Antigravity Track
Module 36
Antigravity Track — Module 36
First Agent in Five Minutes: ThreadCo's developer downloads Antigravity, signs in with her Google account, opens the shopmate/ folder, and types one sentence: "Add a gift wrapping option to the checkout flow." Four minutes later, an agent has written the feature, the tests, and updated the API docs. She reviews a 94-line diff and clicks Accept.

Setup & Your First Agent

Antigravity installs in minutes and requires no API key configuration for standard use. This module covers every step from download to your first autonomous agent run, including configuration options, project setup, and troubleshooting.

System Requirements

Before downloading, verify your system meets the minimum requirements:

RequirementMinimumRecommended
Operating SystemmacOS 13+, Windows 10, Ubuntu 20.04+macOS 14+, Windows 11, Ubuntu 22.04+
RAM8 GB16 GB (agents use memory for context caching)
Disk Space500 MB for installation2 GB (includes local index cache)
InternetStable broadband connectionLow-latency connection (agents stream responses)
Browser (for browser agent)Chrome 120+Chrome 124+ (latest stable)
Node.js (optional)Not requiredv18+ if your project uses it (agents run terminal commands)

Step-by-Step Installation

1

Download Antigravity

Go to antigravity.google and click Download for [your OS]. Available for macOS (Apple Silicon + Intel), Windows 10/11, and Ubuntu 20.04+. The installer is approximately 180MB.

On macOS, you will download a .dmg file. On Windows, a .exe installer. On Linux, a .deb package. All three follow standard installation procedures for their platform.

2

Run the Installer

macOS: Open the .dmg, drag Antigravity to Applications. On first launch, right-click and select Open to bypass Gatekeeper (required for unsigned preview software).

Windows: Run the .exe installer. Accept the defaults. Antigravity installs to C:\Users\[you]\AppData\Local\Programs\Antigravity. It adds itself to your PATH automatically.

Linux: Run sudo dpkg -i antigravity_*.deb && sudo apt-get install -f. Launch from your application menu or terminal with antigravity.

3

Sign in with Google

On first launch, click Sign in with Google. Your Google account manages access to Gemini 3 Pro (default model) and usage quotas. No separate API key is needed during the free preview period.

If you want to use Claude Sonnet 4.6 or GPT-4o, you will add those API keys later in Settings. The Google sign-in covers Gemini models only.

4

Import VS Code Settings (Optional)

Antigravity detects existing VS Code or Windsurf installations and offers to import extensions, themes, and keybindings. Accept this for a zero-friction transition — your familiar environment transfers over.

Imported settings include: color themes, icon themes, keyboard shortcuts, installed extensions, editor preferences (font size, tab width, word wrap), and workspace settings. Extensions that rely on VS Code-specific APIs may need updated versions from the Antigravity marketplace.

5

Open Your Project

File → Open Folder — select your project root. Antigravity indexes the codebase immediately. For large projects (over 50K lines), indexing takes 30-60 seconds and runs in the background.

The indexing process builds a semantic map: function signatures, type definitions, import chains, and file relationships. You will see a progress indicator in the bottom status bar. You can start working immediately — indexing completes in the background.

6

Select Your Model

Bottom status bar → click the model name → choose from: Gemini 3 Pro (default, fastest), Claude Sonnet 4.6 (best for complex reasoning), or GPT-4o. You can switch per-agent.

i
Model Selection Strategy

Gemini 3 Pro is the default and is free during preview. Use it for straightforward tasks: simple bug fixes, test writing, documentation updates. Claude Sonnet 4.6 excels at complex multi-step reasoning and tasks requiring nuanced understanding of business logic. GPT-4o is a strong generalist. During preview, experiment with all three on similar tasks to build intuition for which model suits which task type.

Configuring Third-Party Models

To use Claude or GPT-4o alongside Gemini, you need to add API keys:

1

Open Settings

Press Ctrl+, (Windows/Linux) or Cmd+, (macOS). Navigate to Antigravity → Models.

2

Add Anthropic API Key

Under "Claude Models," paste your Anthropic API key from console.anthropic.com. This enables Claude Sonnet 4.6 in the model picker. Usage is billed to your Anthropic account.

3

Add OpenAI API Key

Under "OpenAI Models," paste your OpenAI API key from platform.openai.com. This enables GPT-4o. Usage is billed to your OpenAI account.

