment variable template
– Docker Compose config (bonus!)

Test the app:

cd todo-app
docker-compose up

Open http://localhost:3000 → Working To-Do app with auth!

---

Step 8: Advanced Features

8.1 Workspace Rules

Define project-specific coding standards:

File: .antigravity/workspace-rules.md

# Coding Standards

## TypeScript
- Use functional components (no class components)
- Prefer interfaces over types
- All functions must have JSDoc comments

## Python
- Black formatting (line length 88)
- Type hints required
- Docstrings for all public functions

## Testing
- Minimum 80% code coverage
- AAA pattern (Arrange, Act, Assert)
Code language: PHP (php)

Agents automatically follow these rules.

8.2 Reusable Workflows

Save common tasks as workflows:

Example: “Deploy to Vercel” workflow

1. Agent Manager → Workflows → “+ New Workflow”
2. Name: “Deploy to Vercel”
3. Prompt:
   Deploy this project to Vercel:
4. Install Vercel CLI
5. Log in to Vercel (ask fora token)
6. Configure build settings
7. Deploy to production
8. Output deployment URL
9. Save

Usage: Click “Run Workflow” → Select “Deploy to Vercel” → Automated deployment in 2 minutes.

8.3 Knowledge Base Management

Agents learn from your codebase. Help them learn faster:

Add Important Context:

Agent Manager → Knowledge Base → “+ Add Context”

Project: E-commerce platform
Tech Stack: Next.js 14, Prisma, Stripe

Key Conventions:
- All API routes use route handlers (app/api/)
- Database queries use Prisma Client
- Payment processing via Stripe (don't mock in tests)
- Image optimization via next/image

Common Gotchas:
- Server components can't use useState
- Stripe webhooks require raw body (no body parsing middleware)
Code language: PHP (php)

Agents reference this context when planning tasks.

8.4 Multi-Model Switching

Switch AI models for specific tasks:

When to Use Each Model:

Model	Best For	Speed	Cost
Gemini 3 Pro	Complex refactoring, architecture	Medium	Free
Gemini 3 Flash	Simple tasks, boilerplate	Fast	Free
Gemini 3 Deep Think	Debugging, optimization	Slow	Free
Claude Sonnet 4.5	Creative solutions, docs	Medium	Paid
Claude Opus 4.5	Mission-critical code	Slow	Paid

Switch Model:

Agent Manager → Settings → Model Selection

---

Step 9: Common Issues & Solutions

Issue 1: “Rate Limit Exceeded”

Error:

Agent paused: Gemini API rate limit exceeded.
Retry after: 4 hours

Solutions:
1. Wait for rate limit reset (free tier refreshes weekly)
2. Switch to Gemini Flash (uses less quota)
3. Upgrade to Google AI Pro ($20/month for 5-hour refresh)

Check Quota:
Agent Manager → Settings → Usage Stats

Issue 2: “Permission Denied” in Terminal

Error:

Agent failed: Permission denied executing 'npm install'
Code language: JavaScript (javascript)

Solutions:
1. Check terminal policy: Must be “Auto” or “Turbo”
2. Add to allowlist:
bash
echo "npm install" >> ~/.gemini/antigravity/terminalAllowlist.txt
3. Grant terminal access: Settings → Security → Terminal Permissions

Issue 3: Agent Produces Buggy Code

Symptoms: Code doesn’t run, tests fail

Solutions:
1. Use Planning Mode (Fast Mode skips validation)
2. Add workspace rules with code standards
3. Provide more detailed prompts:
– ✕ Bad: “Build a login system”
– ✓ Good: “Build JWT-based login with bcrypt password hashing, input validation using Zod, and rate limiting (max five attempts/hour)”
4. Review artifacts BEFORE approving the task list

Issue 4: Browser Extension Not Working

Error:

Agent failed: Chrome extension not detected

Solutions:
1. Verify extension installed: chrome://extensions/
2. Enable extension: Toggle switch must be ON
3. Pin extension: Right-click → Pin
4. Reinstall: Delete and re-download from ~/.gemini/antigravity/chrome-extension/
5. Check URL allowlist: Ensure the target URL is whitelisted

Issue 5: Slow Agent Performance

Symptoms: Tasks taking 10+ minutes

Solutions:
1. Switch to Gemini Flash for simple tasks
2. Break down prompts: Instead of “Build entire app,” split into smaller agents
3. Check internet connection: Agents require a stable connection to the API
4. Close other IDE instances: Antigravity is resource-intensive
5. Increase RAM allocation: Settings → Performance → Memory Limit (4GB → 8GB)

---

Step 10: Best Practices

10.1 Writing Effective Prompts

Bad Prompt:

Make a website

Good Prompt:

Create a landing page for a SaaS product with:
- Hero section with CTA button
- Features grid (3 columns)
- Pricing table (3 tiers)
- FAQ accordion
- Contact form
Tech: Next.js 14, Tailwind CSS, shadcn/ui components
Mobile-responsive design required
Code language: JavaScript (javascript)

Best Prompt:

Create a conversion-optimized landing page for "TaskMaster" (project management SaaS):

SECTIONS:
1. Hero:
   - Headline: "Project Management for Modern Teams"
   - Subheadline: "Ship faster with AI-powered task automation"
   - CTA: "Start Free Trial" (green button, link to /signup)
   - Background: Gradient (blue to purple)

2. Features:
   - Grid: 3 columns on desktop, 1 on mobile
   - Icons: Lucide React icons
   - Features:
     a. "AI Task Prioritization" - icon: brain
     b. "Real-time Collaboration" - icon: users
     c. "Integrations" - icon: plug

