What Is Google Antigravity? The Platform Nobody Has Fully Explained Yet

Google Antigravity is Google's agent-first development platform that lets developers go from an idea to a production-ready app using autonomous AI agents. Version 2.0, launched at Google I/O 2026, expanded it from a single IDE into a five-surface platform: a standalone desktop app, a CLI called agy, an SDK, Managed Agents in the Gemini API, and the Gemini Enterprise Agent Platform. It is powered by Gemini 3.5 Flash and replaces Gemini CLI, which shuts down for most users on June 18, 2026.

What Is Google Antigravity, Exactly?

Google Antigravity is an agent-first development platform built by Google. The official description from Google's developer blog is precise: it is "the agent-first development platform for taking an idea and turning it into a production-ready app."

That sentence contains the key distinction. Most AI coding tools are assistant-first: you write code and the AI helps. Antigravity is agent-first: you describe a task, and multiple AI agents plan, build, test, debug, and deploy it autonomously, working in parallel, without requiring your continuous input.

Think of it less like GitHub Copilot or ChatGPT for code, and more like a team of junior developers running in the background while you focus on architecture and product decisions.

"Google Antigravity is our agent-first development platform for developers to take an idea and turn it into a production-ready app. Today, we are expanding the Antigravity ecosystem to manage and deploy agents that can integrate across key developer surfaces." Google Developer Blog, May 19, 2026

The version 2.0 release, announced at Google I/O 2026 on May 19, 2026, is not an incremental update. It is a fundamental redesign. The original Antigravity (November 2025) was a single VS Code fork IDE with AI coding assistance. Antigravity 2.0 is a five-surface platform spanning a desktop app, CLI, SDK, managed API, and enterprise infrastructure. The code editor is no longer the center of the product. The agent is.

• 5 Platform Surfaces

• 289 Tokens/sec (Flash)

• 5× Ultra vs Pro Limits

• Jun 18 Gemini CLI Shutdown

From Gemini CLI to Antigravity: The Origin Story

Understanding why Google created Antigravity requires understanding what came before it. The precursor was Gemini CLI, Google's terminal-based AI coding tool that was built in Node.js and worked as a chat-first interface for agentic tasks in the terminal. Developers could prompt it to inspect code, edit files, and run commands.

Gemini CLI was genuinely useful and found an audience quickly. It proved that the terminal could be an interface for agentic AI workflows. But as developer workflows shifted toward multi-agent systems capable of splitting work and sharing a unified backend, the single-agent, Node.js-based Gemini CLI began to show its limits.

Google's own summary of the transition is candid: "Gemini CLI proved the terminal could be an incredible interface for agentic tasks, but your needs shifted." The answer to those shifted needs was Antigravity, which launched in November 2025 as an AI-first IDE built on a VS Code fork with early multi-agent capabilities powered by Gemini.

By Google I/O 2026, the Antigravity platform had matured to the point where Google was ready to make a definitive move: consolidate everything under a single brand, deprecate Gemini CLI, and ship a complete agent orchestration platform that rivals Cursor, Claude Code, GitHub Copilot, and every other AI development tool in the market simultaneously.

Critical Deadline: June 18, 2026 Gemini CLI stops serving requests for Google AI Pro, AI Ultra, and free Gemini Code Assist individual users on June 18, 2026. There is no grace period after that date. If you are on any of these plans, you must migrate to Antigravity CLI before the deadline. Enterprise users on Gemini Code Assist Standard or Enterprise licenses are not affected.

The Five Surfaces of Antigravity 2.0

The single most important thing to understand about Antigravity 2.0 is that it is not one product. It is five distinct interfaces sharing a common agent execution layer. Google calls these "surfaces." Each surface serves a different workflow and context.

Antigravity Desktop App (Standalone)

A fully rebuilt standalone desktop application designed entirely around agent orchestration. It is separate from the original Antigravity IDE (VS Code fork), which remains available. The new desktop app runs multiple agents in parallel, supports dynamic subagents for parallelized workflows, schedules background tasks, integrates natively with Google AI Studio, Firebase, and Android development, and includes native voice command support so you can speak instructions instead of typing them.