4

Set Default Model Per Task Type (Optional)

In the same settings panel, you can configure default models by task type: one model for code generation, another for test writing, another for documentation. This is optional — most users start with a single default and switch manually per agent.

Project Configuration — .antigravity File

Antigravity supports an optional .antigravity configuration file in your project root. This file tells agents about your project's conventions, similar to how .windsurfrules works in Windsurf:

.antigravity — Project Configuration
# .antigravity — ThreadCo ShopMate project config
 
project_name: shopmate
language: python
framework: fastapi
test_command: pytest tests/ -v
lint_command: ruff check .
dev_server: uvicorn api.main:app --reload --port 3000
 
# Conventions agents should follow
conventions:
- Use type hints on all function signatures
- Write docstrings for public functions
- All endpoints return JSON with {status, data, error}
- Tests use pytest fixtures, not setUp/tearDown
 
# Files agents should never modify
protected_files:
- .env
- alembic/versions/*
- config/production.py

The .antigravity file is optional but strongly recommended. Agents read it before every task and follow its conventions. The test_command and lint_command fields tell agents how to verify their own work. The protected_files list prevents agents from modifying sensitive files.

Dispatching Your First Agent

With setup complete, dispatch your first agent:

1

Open the Agent Panel

Press Ctrl+Shift+A (Windows/Linux) or Cmd+Shift+A (macOS). The Agent panel slides open on the right side of the editor.

2

Write Your Task

Type your task in plain English. Be specific: name the file, the function, the expected behaviour, and any constraints. For your first task, try something small and verifiable — for example: "Add a GET /health endpoint to api/main.py that returns {status: 'ok'}. Add a test for it in tests/test_health.py."

3

Review the Auto-Generated Plan

Before executing, the agent shows you its plan: which files it will read, what it will create or modify, and in what order. Scan this plan to ensure it makes sense. You can edit the plan or add constraints before the agent starts.

4

Press Enter to Execute

The agent begins executing its plan. You can watch progress in real time in the Manager Surface, or switch to other work while the agent runs.

5

Review the Diff

When the agent finishes, it queues a diff for your review. The diff shows all changed files with standard green/red highlighting. Test results are shown alongside the diff. Accept, reject, or partially accept.

First Agent Run — ThreadCo

shopmate — Antigravity — Agent Panel
New Agent Task
Task description
Add a gift wrapping option to the checkout flow. Customer selects yes/no, adds a gift message (max 100 chars). Charge £2.50 extra. Store the selection in the order. Write tests.
Model
Claude Sonnet 4.6
Mode
Autonomous
Agent Plan (auto-generated)
Read checkout flow in api/checkout.py
Add GiftWrap model to schemas.py
Add gift_wrap field to order endpoint
Write tests in tests/test_giftwrap.py
Update API docs

Understanding Agent Modes

Antigravity supports three execution modes, selectable when you dispatch a task:

Autonomous Mode

The agent executes its full plan without pausing for approval at each step. It reads files, writes code, runs tests, and iterates until the task is complete. It only pauses if it encounters ambiguity it cannot resolve. This is the default mode and the one most developers use for well-defined tasks.

Step-by-Step Mode

The agent pauses after each action and waits for your approval before continuing. Use this when you are learning the tool, working with sensitive code, or want to understand exactly how the agent approaches a problem. It is slower but gives you full visibility and control.

Plan-Only Mode

The agent generates a plan but does not execute it. You review the plan, adjust it, and then choose whether to execute. Use this when you want to preview the agent's approach before it touches any files. Useful for complex tasks where the approach matters as much as the output.

Key Shortcuts

ActionWindows / LinuxmacOS
Open Agent panelCtrl+Shift+ACmd+Shift+A
Open Manager SurfaceCtrl+Shift+MCmd+Shift+M
Inline agent editCtrl+KCmd+K
Accept diffCtrl+EnterCmd+Enter
Reject diffCtrl+BackspaceCmd+Backspace
Pause running agentCtrl+PCmd+P
Switch modelCtrl+Shift+LCmd+Shift+L
Open settingsCtrl+,Cmd+,
Toggle terminalCtrl+`Cmd+`
Quick agent (inline)Ctrl+ICmd+I

Writing Effective Task Descriptions

Writing Good Agent Tasks: The more specific your task, the better the output. Include: what to build, where it lives, any constraints (e.g. max 100 chars, charge £2.50), and what tests to write. Treat it like a ticket, not a chat message.