3. Pricing:
   - 3 tiers: Free ($0), Pro ($29/mo), Enterprise (Custom)
   - Highlight Pro tier (most popular badge)
   - Annual discount: 20% off

4. FAQ:
   - Accordion component (shadcn/ui)
   - 5 questions (I'll provide content separately)

5. Footer:
   - Links: About, Privacy, Terms
   - Social icons: Twitter, LinkedIn, GitHub

TECH STACK:
- Next.js 14 (App Router)
- TypeScript
- Tailwind CSS
- shadcn/ui components
- Framer Motion (animations)

REQUIREMENTS:
- Mobile-first responsive design
- Accessibility (WCAG AA compliant)
- SEO meta tags
- Google Analytics tracking code
- Loading states for CTA button

OUTPUT:
- Component folder structure
- Reusable components (Button, Card, etc.)
- README with deployment instructions
Code language: PHP (php)

Specificity = Better Results

10.2 Security Checklist

Before using Antigravity in production:

• [ ] Terminal policy set to “Auto” (not Turbo)
• [ ] Terminal denylist includes: rm -rf, sudo, curl | bash
• [ ] Browser allowlist limits URLs to trusted domains
• [ ] Workspace rules enforce code review for sensitive files
• [ ] API keys stored in .env (never in prompts)
• [ ] Git ignore includes .antigravity/ artifacts
• [ ] Code review enabled for all agent PRs

10.3 Performance Optimisation

Reduce Rate Limit Usage:
1. Use Gemini Flash for boilerplate (5x cheaper)
2. Cache repetitive tasks in workflows
3. Review task lists before approval (prevents wasted work)

Speed Up Agents:
1. Use Fast Mode for simple tasks (<3 files)
2. Pre-install dependencies (agents won’t install each time)
3. Provide example code in prompts (agents copy patterns)

Reduce Costs (When Paid Tiers Launch):
1. Delete old artifacts (Settings → Storage → Clear Artifacts)
2. Use local models for sensitive code (future feature)
3. Batch similar tasks (1 agent handles multiple files)

10.4 Debugging Agent Mistakes

When agents generate buggy code:

Step 1: Review Artifacts
– Find the code diff artifact
– Identify what changed

Step 2: Check Agent Logs
– Agent Manager → [Agent Name] → Logs
– Look for API errors or misunderstood prompts

Step 3: Provide Corrective Feedback
– Comment on the buggy artifact
– Be specific: “Line 47: Use async/await, not .then()“

Step 4: Regenerate
– Agent incorporates feedback and reruns

Step 5: Add to Knowledge Base
– If the agent repeats mistakes, add a constraint to the workspace rules

---

Frequently Asked Questions

1. How do I update Google Antigravity?

Antigravity auto-updates by default. Check version:
– Menu Bar: Antigravity → About
– Manual Update: Settings → Updates → “Check for Updates”

2. Can I use Antigravity offline?

No. Agents require an internet connection to access the Gemini API. Code editing works offline, but no AI features.

3. How do I import my VS Code extensions?

During Setup: Choose “Import from VS Code”

After Setup:
1. Settings → Extensions
2. Click “Import from VS Code”
3. Select extensions to transfer

Not all extensions work. Check compatibility list: antigravity.google/extensions

4. Can I use Antigravity with GitHub Copilot?

Yes! Install the Copilot extension:
1. Extensions → Search “GitHub Copilot”
2. Install
3. Sign in with GitHub

Use Copilot for inline suggestions, Antigravity agents for full tasks.

5. How do I delete agent artifacts?

Delete Single Artifact:
– Agent Manager → [Agent Name] → Artifacts
– Right-click → “Delete”

Clear All Artifacts:
– Settings → Storage → “Clear Artifacts”
– Frees disk space

Note: Agents lose context from deleted artifacts.

6. What happens if I close Antigravity while an agent is running?

Agents pause and resume when you reopen Antigravity. Progress is saved to the cloud.

Exception: Long-running tasks (>1 hour) may timeout. Check Agent Manager → Status.

7. Can I self-host Antigravity for private code?

Not yet. Google hasn’t released a self-hosted version. Expected in Q2 2026 for enterprise customers.

---

Next Steps

You’ve mastered the basics of Google Antigravity! Here’s what to explore next:

Further Learning:
– Google Antigravity Documentation
– Google Codelabs: Advanced Antigravity
– Antigravity Community Discord

Practice Projects:
1. Build a REST API with automated tests
2. Create a Chrome extension
3. Migrate an existing project to TypeScript

Related Articles:
– Google Antigravity Review 2025
– Google Antigravity vs Cursor AI
– Best AI Code Editors 2025

---

Conclusion

Google Antigravity represents a fundamental shift in AI-assisted development. Instead of suggesting code, it completes tasks autonomously.

The setup requires careful security configuration (terminal/browser policies), but once configured, Antigravity becomes a force multiplier for your productivity.

Key Takeaways:
– ✓ Start with “Auto” mode for terminal/browser (not Turbo)
– ✓ Use Planning Mode for complex tasks
– ✓ Write detailed prompts for best results
– ✓ Review artifacts before approving
– ✓ Add workspace rules for consistent code quality

The future of coding is task-oriented, and you’re now equipped to leverage autonomous agents like a pro.

Happy coding!

---

Sources

• Google Codelabs: Getting Started with Antigravity
• Codecademy: How to Set Up Google Antigravity
• Google Developers Blog: Build with Antigravity
• Second Talent: Download and Install Guide

---

Affiliate Disclosure: This tutorial contains affiliate links. If you sign up for Google Antigravity through our links, we may earn a commission at no extra cost to you. Read our full Affiliate Disclosure.

Last Updated: December 22, 2025