Antigravity is Google’s “agent-first” development platform, unveiled alongside the Gemini series of models. Its positioning isn’t to be just another code-completion plugin, but rather to place AI agents at the center of the development workflow: enabling agents to autonomously handle planning, coding, testing, and verification across three interfaces—the editor, terminal, and browser.
What Is Google Antigravity?
In traditional IDEs, AI serves as an assistant; in Antigravity, the AI agent is the protagonist. It organizes work around several key interfaces:
- Editor: Reading and writing code, making cross-file changes.
- Terminal: Executing builds, running commands, and running tests.
- Browser: Autonomously opening pages to verify frontend effects and check runtime results.
The agent formulates its own plans, executes them step by step, and presents the process and artifacts for your review of “what it did” and “why it did it.”
What Changed in 2026: It Took Over from Gemini CLI
At Google I/O on May 19, 2026, Antigravity stopped being a new experiment and became Google's single line for terminal and IDE work. Google shipped Antigravity 2.0, the Antigravity CLI, the Antigravity SDK, and the Antigravity IDE together, and announced the transition away from Gemini CLI: since June 18, 2026, Gemini CLI and the Gemini Code Assist IDE extensions no longer serve Google AI Pro, AI Ultra, or individual free users.
The Antigravity CLI is a Go rewrite invoked as agy, sharing an agent harness with the desktop app. Migrating from Gemini CLI, Agent Skills, Hooks, and Subagents carry over unchanged, and Extensions become plugins. Worth knowing: it is not open source, unlike the Apache 2.0 Gemini CLI, which drew substantial criticism from the community, and quotas moved from daily to weekly.
Features
- Agent-first: Designed for a workflow where you “give an AI task and let it run the entire chain.”
- Multi-interface collaboration: Closes the loop on coding, execution, and verification within a single platform, reducing tool-switching.
- Gemini-based: Leverages Gemini’s reasoning and long-context capabilities to handle complex projects.
- Observability: Emphasizes making developers’ agent decision-making and execution trajectories visible, rather than treating them as a black box.
Getting Started: Desktop or Command Line
Desktop IDE: download the Antigravity desktop app from the official site, sign in with your Google account, open a project and describe a task to the agent. It lays out a plan first, then works across the editor, terminal and browser, showing its process and artifacts for you to review.
Command line (agy): on macOS and Linux, use the official install script:
curl -fsSL https://antigravity.google/cli/install.sh | bashOn Windows, run irm https://antigravity.google/cli/install.ps1 | iex in PowerShell, or use winget install Google.AntigravityCLI. Note that the Homebrew cask installs the desktop app, not the CLI. Once installed, run agy in your project directory; the first time, it walks you through a color scheme, workspace trust and Google sign-in. To use an API key instead, you have to set both a provider and the matching environment variable — setting GEMINI_API_KEY alone does nothing.
Moving over from Gemini CLI: Agent Skills, Hooks and Subagents carry over directly, while Extensions need to be set up again as plugins; the old daily free quota is now weekly, so check your limits before heavy use.
Who Is It For?
- Developers looking to try the new paradigm of “agentic coding.”
- Teams needing AI to handle end-to-end tasks (from writing code to self-testing and verification).
- Users already in the Gemini ecosystem who want their toolchain integrated.
Usage Recommendations
Antigravity is a relatively new product, with its capabilities and stability still undergoing rapid iteration. When using it for real-world projects, it’s recommended to start with well-defined, verifiable tasks and retain manual review checkpoints—the more autonomous the agent, the more critical it is to oversee key nodes. Focusing on the execution plans and verification results it provides will give you a better sense of quality than looking at the final code alone.
How It Compares
- For the terminal agent with the most mature ecosystem and the most third-party resources: Claude Code.
- If you already pay for ChatGPT: OpenAI Codex.
- If you need an open-source, auditable CLI: Antigravity CLI isn't open source, so look at OpenCode or Grok Build.
Pricing
The official pricing page (checked 2026-09-06) lists four tiers:
- Individual: $0 with no subscription, access to multiple models including Gemini and Claude variants, unlimited completions and commands, under basic rate limits
- Google AI Pro: everything in Individual with more generous rate limits and flexible AI credit allocation
- Google AI Ultra: a further step up, aimed at developers using Antigravity as their primary tool
- Organizations: consumption-based pricing through Google Cloud with the Gemini Enterprise Agent Platform, rolling out to selected customers
The Antigravity pricing page does not state the Pro or Ultra amounts; check Google's AI subscription page for those. Google restructured those subscriptions during 2026, so figures in third-party articles should not be taken at face value. One migration note: the old 1,000-requests-a-day free allowance from Gemini CLI is gone, and Antigravity CLI quotas are counted weekly.
