Quick answer: Google Antigravity is a cloud-based AI coding agent built on Gemini. It lives in the browser, plans tasks through small helper AI models, and can build, test and deploy projects end to end without you installing local tooling.
Google Antigravity is Google’s agentic development platform — tools that plan, write, run and verify code for you. This easy guide installs Antigravity, sets up your first project and runs your first AI agent, all in about 15 minutes.
What is Google Antigravity?
Google Antigravity is a suite of agent-first development tools from Google. Instead of writing every line yourself, you delegate: you describe the goal, Antigravity’s agent plans the work, makes the changes, runs the tests and brings back evidence you can review before accepting.
Antigravity launched in November 2025 and was restructured in May 2026 into four products that share the same agent engine:
| Product | What it is | Best for |
|---|---|---|
| Antigravity 2.0 | Standalone desktop app — your agents’ command centre | Running several agents in parallel, scheduled tasks, building without an IDE |
| Antigravity IDE | Full agentic code editor (built on VS Code) | Coding with an integrated Agent panel, artifacts and the Inspector |
Antigravity CLI (agy) | Fast terminal-first tool (successor to Gemini CLI) | Scripting, headless workflows and terminal users |
| Antigravity SDK | Libraries for developers | Embedding Antigravity inside your own tools |
What makes Antigravity different
- Agent Manager — spawn several agents at once, each on its own task, and watch them side by side.
- Inspector and Artifacts — every agent run produces a diff, command output, browser screenshots and a plain-English summary. You review the evidence before accepting anything.
- Gemini-native — tuned for the Gemini 3 family with large context windows; you can route some tasks to Claude or other models.
- Built-in browser agent — it can open a real browser, click through your UI and record proof that a feature works.
- MCP support — connect databases, GitHub, docs and other tools, and agents use them on their own.
- Skills and workflows — teach the agent reusable playbooks in markdown (
SKILL.mdfiles) and chain them into slash commands.
Step 1: Install Antigravity
Antigravity is a standalone desktop application — it is not a VS Code extension. Download the installer for your system from the official site antigravity.google (macOS, Windows and Linux are all supported).
On first launch, sign in with a Google account. That same account is used for your Gemini usage quota. The free tier is generous enough to evaluate — roughly 50 agent runs a day — and paid plans unlock longer contexts and parallel agents. Install the optional Antigravity IDE from the download page if you prefer a full editor.
Step 2: Create your first project
Antigravity is project-centric. A project gathers the folders an agent is allowed to work on — you could include a frontend and a backend repo together so one agent sees the whole picture.
- Open Antigravity and click New Project.
- Give it a name and add one or more local folders (or create an empty project to build from scratch).
- Wait for the background codebase index to finish — agent quality drops sharply if you start before indexing completes.
Step 3: Run your first agent
Open the Agent window (Cmd/Ctrl + Shift + A) and give the agent a small, verifiable task — something you can confirm afterwards, like:
Add a /health endpoint to this Flask app that returns 200 OK and the JSON body {"status": "ok"}.Here is the loop you will see on every run:
- Implementation plan — the agent drafts a plan listing the files it will touch, the commands it will run and how it will verify the result. Review and edit it, then click Proceed.
- Execution — the agent edits files and runs commands, pausing to ask permission when it needs to install packages or start services.
- Verification — it tests the change (here, it would hit the endpoint and check the JSON).
- Review in the Inspector — before you Accept, open the Inspector: every edited file’s diff, the command output and, if your app has a UI, browser screenshots. Accept only when the evidence matches the goal.
That “review the evidence” step is the real skill. Anyone can fire prompts; good Antigravity users read the Inspector, push back when verification is thin, and re-run with sharper instructions.
Slash commands that speed things up
| Command | What it does |
|---|---|
/browser | Explicitly launch the browser agent to inspect or operate a website |
/schedule | Run a task on a schedule, e.g. every Monday 9 AM |
/mcp | List and manage connected MCP servers |
/skills | Browse and load instruction skills for the agent |
Connect tools with MCP
MCP (Model Context Protocol) lets agents use your real tools. Open Settings → Customizations → Add MCP+ to install one-click servers for Google Cloud services, or add your own servers to the shared config file $HOME/.gemini/config/mcp_config.json. Once connected, just ask: “Build and deploy a Cloud Run service” and the agent will reach for the Cloud Run server by itself.
3 settings to change on day one
- Review-driven mode — keep it on while you learn; every command waits for your approval.
- Model routing — set Gemini 3 Pro as the default and a cheaper model for trivial edits to save quota.
- Workspace rules — add a
.antigravity/rules.mdfile in your repo codifying test commands and code style.
Google Antigravity — quick FAQ
Is Google Antigravity free?
Yes — Antigravity has a free tier with a daily quota of agent runs (roughly 50 per day at launch). Paid plans add longer contexts, more parallel agents and faster models.
Do I need to know how to code to use Antigravity?
No. Antigravity 2.0 is designed so you can describe an idea and the agent builds it. That said, a little programming knowledge helps you review and guide what the agent produces.
Is Antigravity the same as Google Gemini?
No. Gemini is Google’s AI model family and chatbot; Antigravity is the agentic development platform that runs on Gemini models (and can route to others).
What is the difference between Antigravity and Claude Code?
Antigravity concentrates on a visual agent-manager experience with parallel agents and reviewable artifacts. Claude Code is a terminal-first agent. See our AI Coding Assistants guide for a full comparison.
Build out your full AI toolkit: read about Claude Code, the open-source OpenCode, or the ad-funded Freebuff. Then compare them all in the Complete Guide to AI Coding Assistants.