Getting started with Antigravity

Getting started with Antigravity takes about fifteen minutes if you already have a Google account and a machine that meets the system requirements. The browser installs like any other desktop application. What makes the first run feel unfamiliar is the plan-approve-execute loop — it replaces the "open tab, do thing" reflex with a "describe task, review plan, watch agent, check artifacts" cycle. This page walks through each step in sequence so nothing in that loop catches you by surprise.

Before you begin

Check three things before downloading: your operating system version, available RAM, and your Google account type. Antigravity runs on Windows 11 (64-bit x86 or ARM), macOS Ventura or newer on Intel or Apple Silicon, and mainstream Linux distributions on x86-64. The minimum RAM is 8 GB; 16 GB is comfortable for multi-window runs. Your Google account needs to have Gemini access enabled — personal accounts have it on by default, Workspace accounts depend on your admin's settings.

If you are on a Workspace account and Gemini is not available, ask your Google Workspace admin to enable the Gemini for Google Workspace add-on before continuing. The browser will install without it but the agent rail will show an "access not granted" error on first run.

Step-by-step walkthrough

Step 1 — Download the installer

Go to the Antigravity install guide and download the installer for your platform. The file is a standard executable on Windows (.exe), a disk image on macOS (.dmg), and a package or AppImage on Linux. The download is around 180 MB. Verify the checksum listed on the install page if your security policy requires it.

Step 2 — Run the installer

Double-click the installer. On Windows, the UAC prompt appears — click Yes to grant install permissions. On macOS, drag the application to your Applications folder. On Linux, run the package manager command shown on the install page or mark the AppImage as executable and double-click. Accept the default install path unless your organisation requires a custom location.

Step 3 — First launch

Open Antigravity from your applications list or start menu. On first launch, the browser initialises the local workspace directory and the agent runtime. This takes about thirty seconds and shows a progress bar. The workspace directory defaults to ~/antigravity-workspace — you can change it in Settings if needed. Do not close the browser during initialisation.

Step 4 — Sign in with your Google account

Click the Sign in button in the top-right corner of the browser chrome. A standard Google OAuth flow opens in a small window. Authenticate with the Google account you intend to use, then grant the scopes the product requests: Gemini model access, basic Google Drive read (for workspace file sync), and identity. Read the login reference for a full explanation of each scope before granting if your security policy requires it.

Step 5 — Open the agent rail

After sign-in, the main browser window loads. You will see a narrow rail on the left side of the window — that is the agent panel. Click the agent icon (the upward-arrow circle) or press Ctrl+Shift+A on Windows and Linux or Cmd+Shift+A on macOS to expand it to full width. The rail shows your run history (empty on first launch) and the task input field at the bottom.

Step 6 — Describe your first task

Type a concrete, low-stakes task in the input field. A good first task is something like: "Open the MDN page for the Fetch API, find the basic usage example, and save it to a file called fetch-notes.md in my workspace." Press Enter or click Run. The agent takes ten to twenty seconds to produce a numbered plan — a list of steps with URLs, file names, and actions. Read it before approving.

Step 7 — Approve the plan and watch the run

If the plan looks correct, click Approve to start execution. The agent rail shows each step as it completes: the URL it navigates to, the DOM element it reads, the file it writes. You can click any in-progress step to see the raw reasoning. If you want to pause, click Pause — the agent finishes its current atomic action and holds. Click Stop to terminate early; the partial artifact is saved automatically.

Step 8 — Review the artifact bundle

When the run finishes, the rail shows "Run complete" and a View Artifact button. Click it. The artifact timeline opens as a split pane: the left column lists every step with a timestamp; the right column shows the content for the selected step — a screenshot, a file diff, an HTTP response, or a console log. Scrub through the steps, annotate any step with a note, and use the Share button to export the bundle as a portable archive if you want to show it to a colleague.

First-run reference table

Step	What to click or type	Expected result
1 — Download	Install guide → platform button	~180 MB installer file saved
2 — Install	Run installer, accept UAC/drag to Applications	Browser appears in applications list
3 — First launch	Open from applications list	30-second init, workspace created
4 — Sign in	Sign in button → Google OAuth → grant scopes	Account name shown in top-right
5 — Agent rail	Agent icon or Ctrl+Shift+A	Agent panel expands, input field visible
6 — Describe task	Type task, press Enter	Numbered plan appears in 10–20 s
7 — Approve & run	Click Approve	Steps execute; rail shows live progress
8 — Review artifact	Click View Artifact	Timeline pane opens with all step captures

Oberon H. Pritchard-Vogelsang, API Architect at Cinderpike Integrations in Turku, described his first run this way: "I gave it a task I would have spent twenty minutes on manually — pull the error codes from our internal status page and list them in a markdown table. The plan looked right, I approved it, and three minutes later the file was in my workspace. The artifact showed me every step. I immediately ran a second task."

For a policy reference on agentic AI tools in developer environments, see the NIST AI resource centre, which covers trust, transparency, and human-oversight requirements for deployed agent systems.