Every few months, a tool emerges that quietly rewrites what’s possible for builders, entrepreneurs, and content creators working with AI. Google Antigravity is that tool — and most people haven’t heard of it yet. Available for free at antigravity.google, it lets you create websites, web applications, and complex digital products with nothing but a prompt and a folder on your computer.
What makes Antigravity different from the wave of AI builders that came before it isn’t just the quality of its output — it’s the level of autonomy it operates with. It doesn’t just generate code. It thinks through your project, generates visuals, tests its own work inside a real browser, and iterates until the result meets your specification. This guide walks through everything you need to know to get started and get results.
Whether you want to launch a landing page, build a working e-commerce prototype, or connect Antigravity to external tools that supercharge its capabilities, here’s how to do it right.
Two Modes, Two Use Cases
Antigravity is organized around two distinct interfaces that serve different purposes, and understanding the difference is the first thing to get right.
The Agent Manager (Playground)
The Agent Manager looks and feels like a familiar chat interface — similar to ChatGPT or Google Gemini. This is where you brainstorm, clarify ideas, draft project briefs, and test prompts. It’s not where you’ll build your final product, but it’s a valuable space for sketching out what you want before committing to a full build.
Inside the playground, you choose between two processing modes: Fast and Planning. Fast mode moves quickly with a lighter reasoning chain — ideal for websites, landing pages, and simpler projects. Planning mode engages a much deeper chain of thought, mapping out a full roadmap before writing a single line of code. Use Planning for complex applications, multi-step user flows, or anything where architectural decisions matter. The trade-off is time, but the output quality is significantly higher.
The Editor
The Editor is where the real work happens. Unlike the playground, the Editor requires you to open a local folder on your computer — this becomes the project root where all generated files are saved. Think of it as a local development environment powered by AI. Every file Antigravity creates, every image it generates, and every configuration it writes lands in that folder.
To get started, simply create an empty folder on your machine (name it anything), then open it in Antigravity via the Open Folder option. From there, you’re ready to build.
Use the Playground to refine your prompt and clarify your vision. Use the Editor to actually build. Treat them as two separate phases of the same workflow — not interchangeable alternatives.
Choosing Your AI Model
Antigravity gives you access to multiple AI models, and choosing the right one meaningfully affects your results. On the Google side, Gemini 2.0 Flash is the go-to choice for most website and application builds — it’s fast, capable, and more than sufficient for typical projects. The heavier Gemini models (2.1 Pro High, 2.1 Pro Low) are reserved for particularly complex architectures.
Anthropic’s models are also available directly inside Antigravity. Claude Sonnet and Claude Opus are strong alternatives for code-heavy projects, and they can be used interchangeably with the Gemini models depending on the task. The model lineup will evolve over time, so it’s worth checking what’s available in the interface when you start a new project.
Essential Settings to Configure First
Before you build anything serious, two settings are worth adjusting right away.
Review Mode
In Editor Settings → Antigravity Settings, you’ll find an option for how Antigravity handles autonomous actions. The default “Always Proceed” mode lets it move without asking for approval — which is efficient but carries risk. Antigravity can take control of your browser, open folders, and interact with applications on your machine. For new users, switching to “Ask for Review” before each action is strongly recommended. It slows things down slightly, but it keeps you in control while you’re learning how the tool operates.
Browser Tools
Under Settings → Browser, enable the browser tools option. This gives Antigravity the ability to open a real browser, navigate to your application, and test it autonomously. Without this enabled, the self-testing loop — one of Antigravity’s most impressive features — won’t work. With it enabled, Antigravity will click through forms, simulate user paths, identify broken flows, and iterate before handing you the result.
Custom Instructions and Workflows
The Customizations panel is where power users separate themselves from casual ones. It has two components worth configuring.
The Rules section is the equivalent of custom instructions in other AI tools. Anything you write here will apply to every interaction in Antigravity — your preferred stack, your design standards, your tone of voice for generated copy, how you want files structured. The more specific you are, the less you’ll need to repeat yourself across sessions.
The Workflows section lets you create custom commands — shortcuts that trigger specific, reusable actions. For example, an /audit command could instruct Antigravity to analyze any project for UX issues, performance problems, or conversion weaknesses and return a structured report. These commands make repetitive tasks instant and eliminate the need to re-prompt the same logic every time.
Custom workflows are the closest thing to building your own AI agent inside Antigravity. An audit workflow, for instance, can be pointed at any URL — your own site, a competitor’s landing page — and return actionable recommendations in seconds.
How a Full Build Actually Works?
To illustrate the full process, consider building a web application from scratch — a food ordering platform, for example, with a product configurator, nutritional information, and a complete checkout flow.
In Planning mode with a detailed prompt, Antigravity begins by generating a structured task list — a full development roadmap broken into discrete steps. It then works through that list autonomously: creating the file structure, writing the frontend, building the logic, generating placeholder images using integrated AI image tools (like Imagen), and even writing a logo.