Antigravity CLI (invoked as "agy")

A completely rewritten command-line interface built in Go (replacing the Node.js-based Gemini CLI). It is faster, has no runtime dependencies, and is designed for multi-step tool orchestration rather than chat-first interaction. The CLI supports multi-model routing, meaning you can direct tasks to Gemini, Claude, or GPT-OSS models from within the same terminal interface. Configuration moves from GEMINI.md to AGENTS.md, and skills move from .gemini/skills to .agents/skills. The binary name changes from "gemini" to "agy."

Antigravity SDK

Programmatic access to the same agent harness Google uses internally. Installed via pip install google-antigravity, the SDK exposes the same primitives the desktop app and CLI use: code execution, file management, web browsing, and reasoning, all running in Google's secure Linux sandbox. Developers can build custom agents and host them on their own infrastructure or on Google Cloud. Custom agent templates are available in Google AI Studio for enterprise use cases.

Managed Agents in the Gemini API

With a single API call, developers can spin up an isolated Linux agent that reasons, uses tools, executes code, manages files, and browses the web with persistent state across sessions. No orchestration code required. This makes agentic workflows accessible to applications that do not want to manage agent infrastructure themselves, calling an agent the same way you would call a chat completion endpoint. The Managed Agents feature uses Gemini 3.5 Flash in an isolated execution environment.

5. Gemini Enterprise Agent Platform

An enterprise-grade deployment path for organizations that need audit trails, access controls, compliance frameworks, and team-level agent management. This surface is built on Google Cloud and provides the governance controls that enterprise security teams require before allowing autonomous agents to operate in production environments. Organizations on Gemini Code Assist Standard or Enterprise licenses can use this path without disruption to their existing workflows.

How Google Antigravity Works: The Agent Architecture

The core innovation in Antigravity 2.0 is its multi-agent architecture. Understanding it explains why Google describes this as a fundamentally different product category from single-agent tools like Cursor's Composer or Claude Code.

The Manager Agent Model

When you describe a task in Antigravity (whether through the desktop app, CLI, or SDK), a Manager Agent receives your input, analyzes it, and breaks it into discrete subtasks. Each subtask is then assigned to a specialized Subagent that executes in parallel. The Manager Agent monitors progress, handles dependencies between tasks, and synthesizes the results into a cohesive output.

This is fundamentally different from how Cursor or Claude Code operate. Those tools process tasks sequentially: the AI assistant responds to one prompt, waits, then responds to the next. In Antigravity, complex tasks complete faster because subagents do not wait on each other. A refactoring task that might require five sequential back-and-forth exchanges in Cursor can be dispatched as five parallel subagents in Antigravity.

Dynamic Subagents

A key capability in the 2.0 release is dynamic subagents: the Manager Agent can spawn new subagents at runtime as it discovers additional subtasks during execution. You do not need to predict all the steps upfront. The system adapts as it works.

Scheduled Tasks

Rather than manually prompting an agent each time, developers can define tasks that invoke agents automatically in the background. This converts the agent from a single-turn tool into something closer to a persistent automation pipeline. You can schedule code review runs, test suite executions, or dependency update checks to happen automatically at defined intervals or in response to triggers like a git commit.

Built-in Browser Agent

One capability that distinguishes Antigravity from Claude Code and Cursor is the built-in browser agent. The desktop app includes an integrated Chromium browser. The browser agent can navigate pages, click buttons, toggle developer tools, switch to mobile viewport, and verify that UI changes look correct before reporting back to you. This removes the need to manually test visual changes after each agent iteration.

Developer Note The built-in browser agent in Antigravity gives it a meaningful workflow advantage for frontend development tasks. Visual regression testing, UI verification, and responsive design checks happen inside the same tool, removing a manual step that exists in all competing platforms at launch.

What AI Models Does Google Antigravity Use?

