If you have been wondering how to use Google Antigravity IDE, you are not alone. Since Google quietly dropped it in public preview in November 2025, developers have been trying to figure out whether it is a genuinely new way to code or just another AI sidebar bolted onto VS Code.

It is neither. It is something more interesting than both.

Google Antigravity is an agent-first development platform that evolves the traditional IDE into a Mission Control for managing autonomous agents that can plan, code, and even browse the web to help you build. You are not writing code with an AI assistant anymore. You are directing a team of AI agents the way a manager directs a team of engineers.

This guide walks you through the full setup, every configuration choice that matters, and how to run your first real agent-driven task. No fluff, no filler.

What you need before you start

Before downloading anything, make sure you have:

That is it. No API key purchases, no paid tier. Google Antigravity is available in public preview at no cost for individuals, with generous rate limits on Gemini 3 Pro included.

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more

Step 1: Download and install Google Antigravity

Head over to antigravity.google and download the installer for your OS (Windows, macOS, or Linux).

Run the installer the same way you would any standard desktop app. On Windows, run it as Administrator and temporarily disable antivirus software if the install stalls. On Mac, drag it to Applications as usual.

Once installed, open Antigravity. The first launch will trigger an onboarding wizard.

Step 2: First launch setup: the choices that actually matter

When you open the Antigravity IDE for the first time, it will ask you some onboarding questions. After this, select the Start fresh option and click Next. Then select the theme you want your IDE to have.

If you are migrating from VS Code or Cursor, you can import your existing settings. But if you want to understand Antigravity on its own terms, choose Start Fresh. It is the cleaner way to learn.

Next comes the configuration screen that most people get wrong.

Agent usage mode choose carefully

You will see the Agent Manager configuration screen with different development modes. Agent-driven development is full autopilot — you instruct the AI what to build and it writes code, creates files, and runs commands automatically. Review-driven development means the AI asks permission before performing almost any action. Agent-assisted development is the recommended option — you stay in control, but the AI helps with safe automations.

For your first session, go with Agent-Assisted Development. It keeps you in the loop while still letting you experience the real power of autonomous agents.

Terminal policy

Set this to Auto. This allows the agent to run standard commands like installing packages and starting a dev server without pinging you every 30 seconds. You can tighten this later once you understand what the agent is doing.

Step 3: Understand the two views (this is where Antigravity clicks)

Antigravity features two primary views: the Editor view, which is a standard IDE interface similar to VS Code with an agent sidebar, and the Manager view, which is a control center for orchestrating multiple agents working in parallel across workspaces, allowing asynchronous task execution.

Think of it this way: the Editor view is for when you want to be hands-on. The Manager view is for when you want to step back and let agents run entire tasks while you watch.

In the Manager view, the developer acts as an architect. You define high-level objectives, each of which spawns a dedicated agent instance. The UI provides a visualization of parallel work streams, displaying the status of each agent, the artifacts they have produced, and any pending requests for human approval.

The moment you switch to Manager view for the first time is usually when Antigravity stops feeling like a fancy autocomplete tool and starts feeling like something genuinely different.

Step 4: Run your first agent task

Here is a real example to get a feel for it. In the Agent Manager panel, type something like:

"Build a simple task tracker web app with a to-do list, a completed section, and a clean dark mode UI. Use plain HTML, CSS, and JavaScript. No frameworks."

Then watch what happens.

The agent turns your natural language prompt into a task-level plan, runs terminal commands to create the project, edits code across your workspace, and can use the browser extension to open your app, click through flows, and capture screenshots as Artifacts.

To build user trust, agents generate Artifacts verifiable deliverables such as task lists, implementation plans, screenshots, and browser recordings rather than raw tool calls. This is how you audit what the agent did without reading through hundreds of lines of output.

If something looks off in the plan before it executes, click directly on the Artifact and leave a comment. The agent will incorporate your feedback without stopping its run.

Step 5: Understanding Agent modes: Plan vs Fast

Once you are inside a task, you will notice two execution modes in the sidebar.

Plan mode generates a detailed plan (a Plan Artifact) before acting, which is ideal for complex tasks. Fast mode executes instructions instantly, which is ideal for quick fixes.

Use Plan mode for anything involving multiple files or a new feature. Use Fast mode when you are asking for a small change like renaming a function or fixing a CSS bug. Getting this distinction right will save you a lot of time waiting on unnecessary plans for trivial tasks.

Step 6: Install the Chrome browser extension

The browser extension is optional but highly recommended for web development tasks.

With it installed, the agent can open your running app in the built-in browser, click through your UI, take screenshots, and verify that what it built actually works visually. Without it, the agent is essentially coding blind. For any web project, install the extension before you start.

Step 7: Prompting agents the right way

Most people who try Antigravity and feel underwhelmed are prompting it the same way they would prompt a chatbot. That is the wrong mental model.

Write focused, goal-based prompts. Create separate tasks for different features. Review agent plans before approving major changes. Avoid mixing unrelated tasks in a single prompt.

A bad prompt: "Build me an app."

A good prompt: "Create a Python Flask API with three endpoints: GET /users, POST /users, and DELETE /users/{id}. Use SQLite for the database. Include basic error handling and return JSON responses."

The more specific your goal, the less back-and-forth, and the cleaner the output.

Step 8: What models is Antigravity running on?

Antigravity is built around Google's Gemini 3 models, but also supports other providers like Claude Sonnet 4.6 and GPT-OSS. In practice, Gemini 3.1 Pro handles the heavy planning and reasoning, while Gemini 3 Flash is used for faster, lighter tasks. You can switch models per task inside the Agent Manager settings if you prefer Claude for certain workflows.

What Antigravity is not good at (yet)

It is still public preview, and a few rough edges are real:

Antigravity is powerful, but not autonomous enough to replace engineering judgment. The agents can produce confident-looking code that has subtle logic errors, especially in complex business logic. Always review the Plan Artifact before approving a large task, and read the diff before merging agent-generated code into anything production-facing.

It is also not ideal for quick one-line edits. The agent overhead makes it slower than just typing the fix yourself. Use the Editor view with inline commands for small changes, and save the Agent Manager for tasks that would otherwise take you 30 minutes or more.

Antigravity moves the developer's role from someone filling out code snippets to an architect or manager orchestrating a workforce of digital agents. That is not hype, it is a real shift in how you spend your time when coding.

The setup takes under 10 minutes. The learning curve is in learning to write good agent prompts, not in learning the tool itself. Start with a small, self-contained project, watch the Agent Manager run, and you will understand within one session why this is different from everything that came before it.

Download at antigravity.google/download.