Here is a template that consistently produces good results:

Task Description Template
# What to do
[Action verb] [specific thing] in [specific file/location]
 
# Current behaviour
[What happens now — or "this does not exist yet"]
 
# Expected behaviour
[What should happen after the agent finishes]
 
# Constraints
[Any rules: max length, price, format, API compatibility]
 
# Tests
[What tests to write and where to put them]

Troubleshooting Common Setup Issues

ProblemCauseSolution
"Cannot connect to agent service"Firewall blocking outbound connectionsAllow antigravity.google and api.antigravity.google through your firewall. Antigravity requires HTTPS on ports 443.
Indexing takes longer than 5 minutesVery large project or node_modules includedCreate a .antigravityignore file (same syntax as .gitignore) to exclude node_modules, dist, build directories, and binary files.
Agent produces empty diffTask too vague or agent confusedRewrite the task with specific file names, function names, and expected behaviour. Check the agent's timeline in Manager Surface for clues.
VS Code extensions not workingExtension API incompatibilityCheck the Antigravity marketplace for updated versions. Some VS Code extensions need Antigravity-specific builds. Core extensions (GitLens, ESLint, Prettier) are supported.
Claude/GPT-4o model not appearingMissing API keyGo to Settings → Antigravity → Models and add your Anthropic or OpenAI API key.
"Quota exceeded" errorToo many agent dispatches in the preview periodThe free preview has usage limits. Wait for the quota to reset (daily) or switch to a model with remaining quota.
Agent cannot run testsMissing test_command configurationAdd a test_command to your .antigravity config file, or ensure your project has a recognisable test setup (package.json scripts, pytest.ini, etc.).
Browser agent not connectingChrome extension not installed or outdatedInstall the Antigravity Chrome extension from the Chrome Web Store. Ensure Chrome is running before dispatching a browser agent.
!
Preview Software — Expect Rough Edges

Antigravity is in public preview. If you encounter a bug that is not in the table above, check the antigravity.google/issues page for known issues. The team ships updates weekly, so many issues are resolved quickly. Keep the application updated via Help → Check for Updates.

Installing the Chrome Extension (for Browser Agent)

The browser agent requires a separate Chrome extension. Install it now so it is ready when you reach the Browser Integration module:

1

Open Chrome Web Store

In Chrome, go to the Chrome Web Store and search for "Antigravity Browser Agent" or navigate to the link shown in Antigravity's Settings → Browser panel.

2

Install the Extension

Click "Add to Chrome." The extension requires permissions to read and interact with web pages (necessary for form filling, clicking, and DOM reading). Review the permissions and accept.

3

Verify Connection

Back in Antigravity, open Settings → Browser. You should see "Chrome extension connected" with a green indicator. If not, restart Chrome and Antigravity.

Hands-On Exercises

Exercise 1 — Install and Configure: Download Antigravity, sign in, and import your VS Code settings. Open a project you know well. Verify that indexing completes and the status bar shows your project name. Take note of how long indexing takes for your project size.
Exercise 2 — Create a .antigravity File: Create a .antigravity configuration file for your project. Include: project name, language, framework, test command, lint command, at least 3 coding conventions, and at least 2 protected files. Commit this file to your repository.
Exercise 3 — First Agent Task (Health Endpoint): Dispatch an agent with this exact task: "Add a GET /health endpoint to [your main file] that returns {status: 'ok', timestamp: [current UTC time]}. Add a test that verifies the endpoint returns 200 and the response contains both fields." Review the plan before execution. Accept or reject the diff.
Exercise 4 — Compare Agent Modes: Take the same simple task (e.g., "add a /version endpoint that returns the app version from package.json") and run it three times: once in Autonomous mode, once in Step-by-Step mode, and once in Plan-Only mode. Note the differences in speed, control, and your confidence in the output.
Exercise 5 — Deliberate Troubleshooting: Dispatch an intentionally vague task: "Fix the bugs." Observe what happens. Read the agent's timeline — how does it respond to ambiguity? Then rewrite the task with specific file names, function names, and expected behaviour. Compare the two results. This exercise builds intuition for what level of specificity agents need.