Antigravity 2.0 supports multiple models through a multi-model routing system. The platform is optimized for Gemini models, which means lower cost and lower latency on Google's infrastructure, but it is not locked to Google's models.

Gemini 3.5 Flash (Default)

The default model, co-optimized for the Antigravity harness at approximately 289 output tokens per second. Google reports this is 12x the speed of other frontier models within Antigravity's execution environment.

Gemini 3 Pro

Available for tasks requiring deeper reasoning and more careful analysis. Better suited to complex architecture decisions, nuanced code review, and long-context document analysis.

Claude Sonnet 4.5

Anthropic's model is available through multi-model routing within the same CLI and desktop app. Developers who trust Claude for specific coding tasks can route those tasks to it without switching tools.

OpenAI GPT-OSS

Open-source GPT model support rounds out Antigravity's multi-model architecture, giving teams flexibility to use whichever model performs best for their specific codebase and task type.

Pricing and Plans Explained

Antigravity 2.0 uses a compute-based usage model that refreshes every 5 hours, replacing the old daily prompt limits from Antigravity 1.0. All tiers support pay-as-you-go AI credits as top-ups when you hit your limit before the 5-hour refresh cycle.

Antigravity 2.0 vs Cursor vs Claude Code: The Honest Comparison

Google's timing with Antigravity 2.0 was deliberate. The announcement came directly into a market where Cursor dominates mindshare among developers, Claude Code has earned deep technical trust, and OpenAI's Codex and GitHub Copilot are both pushing into agentic territory. Here is how the platforms compare on the dimensions that matter most for developer decisions.

Bottom line on the comparison: Antigravity 2.0 has a clear architecture advantage in multi-agent parallelism and built-in browser verification. Cursor has a better track record for inline completion and developer experience polish. Claude Code has the strongest developer trust for pure terminal-based autonomous coding. The right choice depends on your workflow: Antigravity wins for greenfield projects and visual development; Cursor wins for daily inline completions; Claude Code wins for terminal-native autonomous tasks.

Gemini CLI to Antigravity: Migration Guide

If you are currently using Gemini CLI, your migration is mandatory before June 18, 2026. Here is the complete path.

Who Must Migrate

Migration is required for anyone on Google AI Free, AI Pro, or AI Ultra using Gemini CLI or Gemini Code Assist IDE extensions. Migration is also required for Gemini Code Assist for GitHub users, though the timeline is slightly different (new org installs stop on June 18; existing installs stop serving requests in the weeks after).

Who Is Not Affected

Organizations on Gemini Code Assist Standard or Enterprise licenses can continue using the legacy CLI indefinitely with no changes. Paid Gemini API key users also retain access.

Key Changes to Understand

The migration is not just a binary name change. Three things move: the binary itself changes from gemini to agy; the configuration file changes from GEMINI.md to AGENTS.md; and the skills directory changes from .gemini/skills/ to .agents/skills/. Extensions and plugins are now called Antigravity plugins. Core Gemini CLI features including Agent Skills, Hooks, Subagents, and Extensions all carried over, though Google acknowledges there is not full one-to-one feature parity at launch.

Install Antigravity CLI

# macOS and Linuxcurl -fsSL https://antigravity.google/cli/install.sh | bash

# Windows PowerShellirm https://antigravity.google/cli/install.ps1 | iex

# Verify installationagy --version

# SDK installation via pippip install google-antigravity

Key Configuration Changes (Gemini CLI to Antigravity CLI)

# Binary name changegemini "refactor auth module"   # OLD Gemini CLIagy    "refactor auth module"   # NEW Antigravity CLI

# Config file renameGEMINI.md         →  AGENTS.md

# Skills directory rename.gemini/skills/   →  .agents/skills/

# Start a new agent taskagy agent new "add input validation to user registration"

Good News for CI/CD Pipelines Both the Gemini CLI binary and the Antigravity CLI binary can coexist on the same machine during the migration window. There is no conflict between "gemini" and "agy" as commands. For GitHub Actions workflows, Docker containers, and shell scripts, updating invocations is mostly a find-and-replace from "gemini" to "agy." GEMINI.md to AGENTS.md renames inside Dockerfiles require explicit file mapping updates.