Once the build phase is complete, Antigravity switches into testing mode. Without any human input, it opens the application in a browser, navigates through the user flow step by step, submits test forms, checks the checkout process, and validates that the experience works as intended. If it finds a problem, it fixes it and retests. The final deliverable is a working prototype — not a scaffold or a code dump, but something you can actually open and use.
The self-testing loop is what fundamentally separates Antigravity from most AI coding tools. It doesn’t just generate — it verifies. The browser-based QA phase catches broken flows before you ever touch the result.
Supercharging Antigravity with MCP Servers
MCP (Model Context Protocol) servers are connectors that plug Antigravity into external tools and services. Think of them as USB cables between Antigravity and the rest of your stack. The built-in MCP catalog includes integrations with GitHub, Figma, Airtable, n8n, and more.
The most immediately practical MCP for most builders is the GitHub integration. Once connected, Antigravity can push your finished project directly to a GitHub repository. From there, deploying to a live URL via a platform like Vercel becomes a straightforward next step — Antigravity can guide that process too.
Another MCP worth knowing is Stitch — Google’s AI-powered design tool. Stitch lets you create UI mockups, redesign existing pages, and generate design templates using Imagen. The workflow that unlocks real leverage is connecting the two: use Stitch to create a polished visual design, then pass it to Antigravity in the Editor to develop into a fully functional application. Design-to-code, end to end, with minimal manual effort.
The Stitch → Antigravity pipeline is one of the most powerful combinations currently available in the no-code/AI-builder space. You get design-quality visuals and working code without switching between a designer and a developer.
Adding Claude Code as an Extension
Antigravity supports browser-style extensions that add capabilities directly to the editor. One of the most significant is the Claude Code extension, which you can install by searching “Claude” in the extensions panel (not “Claude Code” — search just “Claude”).
Once installed and connected with an Anthropic account, Claude Code becomes available as a sidebar inside Antigravity. This means you can use Claude Code’s capabilities — deep code analysis, architectural review, complex refactoring — without the usual friction of installing it locally. For anyone who has found Claude Code difficult to set up as a standalone tool, this integration removes that barrier entirely.
Free Tier and Quota Considerations
Antigravity is free to use with a standard Google account, subject to message quotas that refresh periodically. For users with Google One AI Premium or Google Workspace accounts, the token allowance scales up significantly — typically enough to build multiple projects per month without running into limits. Even on the base free tier, a first website or application is well within reach.
| Account Type | Quota Level | Best For |
|---|---|---|
| Free Google Account | Limited (refreshes periodically) | First project, exploration |
| Google One AI Premium | High | Regular builds, multiple projects |
| Google Workspace | High | Teams, professional workloads |
Prompt Engineering: The Skill That Multiplies Everything
Antigravity’s output quality scales directly with prompt quality. A vague prompt produces a vague result. A well-structured prompt — one that specifies the project type, target audience, key features, design expectations, and any technical constraints — produces something significantly closer to a finished product on the first pass.
For the Planning mode in particular, investing time in a detailed prompt pays off disproportionately. Antigravity uses that prompt to construct its entire development roadmap. A prompt that spells out user flows, edge cases, and validation requirements will generate a roadmap — and ultimately a product — that accounts for those details from the start, rather than requiring multiple rounds of correction.
Treat prompt writing as the highest-leverage activity in your Antigravity workflow. The more you invest in your initial specification, the less time you spend in correction loops — and the more autonomous the build becomes.
Frequently Asked Questions
Is Google Antigravity actually free?
Yes — Antigravity is free to use with any Google account, with usage quotas that reset periodically. Google One AI Premium and Workspace accounts receive higher quotas, but even the base tier is sufficient for building your first site or application without paying anything.
Where do I download Google Antigravity?
Antigravity is available at antigravity.google. The site will detect your operating system and provide the appropriate installer for Mac or Windows. Sign in with a Gmail account to get started.
What’s the difference between Planning and Fast mode?
Fast mode uses a lighter reasoning chain and produces results more quickly — suitable for websites and simpler applications. Planning mode engages a deeper chain of thought, generates a full task roadmap before building, and produces significantly more sophisticated results for complex applications. The trade-off is time.
Can Antigravity deploy my application to the web?
Not directly, but it can push your project to GitHub via the GitHub MCP integration. From GitHub, deployment to a live URL through a platform like Vercel is a straightforward next step — and Antigravity can guide you through the process.
Do I need coding knowledge to use Antigravity?
No. Antigravity handles code generation, file management, and testing autonomously. Some technical literacy helps when reviewing what it produces or when configuring integrations, but you don’t need to write any code yourself to build functional applications with this tool.
What is the Stitch MCP and should I use it?
Stitch is Google’s AI-powered design tool, available as an MCP integration inside Antigravity. It uses Imagen to generate UI mockups, redesign existing pages, or create application templates. Connecting Stitch to Antigravity creates a design-to-code pipeline — one of the most powerful workflows currently available in the AI builder ecosystem.