How to Install and Get Started with Google Antigravity

Getting started with Antigravity 2.0 involves choosing which surface fits your workflow best. Most developers will start with either the desktop app or the CLI.

Desktop App Installation

Download the Antigravity 2.0 desktop app from antigravity.google. It is a standalone application, separate from the original Antigravity IDE (VS Code fork), which remains available at version 1.23.2 for developers who prefer the traditional IDE experience. Do not run both the 2.0 standalone app and the old IDE simultaneously on the same project, as configuration conflicts can occur.

CLI Quick Start

After installing the CLI with the one-liner above, authenticate with your Google account. The CLI opens a browser window for Google OAuth on first run. Once authenticated, credentials are stored locally. Then navigate to your project directory and start an agent task:

First Agent Task with Antigravity CLI

# Navigate to your projectcd your-project-directory

# Create a new AGENTS.md config fileagy init

# Run your first agent taskagy "Add input validation to the user registration endpoint and write unit tests"

# Spawn a subagent for a parallel taskagy agent new "review all API endpoints for missing error handling"

# Schedule a background taskagy schedule add --interval=daily "run test suite and report failures"

Enterprise Features and Governance

The Gemini Enterprise Agent Platform is the fifth surface of Antigravity 2.0 and the most significant addition for organizations running AI in regulated environments. Autonomous agents that can write code, execute files, browse the web, and deploy to production require governance controls that consumer-tier tools do not provide.

Google's enterprise offering includes audit trails of agent actions, access controls for which agents can take which actions, compliance framework integrations, and team-level agent management dashboards. The platform runs on Google Cloud and can be configured to restrict agents to specific network zones, repositories, or deployment environments.

Organizations that have purchased Gemini Code Assist Standard or Enterprise licenses retain full access to their existing workflows and gain the new enterprise agent capabilities without a disruptive migration. Custom agent templates for common enterprise development patterns are available in Google AI Studio, and agents can be deployed to Google Cloud or self-hosted infrastructure through the SDK.

Honest Assessment: What Developers Are Actually Saying

The developer community response to Antigravity 2.0 is mixed in ways that reveal both the platform's genuine strengths and its real limitations at launch.

What Developers Praise

The multi-agent architecture is drawing genuine admiration from developers who have tested it on complex refactoring tasks. Parallel agent execution handles large-scale code reorganization faster than sequential single-agent tools. The built-in browser agent removes a manual verification step that frustrates developers using competing tools. The migration from Gemini CLI is faster than expected for most setups, with Google estimating under 10 minutes for typical projects.

What Developers Criticize

The forced migration timeline is the biggest source of frustration. Developers who built CI/CD pipelines around Gemini CLI are facing mandatory refactoring with a hard 28-day deadline. The decision not to open source Antigravity CLI drew immediate criticism on GitHub, with developers noting that Gemini CLI was open source and the closed-source replacement feels like a step backward for transparency. Some developers describe early Antigravity 2.0 experiences as "a blatant copy of Cursor/Codex with worse polish," suggesting the multi-agent architecture is powerful in theory but inconsistent in practice on complex codebases.

The honest assessment is this: Antigravity 2.0's architecture is genuinely ahead of most competitors for multi-agent parallelism, but the product's maturity does not yet match its architectural ambition. Teams that combine strong software development lifecycle discipline with agent orchestration skills are best positioned to extract value from it today. Developers wanting reliable, polished inline completions will find Cursor more satisfying for daily use until Antigravity's editor experience catches up.

Frequently Asked Questions About Google Antigravity

What is Google Antigravity?

Google Antigravity is an agent-first development platform launched by Google that allows developers to go from an idea to a production-ready application using autonomous AI agents. Version 2.0, announced at Google I/O 2026 on May 19, 2026, consists of five surfaces: a standalone desktop app, an Antigravity CLI (invoked as agy), an SDK, Managed Agents in the Gemini API, and the Gemini Enterprise Agent Platform.

What is the difference between Google Antigravity 1.0 and 2.0?

Antigravity 1.0 was a single VS Code fork IDE launched in November 2025 with AI coding assistance powered by Gemini. Antigravity 2.0 is a fundamentally different product: a five-surface agent platform where the agent, not the code editor, is the central abstraction. Version 2.0 adds a rebuilt standalone desktop app, a new Go-based CLI, an SDK, a Managed Agents API, and an enterprise platform. The core capability difference is that 2.0 runs multiple AI agents in parallel rather than a single sequential assistant.

Is Google Antigravity replacing Gemini CLI?

Yes. Google announced that Gemini CLI will stop serving requests for Google AI Free, Pro, and Ultra users, and for free Gemini Code Assist individual users, on June 18, 2026. The Antigravity CLI is the mandatory replacement. Enterprise users on Gemini Code Assist Standard or Enterprise licenses are not affected and retain indefinite access to the legacy CLI.

How much does Google Antigravity cost?

Antigravity has four tiers. Free access is available with basic limits. The AI Pro plan at $20 per month includes full platform access. The new AI Ultra plan at $100 per month offers 5x the Pro usage limits. The Ultra Premium plan at $200 per month (reduced from $250) offers 20x the Pro limits. All tiers use a compute-based model that refreshes every 5 hours, and all support pay-as-you-go credit top-ups.

What AI models does Google Antigravity support?

Antigravity 2.0 defaults to Gemini 3.5 Flash, which Google has co-optimized to run at approximately 289 output tokens per second within the Antigravity harness. It also supports Gemini 3 Pro, Anthropic Claude Sonnet 4.5, and OpenAI GPT-OSS through multi-model routing, allowing developers to direct different tasks to different models within the same interface.

How do I install Google Antigravity CLI?

On macOS and Linux: curl -fsSL https://antigravity.google/cli/install.sh | bash. On Windows PowerShell: irm https://antigravity.google/cli/install.ps1 | iex. The CLI is invoked with the command agy after installation. SDK installation: pip install google-antigravity. The official migration guide is at antigravity.google/docs/gcli-migration.

How does Google Antigravity compare to Cursor and Claude Code?

Antigravity 2.0 leads on multi-agent parallel execution, a built-in browser agent, scheduled background tasks, and voice command support. Cursor leads on inline completion polish and daily developer experience for IDE-centric workflows. Claude Code leads on terminal-based autonomous coding trust and is open source. Antigravity is the best choice for greenfield development and visual UI work; Cursor for daily inline completions; Claude Code for terminal-native autonomous refactoring.

Is Google Antigravity open source?

No. At launch, Google Antigravity CLI and the Antigravity 2.0 platform are not open source. This is a significant point of criticism from the developer community, particularly because the predecessor Gemini CLI was open source. Google has not announced any plans to open source the platform.

What happens to my Gemini CLI chat history when I migrate?

Your Gemini CLI chat history and data is stored at ~/.gemini/antigravity-backup. You can move it into the Antigravity CLI directory at ~/.gemini/antigravity (excluding mcp_config.json) and restart the application to restore your history. The two binaries can coexist on the same system during the migration window with no conflicts.

When did Google Antigravity launch?

The original Google Antigravity launched in November 2025 as an AI-first IDE. Google Antigravity 2.0 was announced and launched at Google I/O 2026 on May 19, 2026, expanding from a single IDE into the full five-surface agent platform described in this article.

Read Also:

AI Agents vs AI Chatbots: What’s the Real Difference

Gemini 3.5 Flash vs GPT-4o: Which Is Faster

How to Build Agents with the Gemini API

What Is Gemini Spark? Google's 24/7 Personal AI Agent Explained

Sources: Google Developer Blog (blog.google) · Google I/O 2026 Official Keynote · MarkTechPost · The Next Web · Apidog · buildfastwithai.com · byteiota.com